Datasets:

pszemraj

/

scientific_lay_summarisation-plos-norm

like 6

Within the genomes of metazoans, nucleosomes are highly organised adjacent to the binding sites for a subset of transcription factors. Here we have sought to investigate which chromatin remodelling enzymes are responsible for this. We find that the ATP-dependent chromatin remodelling enzyme SNF2H plays a major role organising arrays of nucleosomes adjacent to the binding sites for the architectural transcription factor CTCF sites and acts to promote CTCF binding. At many other factor binding sites SNF2H and the related enzyme SNF2L contribute to nucleosome organisation. The action of SNF2H at CTCF sites is functionally important as depletion of CTCF or SNF2H affects transcription of a common group of genes. This suggests that chromatin remodelling ATPase’s most closely related to the Drosophila ISWI protein contribute to the function of many human gene regulatory elements. The genomes of eukaryotes exist predominantly as chromatin. The fundamental subunit of chromatin is the nucleosome which consists of 147 bp of DNA wrapped around an octamer of histone proteins [1]. Typically nucleosomes are distributed along DNA with defined spacing at distinct loci in a given cell type [2]. In addition, nucleosomes exhibit distinct translational positioning with respect to certain genomic features such as promoters [3–5], origins of DNA replication [6,7] and the binding sites for transcription factors such as CTCF [8,9]. CTCF binding has also been found to play a key contribution to the function of insulator elements [10]. Insulators are genetic elements that act to limit the range over which enhancers can act to regulate a gene [11]. Sites occupied by CTCF are frequently observed to also be enriched for subunits of the cohesin complex [12]. Cohesin is a multi- protein complex consisting of two SMC proteins (SMC1 and SMC3) and Rad21 (Scc1) and STAG (Scc3). It forms a ring like complex capable of encircling two DNA strands [13]. This function for cohesin was originally characterised as playing a key role in the association of newly replicated sister chromatids until they segregate in anaphase. However, subsequently it has become clear that cohesin can also mediate interactions between chromosomal loci during interphase. For example, interactions between cohesin and mediator have been found to mediate looping interactions between promoters and enhancers [14]. The combined action of both CTCF and cohesin mediates long range interactions and effects on gene expression [15–18]. In addition, recruitment of cohesin to CTCF binding sites also contributes to insulator activity [19–21]. However, in some cases CTCF sites remain functional following depletion of cohesin [18,22]. ATP-dependent chromatin remodelling enzymes have been found to play an important role in establishing the positioning of many nucleosomes within the genomes of model organisms [23]. More recently several studies have addressed the roles of members of this family of ATPases in the human genome. For example the human ISWI related remodelling enzymes SNF2H (also known as SMARCA5) has been found to contribute to DNA repair [24], and in a partially redundant fashion to the organisation of a subset of DNase hypersensitive sites [25]. This study also found that SNF2H and CHD4 associate with a significant number of CTCF binding sites and a previous study demonstrated a role for the enzyme CHD8 at CTCF sites [26]. Both CHD8 and SNF2H have been shown to affect enhancer blocking mediated by CTCF at individual loci [26][27]. More recently, the bromodomain PHD finger-containing transcription factor (BPTF) subunit of the NURF complex has been observed to contribute to localised chromatin accessibility at CTCF sites and the regulation of CTCF target genes [28]. SNF2H is known to function as the catalytic ATPase in at least five distinct complexes in mammalian cells, namely ACF, CHRAC, WICH, RSF and NoRC [29]. The accessory subunits with which the SNF2H ATPase subunit is associated with varies in the different complexes. For example, SNF2H is found in association with WSTF in the WICH complex, with Tip5 in NoRC, Acf1 in ACF, and with both Acf1 and CHRAC 15/17 in CHRAC [29]. The related ATPase SNF2L is the ATPase subunit in the Cerf and NURF complexes [29]. To our knowledge no studies to date have investigated the contribution of different remodelling enzymes to the establishment of organised nucleosomal arrays adjacent to CTCF and other transcription factor binding sites. Here we find that SNF2H plays a major role in the establishment of ordered arrays of phased nucleosomes flanking CTCF binding sites. The related enzyme SNF2L plays a minor role at CTCF sites, and contributes to nucleosome positioning adjacent to other transcription factors. Depletion of SNF2H results in alterations to the expression of many CTCF dependent genes indicating a role for this enzyme in CTCF function and raising the possibility that nucleosome phasing contributes to function at gene regulatory elements. To investigate the contributions of ATP-dependent chromatin remodelling enzymes in nucleosome organisation, we adopted an siRNA based approach to deplete selected enzymes in cultured HeLa cells. CHD1, CHD2, CHD4 (mi-2), SNF2L and SHF2H could be depleted to between 80% and 96% as judged by western blotting (Fig 1A and S1 Fig). Chromatin isolated from these cells was digested with micrococcal nuclease and the nucleosomal ladder was assessed by gel electrophoresis. Subtle changes in the digestion pattern were observed, but in all cases a distinct species of approximately 150 bp was detected (S1 Fig). In order to characterise the distribution of these nucleosomal DNA fragments, they were subject to high throughput sequencing to a depth of 40–350 million paired reads per repeat. We investigated how depletion of these enzymes affected the organisation of nucleosomes at the promoters of ubiquitously expressed genes. We noticed variation in distribution of nucleosomes across promoters between experimental repeats and realised that this pattern varied with the extent of MNase digestion (Fig 1B and 1C). The extent of MNase digestion could be assessed from the mean length of the mono nucleosome fragments. With fragments digested to a mean length of 147 bp the nucleosome free region was distinct (Fig 1B). With digestion to 169 bp the nucleosome depleted region is partially filled and the -1 nucleosome more prominent (Fig 1C). Using controls with comparable MNase digestion it was not possible to detect significant changes in nucleosome distribution following depletion of SNF2H, SNF2L (Fig 1B and 1C), CHD1, CHD2 or CHD4 (S1 Fig). We next investigated the organisation of nucleosomes adjacent to CTCF binding sites where strikingly well organised arrays of around 20 positioned nucleosomes have been reported previously [8] (Fig 2A). As expected, the organisation of nucleosomes is dependent on CTCF as siRNA depletion of CTCF reduces the nucleosomal pattern (Fig 2A). While depletion of CHD1, CHD2 and CHD4 had little effect on this pattern (S2 Fig), depletion of SNF2H resulted in a significant loss of nucleosome organisation at these sites (Fig 2B). Depletion of SNF2L had a small effect on the nucleosomes adjacent to the CTCF binding site, with progressively weaker effects at nucleosomes located further away (Fig 2C). As SNF2H is present within multiple distinct remodelling complexes in human cells, we next attempted to distinguish which complexes were involved. siRNA depletion of the ACF1, RSF1, TIP5 and WSTF subunits of these complexes did not disrupt nucleosome organisation to the same extent as observed for SNF2H (S3 Fig). We conclude that different SNF2H containing complexes may function with partial redundancy. SNF2L is known to form a complex with subunits of the human NURF complex including BPTF [30,31]. Depletion of BPTF resulted in a change to the organisation of nucleosomes immediately adjacent to CTCF sites related to that observed with SNF2L suggesting that SNF2L functions at CTCF sites as a component of the NURF complex (S3 Fig). As SNF2H affects nucleosome organisation at CTCF sites, we examined whether SNF2H is physically associated with CTCF sites by ChIP. SNF2H is enriched at CTCF sites and enrichment at these sites is reduced following depletion of CTCF (Fig 3A). We also noticed that nucleosome occupancy increased at CTCF sites following depletion of SNF2H (Fig 2B). This led us to investigate whether CTCF occupancy was affected following depletion of SNF2H. The ChIP signal for CTCF was indeed found to be reduced following depletion of SNF2H (Fig 3B). This indicates that in addition to organising nucleosomes adjacent to CTCF, SNF2H acts to maintain high CTCF occupancy. Given that sites bound by CTCF are often also found to be enriched for cohesin, we investigated the effect of depleting CTCF or SNF2H on ChIP enrichment for the cohesin subunit Rad21. Fig 4A shows that enrichment for Rad21 is reduced by approximately 64% following depletion of CTCF. This is consistent with previous studies showing that recruitment of Rad21 to CTCF sites is dependent on CTCF [12,19–21]. Enrichment of Rad21 is also reduced following depletion of SNF2H (Fig 4B). The reduction in occupancy (36%) is likely to be attributable to 68% reduction of CTCF occupancy following depletion of SNF2H rather than a direct role for SNF2H in Rad21 loading. We also observe that depletion of Rad21 had no effect on SNF2H recruitment to CTCF binding sites (Fig 4C) and consistent with this, depletion of Rad21 had little effect on nucleosome organisation at CTCF binding sites (Fig 4D). Previous studies have collated ChIP data characterising the interaction sites for some 119 different transcription factors [32] and this information can be used to align nucleosome distribution adjacent to these factors [9]. Here we select 50 factors for which there are over 1000 genomic binding sites characterised in HeLa cells. Consistent with previous studies we find that binding sites for some factors are located in regions of nucleosome depletion or enrichment without precise positioning of adjacent nucleosomes, whereas other factors such as JUN and RFX5 are flanked by arrays of positioned nucleosomes (Fig 5C and 5D and S4 Fig). While performing this analysis we observed that by ChIP, we could detect enrichment for CTCF at the binding sites for many transcription factors (Fig 5A and 5B and S4 Fig). We reasoned that in some cases CTCF binding sites are located adjacent to the binding sites for other factors. To test this we filtered out any factor binding sites that had a CTCF binding sites within 500 bp. When only binding sites that did not have CTCF sites within 500 bp were considered, CTCF enrichment at the remaining sites was greatly reduced (Fig 5A and 5B). We noticed that the effect of filtering out adjacent CTCF sites had differing effects on the organisation of nucleosomes. RFX5 sites that do include adjacent CTCF sites have well organised arrays of nucleosomes (Fig 5C, red). In contrast the RFX5 sites that are not adjacent to CTCF sites have less well organised adjacent nucleosomes (Fig 5C, blue). For RFX5 this effect is very significant as 38% of RFX5 sites are within 500 bp of a CTCF site. Depletion of CTCF significantly perturbs the organisation of nucleosomes at RFX5 sites with adjacent CTCF sites (Fig 5E). This shows that the correlation between the presence of adjacent CTCF sites is functionally significant for nucleosome organisation. CTCF also contributes to the recruitment of cohesin at RFX5 sites as this is reduced following CTCF depletion (Fig 5H). However, the proportion of Rad21 that remains associated following depletion of CTCF indicates that RFX5 is capable of recruiting some cohesin independently of CTCF. In contrast to the observations at RFX5 sites, the nucleosomes distal to JUN sites are affected in a more complex way. The two nucleosomes immediately adjacent to JUN sites are better organised when there are nearby CTCF sites whilst the extended array of nucleosomes extending beyond the third nucleosome is less ordered as assessed by the depth and periodicity of the normalised read depth (Fig 5D). CTCF depletion results in a modest improvement to nucleosome organisation at JUN sites with adjacent CTCF (Fig 5F). JUN sites lacking adjacent CTCF sites show less change of the distal nucleosomal array following CTCF depletion (Fig 5G). Depletion of CTCF has only a minor effect on Rad21 ChIP at JUN sites indicating that JUN can organise cohesin independently (Fig 5I). The effects of adjacent CTCF sites observed at RFX5 sites are also observed at the binding sites for other transcription factors. For example nucleosomes are better organised adjacent to the binding sites of factors such as BRCA1 and GTF2F1 that have adjacent CTCF sites (S4 Fig). Enrichment of cohesin is often affected in a similar way (S5 Fig). This illustrates a pitfall in the use of averaging to study correlations in the distributions of chromatin associated factors at complex regulatory elements which are likely to include binding sites for many different factors. For this reason we consider only factor binding sites that do not have adjacent CTCF sites for the subsequent analysis. To investigate the involvement of SNF2H and SNF2L in nucleosome organisation at different factor binding sites, we plotted the organisation of nucleosomes following depletion of each enzyme flanking the binding sites for 50 different transcription factors (S6 Fig). As at promoters significant differences in organisation were observed with different levels of MNase digestion. However differences in chromatin organisation are apparent when compared to control digestions with similar nucleosome fragment lengths. Depletion of SNF2H has effects on nucleosome organisation surrounding binding sites of factors such as JUN (Fig 6B). The effects are most pronounced for nucleosomes distal to the factor binding site. For example the nucleosomes distal to the +3 nucleosome are less well organised at JUN sites following SNF2H depletion (Fig 6B). Similar effects are observed surrounding 24 additional transcription factors (S6 Fig). Depletion of SNF2L was observed to result in a small reduction in occupancy of nucleosomes proximal to a subset of factor binding sites (Fig 6A and 6C and S6 Fig). For all factors where SNF2H or SNF2L depletion was observed to affect the generation of extended nucleosomal arrays, SNF2H or BPTF were also observed to be present by ChIP (S5 Fig). For example at JUN sites there is enrichment for SNF2H and BPTF by ChIP. However, enrichment for SNF2H and BPTF was also observed at some factor binding sites where nucleosomes are not well organised, for example GTF3C2 (S4 and S5 Figs). Binding of SNF2H was not enriched at all sites, for example as observed at E2H2 and FAM48A sites (S5 Fig) and there is minimal nucleosome organisation at these sites. To investigate the functional significance of SNF2H dependent phasing of nucleosome arrays we compared the effects of depleting CTCF and SNF2H. Approximately 1000 genes were significantly affected by the transient depletion of either protein. Many of the up and down regulated genes are affected similarly by depletion of CTCF or SNF2H (Fig 7A). This overlap is highly statistically significant with P values lower than 10−50. The most probable explanation for this is that SNF2H is required for the function of a significant subset of CTCF sites. As SNF2H affects both CTCF occupancy and nucleosome positioning it is difficult to distinguish which is dominant. However, it is possible to identify cohorts of genes where CTCF occupancy was either unchanged following SNF2H depletion or reduced. At the genes where CTCF is retained nucleosome positioning is reduced following SNF2H depletion (Fig 7B). In contrast, where CTCF occupancy is lost, nucleosome organisation is completely lost (Fig 7C). To investigate whether changes of CTCF occupancy were correlated with genes that showed changes in expression following depletion of SNF2H, occupancy of CTCF was assessed at all sites within 10kb of genes that changed expression. The changes in CTCF occupancy at genes that changed expression were indistinguishable from the changes observed at all genes (Fig 7D). This suggests that the overlap between genes affected by CTCF and SNF2H depletion cannot accounted for by a simple change in CTCF occupancy. At CTCF sites we observed that depletion of SNF2H resulted in a substantial reduction to nucleosomal pattern flanking these sites. This establishes that SNF2H plays the major role in the establishment of the remarkably well organised arrays of nucleosomes observed flanking CTCF sites. Several additional remodelling ATPases including CHD4, CHD8 and SNF2L have also been reported to be recruited to CTCF sites [25,26,28]. These enzymes are unlikely to have major roles in the establishment of extended nucleosomal arrays adjacent to CTCF sites as this is so strongly dependent on SNF2H. Depletion of SNF2L or BPTF had minor effects on the nucleosomes proximal to CTCF sites. This localised effect in the SNF2L depletion is consistent with a local alteration to digestion detected using microarrays [28]. The SNF2H ATPase is present within multiple distinct complexes. The effects of depleting distinguishing subunits of these complexes were inconclusive suggesting that there may be some redundancy between different complexes. However, depletion of the ACF1 subunit of the SNF2H containing human ACF complex and the WSTF subunit of the human WICH complex resulted in subtle reductions to nucleosome organisation especially at locations distal to CTCF (S3 Fig). Both the ACF and WSTF complexes have the biochemical capability to organise chromatin [35,36]. SNF2L has been purified as a component of a distinct remodelling complexes, NURF [31] and CERF [37]. Expression of SNF2L was originally thought to be restricted to brain and gonadal tissue [38] however, more recent studies indicated that it is ubiquitously expressed [39] and has functions in Wnt signalling and at CTCF sites [28,39,40]. The CERF complex is found in neural tissues [37]. The biochemical activities of NURF are distinct from those of ACF in that NURF was originally purified based upon its ability to disrupt nucleosomal arrays [30]. The role that human NURF plays in nucleosome positioning adjacent to human transcription factors is consistent with the original assays for DNaseI hypersensitivity. Although NURF repositions nucleosomes, it also interacts with transcription factors [41] and this can result in directional repositioning of nucleosomes adjacent to factor binding sites [42]. Thus, the NURF complex has the biochemical properties to direct the positioning of nucleosomes immediately adjacent to factors such as CTCF. We also investigated the effects of chromatin remodelling enzymes on nucleosome organisation at the binding sites for 49 additional transcription factors and at promoters. 29 of these organise extended arrays of nucleosomes and SNF2H contributes to nucleosome organisation at most of these (24/29) (S6 Fig). Typically the nucleosomes immediately flanking the factor binding site are best organised. The distance between these +1 and -1 nucleosomes flanking factor binding sites ranges from 258 bp (REST) to 364 bp (RCOR1). This is substantially larger than would be anticipated based on steric occlusion at the factor binding site and the linker observed between adjacent nucleosomes. It may be that many of these factors are bound by additional cofactors. For example, CTCF is known to associate with TAF3 [43]. Following depletion of SNF2H the distance between the +1 nucleosomes flanking factor binding sites increasing by on average 25bp, in addition the average separation between flanking nucleosomes increases from 176 bp to 183 bp. This is consistent with a role for SNF2H in driving nucleosomes together and towards the factor binding site. The effects following SNF2L depletion are relatively minor, but also distinct in that the distance between adjacent +1 nucleosomes reduces by 10bp and the separation between adjacent nucleosomes is reduced from 176 to 173 bp. This suggests that SNF2L complexes may act to move nucleosomes away from bound factors. The finding that different remodelling enzymes act to alter nucleosome positioning with different directionality is reminiscent of the way remodelling enzymes act with different directionalities at yeast promoters [44] and suggests that a similar interplay operates at the binding sites for a range of transcriptional regulators. With the possible exception of SNF2H at CTCF sites, the effects of depleting enzymes result in alterations to the distributions of nucleosomes rather than complete loss. This suggests that as yet unidentified factors are likely to function in a partially redundant fashion with SNF2H and SNF2L. In vitro, it has been observed that bound transcription factors act as a barrier restricting the positioning of nucleosomes remodelling enzymes [45]. The observations made here provide evidence that the biophysical interplay between bound factors and nucleosome repositioning characterised in vitro is likely to contribute to nucleosome organisation at functional regulatory elements. Nucleosomes positioned adjacent to such barriers could act as a reference point from which progressively distal nucleosomes are organised [46], potentially providing a means of organising chromatin adjacent to any bound factor. This raises the question why are nucleosomes much better organised adjacent to some bound factors than others? It is possible that targeted recruitment of remodelling enzymes is required in addition to the presence of a barrier. For example both SNF2L and SNF2H interact with CTCF [27,28]. However, we also observe dependency upon SNF2L and SNF2H at the binding sites for an additional 24 transcription factors. It is difficult to imagine that SNF2H and SNF2L containing complexes possess the capability of recognising such a structurally diverse range of factors. For this reason we consider it likely that in addition to direct association with transcription factors other interactions contribute to the recruitment of these enzymes. A prime candidate would be modification to histones such as H3 K4 trimethylation which is enriched at the binding sites for many transcription factors [9]. The SNF2L containing NURF complex has specificity for histone H3 methylated at lysine 4 [47] and so this modification is likely to contribute to recruitment. In budding yeast, ISWI chromatin remodelling enzymes have been shown to be recruited by a looping mechanism [48]. As CTCF sites are also sites of gene looping, this mode potentially provides an additional means via which human ISWI containing enzymes could be recruited. Eukaryotic regulatory elements seldom consist of the binding sites for a single sequence specific regulatory factor in isolation. Instead binding sites for disparate factors are often found within close proximity. This complicates the interpretation of which chromatin features are directly recruited by a transcription factor and which are influenced by neighbouring factors. Our own data show that both nucleosome organisation and cohesin enrichment at many factor binding sites is influenced by the presence of adjacent CTCF sites (S4 and S5 Figs). CTCF is an unusual transcription factor in that its interaction with DNA is very stable [49] and it plays a major role in determining the distribution of cohesin during interphase [15–18]. This means that interference from adjacent CTCF sites may have especially strong effects. Nonetheless, in principle, similar forms of interference could occur between any two transcription factors and affect the distribution of any chromatin feature such as a histone modification or a cofactor. The scope for misinterpretation as a result of this type of interference is especially high when averaged enrichment is considered for many sites. The use of averaging in metazoan genomic datasets is often essential as the read depth with which data has been collected is in many cases not sufficient for analysis at single genes. This is especially relevant for high resolution studies of nucleosome positioning as this requires a high read depth at each base pair. Obtaining data with the required depth is in most cases impractical and averaging at many related sites provides a way round this. To our knowledge, there is one dataset, with a depth of 3. 6 billion reads, that potentially does have the depth required to call nucleosome locations at single loci in human cell lines [50]. However, we could not use this to study alignment to CTCF or other transcription factors as ChIP has not been performed to determine the occupancy of these factors in the cell lines used. The enhanced nucleosome phasing adjacent to sites such as RFX5 and ZNF143 that have adjacent CTCF sites is best explained if these different factors constructively interfere with each other to generate stronger nucleosome phasing (Fig 5 and S4 Fig). This would require that the binding sites are in phase with the nucleosomal repeat. A large proportion of the CTCF sites are immediately adjacent to RFX5 and ZNF143, so this is feasible. At JUN sites nucleosome organisation increases following removal of CTCF sites in silico or following depletion of CTCF (Fig 5 and S4 Fig). This suggests that in this case CTCF destructively interferes with the phasing of nucleosomes established by JUN. Constructive and destructive interference in the phasing of nucleosomes by different factors has also been observed on adjacent coding genes in yeast [51]. These findings indicate the potential for complexity in the way that chromatin is arranged over regulatory elements that contain binding sites for many different factors. We observed that depletion of SNF2H results in a reduction in the occupancy of CTCF at many sites (Fig 3). Consistent with this depletion of SNF2H has previously been observed to result in decreased binding of CTCF at the H19/Igf2 locus [27]. This suggests that the action of SNF2H promotes CTCF binding. There is a literature supporting a role for ATP-dependent remodelling enzymes in facilitating the binding of transcription factors to chromatin [52] however, this has typically involved enzymes such as SWI/SNF that disrupt chromatin organisation rather than ISWI containing enzymes such as ACF that space nucleosomes evenly. How then could a nucleosome spacing enzyme act to promote factor occupancy? Currently favoured mechanisms for nucleosome spacing involve the enzyme sensing DNA adjacent to nucleosomes such that repositioning occurs towards the side of a nucleosome with a long accessible linker [23]. This results in the repositioning of nucleosomes with a mean location equidistant between neighbouring nucleosomes. Strongly bound transcription factors such as CTCF also potentially reduce access to linker DNA. In this situation a spacing enzyme would be anticipated to move a nucleosome away from the factor bound linker. Indeed the positioning of nucleosomes by ISWI-related complexes has been observed to be affected by transcription factor binding in vitro [53,54]. The repositioning of nucleosomes away from factor bound sites effectively partitions DNA sequences occupied by transcription factors and nucleosomes. As a result of reduced competition with nucleosomes the factor bound state would be favoured. This contrasts with the action of complexes such as SWI/SNF which move nucleosomes across factor binding sites resulting in dissociation [54]. Reducing competition with nucleosomes may be especially important at CTCF sites as the binding consensus sequence has high GC content and high inherent affinity for nucleosomes [9]. Therefore, unbound CTCF sites are likely to be occupied by nucleosomes. Supporting this increased nucleosome occupancy is observed at CTCF sites that are only occupied in specific cell lines [55] and in our own data following depletion of CTCF or SNF2H (Figs 2A, 2B and 7C). On the other hand when bound by CTCF the action of SNF2H acts to reduce competition with nucleosomes and further stabilise the bound state. The positive feedback favouring both bound and non-bound states may help to explain how the subset of CTCF consensus sequences that are actually bound varies between different cell lines [55] and during differentiation [56]. Following depletion of SNF2H a quite striking increase in the occupancy of nucleosomes well positioned over CTCF sites is observed at locations where CTCF occupancy is reduced (Fig 7C). These well positioned nucleosomes do not by themselves result in the establishment of well-ordered arrays of flanking nucleosomes. This suggests that the level of non-targeted nucleosome spacing activity in human cells is insufficient on its own to establish ordered arrays of nucleosomes. At the sites where well-ordered arrays are observed, there is likely to be a requirement for both a barrier from which the array is established and targeted recruitment of enzymes such as SNF2H and SNF2L to propagate and maintain spaced chromatin. As CTCF acts to recruit cohesin and SNF2H promotes CTCF occupancy, SNF2H would be expected to influence cohesin occupancy at CTCF sites. This is indeed the case as we observe a reduction in cohesin by ChIP at CTCF sites following SNF2H depletion (Fig 4). A previous study also observed that loading of cohesin was reduced following inactivation of SNF2H [57]. To investigate whether SNF2H contributes to cohesin loading independently of its effect on CTCF binding, the enrichment of Rad21 was plotted at CTCF sites that remain occupied and adjacent to the binding sites for other transcription factors. Enrichment for Rad21 is not affected at these locations following depletion of SNF2H (S7 Fig). This suggests that SNF2H is not a general loading factor for cohesin, but affects its loading at a subset of CTCF sites. We do not believe that a remodelling complex containing cohesin contributes to nucleosome organisation at CTCF sites as depletion of Rad21 has no effect on nucleosome organisation (Fig 4D). Following SNF2H depletion we observe that nucleosomes become disorganised and CTCF occupancy is reduced. As many CTCF dependent genes show changes in expression following SNF2H depletion, in principle either or both of these effects could contribute to SNF2H function. However, the lack of any difference in CTCF occupancy at the CTCF target genes affected by SNF2H depletion (Fig 7D) suggests that changes in CTCF occupancy do provide a simple means of explaining the effects on transcription. This raises the possibility that nucleosome positioning is functionally significant, but further investigation will be required to establish this rigorously. The internucleosome spacing adjacent to CTCF sites is 176 bp, 19 bp shorter than the major internucleosome spacing of 198 bp detected in mammalian cells [2]. In addition, the nucleosomes adjacent to CTCF binding sites are unusually well translationally positioned. The presence of similarly well organised nucleosomes over yeast coding genes is correlated with low histone turnover, histone modification and reduced non coding transcription [58–60]. As non-coding transcription also contributes to enhancer function [61] it is possible that the organised nucleosomes adjacent to CTCF sites also affect enhancer function via RNA mediated pathways. HeLa cells originally obtained from the ATCC Global Bioresource Center were cultured in DMEM (Invitrogen) supplemented with 0. 2 mM l-glutamine and 10% FBS. The siRNA oligonucleotides were purchased from Eurofins MWG used at a final concentration of 7. 8 nM. The siRNA transfections were performed using INTERFERin (Polyplus Transfections). All siRNA sequences are listed in Table 1. Cells were transfected three times according to the INTERFERin protocol with 72hours of growing in between transfections. To check for depletion of proteins after siRNA transfections, whole cell extracts of HeLa cells were prepared by lysing cells in WCE-buffer (20mM Hepes pH7. 6,400mM NaCl, 1 mM EDTA, 25% glycerol, 0. 1% NP-40, protease inhibitors) followed by homogenization using a syringe. SNF2L depletion was checked by directly lysing counted cells using urea sample buffer as described by Eckey M. et al, [39]. Primary antibodies for Western blots used were rabbit anti-human SNF2H (Bethyl Laboratories, A301-081A), mouse anti-CTCF (Abcam, ab37477), rabbit anti-RAD21 (Abcam, ab992), rabbit anti-CHD1 (Bethyl Laboratories, A301-218A), rabbit anti-CHD2 (Active Motif, 39364), rabbit anti-CHD4 (Bethyl Laboratories, A301-081A), rabbit anti-ACF1 (Bethyl Laboratories, A301-318A), rabbit anti-WSTF (Cell Signaling, 2152), rabbit anti–TIP5 (Invitrogen, 491037) and mouse anti-beta actin (Sigma, A2228). Primary antibodies for ChIP used were rabbit anti-human SNF2H (Abcam, ab72499), mouse anti-CTCF (Millipore, 17–10044), rabbit anti-RAD21 (Abcam, ab992), rabbit anti-BPTF (Millipore, Abe24), and rat anti-SNF2L (2C4, [39] which has been kindly provided by P. Becker). Secondary antibodies for Western blots used were Alexa Fluor 680 goat anti-mouse (Invitrogen) and Alexa Fluor 790 goat anti-rabbit (Invitrogen) for immunofluorescence staining and analysis using the LI-COR Odyssey CLx. For verification of the RNAi depletion of TIP5, and SNF2L RNA was isolated using the QIAGEN RNeasy kit according to the manufacturer. 2μg RNA was reverse transcribed into cDNA (QIAGEN, QuantiTect kit). PCR was carried out in a total volume of 15 μl by using 2 μl of cDNA with the Quanta PerfeCTa SYBR Green FastMix and TIP5 transcript primers ( (1) for TTCTCCTATGTTGGGATCTAGCA/ rev CAGTGCCATTCTCTGCCACA and (2) for GGCCTACGACTGTCTCTGGAA/ rev TTGGGGATGAAGGTTGCCG) or SNF2L transcript primers ( (1) for AAGCGCCTAAATATGAAAAGGA/ rev GCGGTAGTCTCCAGCAGAAAT and (2) for GCTGGAGACTACCGCCCATAG/ rev CAACCAATTCAGTAATCGAATAT) according to Quanta standard protocol using an AB 7500 Real Time PCR Cycler. Beta-actin transcript was used for normalization. ~8 x105 siRNA-transfected cells were crosslinked with 1% formaldehyde for 10 min and quenched for 5 min with 125 mM glycine at room temperature. After washing cells with cold PBS, cells were lysed using cold NP40-lysis buffer (10 mM Tris pH 7. 5,10 mM NaCl, 3 mM MgCl2,0. 5% NP40,0. 15 mM spermine, 0. 5 mM spermidine) for 5 min on ice. Cells were pelleted and washed with MNase digestion buffer (10 mM Tris pH 7. 5,15 mM NaCl, 60 mM KCl, 1 mM CaCl2,0. 15 mM spermine, 0. 5 mM spermidine) and resuspended in 50 μl MNase digestion buffer. For the digest, 3units MNase S7 (Roche) were added and incubated for 2 min (low digest) or 4 min (high digest) at 37°C. The digest was stopped adding 1/10 vol 10% SDS and 1/10 vol 250 mM EDTA. NaCl was added at a final concentration of 0. 2M to reverse the crosslinking at 65°C overnight. The samples were treated with 40 μg proteinase K for 30 min at 45°C and 10 μg RNase A for 30 min at 37°C, followed by phenol-chloroform extraction and purification using a PCR purification kit (QIAGEN). The samples were eluted from the columns with 50 μl 10 mM Tris pH 7. 5 and run on a 1. 2% agarose gel in 1 x TAE. The gels were stained using SYBR Safe DNA gel stain (Invitrogen) and mono nucleosomes were cut out. Bands were gel extracted with the QIAGEN gel extraction kit. The resulting DNA fragments were used for Illumina library preparations. ~3. 2 x107 siRNA-transfected cells were used per ChIP. Cells were cross-linked with 1% formaldehyde for 10 min and quenched for 5 min with 125 mM glycine at room temperature. After washing cells with twice with ice-cold PBS, cell pellets were flash frozen in liquid nitrogen and stored at -80°C. Frozen cell pellets were lysed in 1. 8 ml lysis buffer containing 1% SDS, 10 mM EDTA, 50 mM Tris pH 8. 1 and protease inhibitors. To shear chromatin to fragments of about 200–500 bp size, samples were sonicated in 300 μl volumes for 15 cycles (7. 5 min total sonication time) at high setting using a Bioruptor (Diagenode). Sonicated lysates were then cleared by centrifugation for 10 min at high speed, diluted 1/10 with dilution buffer (1% Triton X-100,2 mM EDTA, 150 mM NaCl, 20 mM Tris at pH 8. 1,0. 1% Brij-35) and incubated with 12 μg of the respective antibody overnight at 4°C. For each ChIP, 200 μl of Protein G Dynabeads (Life Technologies) were pre-incubated with 0. 5% (w/v) BSA in PBS overnight. To capture antibody-bound protein-DNA complexes, lysates were incubated with the prepared beads for 3 hrs and subsequently washed twice with 6ml of each wash buffer I (0. 1% SDS, 1% Triton X-100,2 mM EDTA pH 8. 0,20 mM Tris pH 8. 1,150 mM NaCl), wash buffer II (0. 1% SDS, 1% Triton X-100,2 mM EDTA pH 8. 0,20 mM Tris pH 8. 1,500 mM NaCl) wash buffer III (0. 25 mM LiCl, 1% NP-40,1% sodium deoxycholate, 1 mM EDTA pH 8. 0,10 mM Tris pH 8. 1) and TE buffer (10 mM Tris pH 8. 1,1 mM EDTA pH 8. 0) in the cold. To elute, reverse-crosslink and purify ChIP DNA the IPure kit from Diagnode was used according to the manufacturer. The resulting DNA was used for Illumina library preparations. Libaries from ChIP DNA or mono nucleosomal DNA resulting from MNase digests were prepared using the protocol published (Bowman SK et al, BMC Genomics 2013) with modifications. All enzymes, buffers and nucleotides were purchased from Fermentas unless stated differently. In short, DNA was end repaired in 50 μl reactions containing 1x T4 ligase buffer, 0. 4 mM dNTPs, 7. 5 U T4 DNA polymerase, 5 U Klenow polymerase, 15 U T4 polynucleotide kinase for 30 minutes at room. To purify DNA, 1. 8 volumes Agencourt AMPure XP beads were used according to the manufacturer. A-tailing reactions (50 μl) contained cleaned up DNA, 1x Klenow buffer, 2 mM dATP, 15U Klenow 3’-5’ exo—and were incubated for 30 minutes at 37°C. DNA purification was performed using 1. 8 volumes Agencourt AMPure XP beads. Adapter ligation reactions in 50 μl volumes contained DNA from A-tailing, 1x T4 ligase buffer, 0. 04 μM annealed universal adapter for ChIP samples and 2 μM adapter for mono nucleosomal DNA, 5 Weiss U T4 ligase and were incubated overnight at 16°C. This time DNA was purified using 1. 1 volumes of Agencourt AMPure XP beads to avoid co-purifying excess adapter. DNA was eluted using 20 μl H2O of which half was used to amplify DNA in the next step. For library amplification the PCR reactions contained 5 μl adapter ligated DNA, 1x Phusion HF buffer (Thermo Scientific), 0. 3 μM Illumina universal primer, 0. 3 μM Illumina barcoded primer, 0. 4 mM dNTP, 200mM Trehalose and 3 U Phusion Hot Start II High Fidelity DNA Polymerase (Thermo Scientific). Thermocycling was performed by denaturing for 3 minutes at 98°C; followed by 20 cycles for ChIP DNA and 10 cycles for mono nucleosomal DNA of: 15 seconds at 98°C, 25 seconds at 60°C, and 1 minute at 68°C, and a final extension of 5 minutes at 68°C. PCR products were resolved on a 1. 2% agarose gel in 1x TAE. ~250 bp to 700 bp of ChIP DNA and ~300 bp of mono nucleosomal DNA was extracted using QIAGEN MinElute Kit and sent for sequencing. Paired end libraries of MNase digested chromatin ChIP DNA were sequenced using illumine HiSeq technology. Fastq files containing raw reads were aligned to human reference genome (ftp: //ftp. ccb. jhu. edu/pub/data/bowtie2_indexes/hg19. zip) hg19 by Bowtie2 with option of maximum fragment length of 1500 for chip data and 500 for nucleosome fragments. Fragment length distributions for each sample used are shown in S1 Table. The midpoints of uniquely mapped nucleosomal or ChIP reads were used for further analysis. Transcription factor tracks [62] in HeLa cells were downloaded using the UCSC table browser [63] of the encode database (http: //genome. ucsc. edu/ENCODE/dataMatrix/encodeChipMatrixHuman. html) as narrow Peak file formats. A 2kb region flanking the TFB site was selected for nucleosome or ChIP enrichment analysis. The nucleosome dyads/chip fragment reads coverage was calculated for each base in the 2kb region. This enrichment value at each base was then divided by number of TFB sites and total number of reads in the experiment to obtain normalised reads. The plotted data was normalised to have same mean read counts in the plotted window. The data was smoothed using a 50 bp sliding window for graphical representation. Plots were generated with python’s plotting modules matplotlib and pylab. All of the data shown in the manuscript was established as being reproducible between repeats of genome wide experiments. In most cases the data plotted is the average of appropriately digested biological repeats. A full description of the data included in each figure is provided in (S1 Table). In some figures the same control enrichments are re-plotted in different panels. Sequence data is accessible at the European nucleotide archive (ENA) http: //www. ebi. ac. uk/ena/data/view/PRJEB8713 under accession number PRJEB8713. RNA seq analysis pipeline described in [64] was followed for mapping and measuring differential expression of genes. In brief, the paired end reads of each biological replicate was mapped to hg19 human reference genome independently with TopHat [tophat2 -p 8 -r 200 -g 2 -o output folder hg19 reads_R1_001. fastq reads_R2_001. fastq]. The assembled reads were the provided as input in Cufflinks which generates assembled transcripts for each replicate [cufflinks -p 8 -g hg19_genes. gtf -o output folder mapped reads. bam]. Mapped reads were the used as input in Cuffdiff to obtain differential expression results in tabular format [cuffdiff -p 8 -o output cuffdiff hg19_genes. gtf siScr_1. bam, siScr_2. bam siSNF2H_1. bam, siSNF2H_2 bam]. Transcripts having >1. 5 fold changes in their expression were selected as differentially regulated and have uncorrected p-value of the test statistic <0. 05 and FDR-adjusted p-value of the test statistic <0. 05 were used for further analysis.

CTCF is a transcriptional regulator acting as an insulator element interfering with enhancer function and as a boundary between chromatin domains. CTCF has been shown to organise an exquisite array of phased nucleosomes flanking its binding sites. Here we identified SNF2H as the enzyme primarily responsible for organising the extended arrays of nucleosomes adjacent to CTCF sites. We find that SNF2H acts to maintain the occupancy of CTCF at its binding sites, but does not act as a general loading factor for CTCF’s binding partner cohesin. SNF2H’s action at CTCF sites is functionally important as overlapping cohorts of genes are affected by depletion of CTCF or SNF2H. Other transcription factors also organise nucleosomes and we find that the SNF2H and the related enzyme SNF2L contribute to organising nucleosomes at many of these sites.

Abstract Introduction Results Discussion Materials and Methods

gene regulation regulatory proteins dna-binding proteins nucleosome mapping transcription factors epigenetics molecular biology techniques chromatin research and analysis methods small interfering rnas gene mapping chromosome biology proteins genetic interference gene expression molecular biology nucleosomes biochemistry rna cell biology nucleic acids genetics biology and life sciences non-coding rna

2016

The Chromatin Remodelling Enzymes SNF2H and SNF2L Position Nucleosomes adjacent to CTCF and Other Transcription Factors

Key steps in a viral life-cycle, such as self-assembly of a protective protein container or in some cases also subsequent maturation events, are governed by the interplay of physico-chemical mechanisms involving various spatial and temporal scales. These salient aspects of a viral life cycle are hence well described and rationalised from a mesoscopic perspective. Accordingly, various experimental and computational efforts have been directed towards identifying the fundamental building blocks that are instrumental for the mechanical response, or constitute the assembly units, of a few specific viral shells. Motivated by these earlier studies we introduce and apply a general and efficient computational scheme for identifying the stable domains of a given viral capsid. The method is based on elastic network models and quasi-rigid domain decomposition. It is first applied to a heterogeneous set of well-characterized viruses (CCMV, MS2, STNV, STMV) for which the known mechanical or assembly domains are correctly identified. The validated method is next applied to other viral particles such as L-A, Pariacoto and polyoma viruses, whose fundamental functional domains are still unknown or debated and for which we formulate verifiable predictions. The numerical code implementing the domain decomposition strategy is made freely available. The genomic material of many viruses is encapsidated inside icosahedral protein shells with diameters in the 20–100 nm range. The number of structurally inequivalent protein units that tessellate these capsids is usually very small [1], [2]. This, in turn, is reflected in the limited repertoire of viable capsid shapes with icosahedral symmetry [3]. Understanding the organization of viral capsids at levels that are intermediate between the single protein units and the fully assembled, infectious particles is crucial to elucidate key aspects of the viral life cycle. These include the molecular basis of capsid conformational changes, such as swelling or maturation events [4], as well as the assemby/disassembly of virion particles [5]–[8]. Both these processes, in fact, are best characterised and rationalised in terms of the typically multimeric protein units [9] that behave as approximately rigid units in the capsid' s conformational mechanics or act as basic assembly/disassembly units. The identification of these units has so far been carried out for few viruses using advanced experimental or numerical techniques for probing and modelling capsids assembly/disassembly kinetics and thermodynamics, internal dynamics and response to mechanical stress [10]–[21]. These approaches have proved extremely valuable to gain insight into various mechanisms controlling the physico-chemical behaviour of few specific viruses [10]–[12], [14]–[16], [22]–[24]. For instance, nano-indentation experiments, where viral particles are subject to mechanical stress and fatigue by atomic force microscopy, have singled out the mechanical building blocks of viral capsids and elucidated the mechanisms of genome uncoating [25]. However, the systematic application of these techniques has been hindered either by the difficulty of transferring the methodologies across different virus types or by their severe experimental/computational demands. As a step towards developing a general scheme for identifying functional and structural units in viral shells, here we introduce and apply a novel and efficient computational strategy that can single out capsid domains that, according to various criteria, are expected to be mechanically stable. The method consists of a decomposition of the capsid into quasi-rigid units based on a suitable analysis of its internal dynamics. In accord with the mesoscopic spirit of the approach, the sought internal dynamics can be efficiently obtained from elastic network approaches, in place of computationally-demanding molecular dynamics simulations. The variational decomposition strategy is applied to several viruses covering a wide range of sizes and capsid classes, from T = 1 to pT = 7. For validation purposes, the set includes several well-characterised instances: the cowpea chlorotic mottle virus (CCMV), the MS2 virus, the satellite tobacco necrosis virus (SNTV) and satellite tobacco mosaic virus (STMV). The units obtained from the decomposition are in excellent agreement with known basic blocks of the assembly/disassembly process or of the structural transitions. These successful comparisons give confidence in the viability of the strategy for identifying putative functional units of viral capsids. This suggests that the method could be profitably used for interpreting viral assembly, disassembly and genome uncoating experiments or as a predictive tool. Towards this latter goal, we conclude the present study by formulating predictions for a number of viruses whose capsid structure is available but whose functional units are still unknown, or debated. This prediction set includes the L-A (pT = 2), Pariacoto (T = 3) and polyoma viruses (pT = 7). The decomposition algorithm, which is formulated in a general and hence transferable way, is made freely available for academic use at the link: http: //people. sissa. it/~michelet/vircapdomains. We now turn to discuss three viruses for which the basic, mechanically stable functional units are not conclusively known. The following viruses are considered, chosen in order of increasing complexity of the capsid type (T-numbers): the L-A (pT = 2), Pariacoto (T = 3) and polyoma (pT = 7) viruses. We recall that the pT = 2 and pT = 7 cases refer to non-standard Caspar-Klug geometries. Identifying the fundamental, and typically multimeric, protein units that control the mechanical response of viral capsids or its assembly and disassembly is important both for rationalizing and for modeling key steps of viral life cycles [15]. Here we introduced and applied a novel computational strategy that, to our knowledge, represents the first attempt to develop a general and efficient method for identifying the basic, mechanically stable protein units starting from the sole input of the fully-assembled protein capsid. The method relies on the characterization of the internal dynamics of the capsid by means of elastic network models and uses it to optimally decompose the protein shell into blocks that have the characteristics expected for genuine capsid functional units, such as mechanical stability (quasi-rigidity), structural integrity of the constitutive proteins, or small numbers of inequivalent block types etc. The viability of the scheme was first assessed and validated by considering a set of four viruses (CCMV, MS2, STNV, STMV) for which the fundamental functional units are known. In all cases, the results of the optimal decomposition scheme were fully consistent with available experimental or numerical results for the known mechanical and/or assembly protein units. We next turned to a further set of three viruses, namely polyoma, Pariacoto and L-A virus, whose functional units are debated or not known, and for which we formulate verifiable predictions. The positive validation of the method and its affordable computational cost (the first hundred ENM modes of the internal dynamics of capsids of about 60000 amino acids can be obtained in hours on a single Intel Xeon 2. 40 GHz processor) demonstrate that simple structure-based strategies can provide considerable information on the basic functional units. In particular, they not only aid the understanding of various viral processes but can also guide the development of their multiscale modelling. We envisage two natural extensions of this first study. On the one hand it would be important to explore the possibility to include, even approximately, the interaction of the coat proteins with the packaged genome. This would be an apt complement of previous studies which considered the viability of ENM characterizations of empty capsid shells as proxies for the genome-loaded virion particles. On the other hand, it would be most interesting to extend considerations systematically to larger and more complex capsid geometries in order to understand how the functional units change as one goes from small- or medium-sized capsids (where the discrete protein nature of the capsid is visible) to larger structures that are well approximated by continuum theory [9]. Proteins and protein assemblies in thermal equilibrium can sustain structural fluctuations of appreciable amplitude. A large body of experimental and numerical evidence has indicated that the principal fluctuation modes, those of lowest energy, have a collective character. This means that the structural deformations associated to these modes entail the concerted displacements of groups of several amino-acids. As was first shown by Tirion [41], the collective character of the modes justifies the use of simplified, coarse-grained models (rather than atomistically-accurate ones) for calculating the principal modes of fluctuation of a protein around its reference, native structure. A commonly used framework for such coarse-grained calculations is provided by elastic network models. The latter rely on a quadratic approximation of the near-native protein free energy, (1) where is the number of amino acids, is the vector displacement from the native position of the main chain (backbone) centroid (typically the atom) and is the effective symmetric interaction matrix of linear size. Within the quadratic approximation of Eq. (1), the principal modes of structural fluctuations can be calculated exactly with minimal computational expenditure, and they correspond to the eigenvectors of having the lowest non-zero eigenvalues. In the following we shall indicate by, the non-zero eigenvalues ranked according to increasing magnitude (they are all positive) and with the corresponding orthonormal eigenvectors. It can be shown that corresponds to the total mean square structural fluctuation projected on the mode, , where denotes the canonical equilibrium average and is the displacement of the amino acid projected on the mode. In this study, we shall resort to the beta-Gaussian network model [45] to compute the matrix and its eigenvalues and eigenvectors. The model, which is implemented in a freely-available numerical code [45], was previously successfully validated against extensive molecular dynamics simulations of various proteins and protein complexes. At variance with most elastic network approaches it uses not one, but two interaction centers per amino acid: one for the main chain, the other for the side-chain (omitted for glycine). As customary, the centroids' interaction range was set equal to Å. Because the side-chain degrees of freedom are integrated out analytically, the linear size of the matrix is still equal to, as in single-centroid schemes. The computational burden associated with the memory storage and diagonalization of the matrices for the capsids (is in the range) was limited by taking advantage of the sparse character of and calculating its lowest-energy eigenvectors using the shift-inverse Arnoldi method, as implemented in the Arpack routines [82]. These algorithmic techniques (which could be further aided by symmetry considerations [83]) sufficed to compute the relevant low-energy modes of all capsids, except for L-A, using less than 24 Gb of RAM and a single 2. 4 GHz Intel processor. The modes calculation is the slowest computational step in the whole decomposition procedure for larger viruses (for instance, it took about 3 hours for the L-A case). For the polyoma capsid alone which, at, is the largest entry in our set, we found it necessary to adopt a coarser ENM description. Specifically, we used one centroid per two amino acids by retaining only one for every other. The interaction range was rescaled accordingly and set equal to 15 Å. Consistent with established results for the case of globular proteins [84], this coarse-graining procedure has no effect on the optimal quasi-rigid domain decomposition of smaller capsids. This is illustrated in Fig. S4 for the STMV capsid which, being the smallest considered here, is expected to be the most susceptible to the coarse-graining level. This validation and the considerations of [84] provide a justification for the use of the coarse-grained description for the polyoma capsid. The subdivision of viral capsids into quasi-rigid domains is based on the PiSQRD strategy introduced in refs. [33], [34]. The approach relies on the notion that for a genuine rigid-body the modulus of the distance of any two points remains constant as the body is moved in space. Accordingly, one can quantify the viability of a tentative capsid subdivision into putative quasi-rigid domains by comparing amino acids' pairwise distance fluctuations within each domain with those across domains. For good subdivisions, the former should be much smaller than the latter, see sketch in Fig. 5. To turn this observation into a quantitative scheme amenable to numerical implementation, we consider the geometric strain, , for a given pair of amino acids, and. Using the same notation introduced after equation (1) for the principal modes of structural fluctuations, , and their associated amplitudes, (2) where is the reference, native distance vector of the and amino acid, and is the number of retained principal modes. is chosen by retaining all the modes with energy lower than the fifth non-zero mode of a single coat protein, thus ensuring a sufficient level of detail while minimizing the computational effort and discarding the mostly irrelevant high-frequency details. Accordingly, the internal strain of the domain is defined aswhere the sum runs over all the pairs belonging to that domain, and the overall strain is thereforeBased on previous considerations, the desired subdivision is the amino acid partitioning into groups that minimizes the overall strain. Notice that the minimization of needs to be performed separately for all possible values of, that is from 2 up to the number of protein units forming the capsid (or even larger values in case the mechanical domains involve protein structural subunits). In fact, the “correct” optimal number of quasi-rigid domains is not known a priori and needs to be found based on physical considerations, see the next subsection. For each explored value of, the minimization of over the amino acids' assignments is performed by a greedy algorithm starting from a random labelling. At each step of the algorithm a randomly-picked amino acid is reassigned to a randomly-chosen domain. The new assignment is accepted if it leads to a decrease of and rejected otherwise. The scheme is repeated until the algorithm is unable to further improve the solution, i. e. the count of systematically rejected moves is comparable with the total number of amino-acids. To reduce the impact of getting trapped in local minima of (whose landscape roughness increases with) the greedy minimization scheme is iterated if the distribution of the domain strain, is highly heterogeneous (which could be a sign of a very asymmetric solution). Specifically, we first compute the average, , and standard deviation, , of the domains' strain and check if one or more residuals is larger than. If so, then the two domains with smallest strain are joined while the one with the largest strain is split in two. This amino acid reassignment clearly preserves the total number of domains, . The greedy minimization of is repeated and the procedure is iterated until one of the following holds: (i) convergence to a minimum which features a sufficiently homogeneous energy distribution, or (ii) the splitting/joining move is unable to improve the solution. It is important to note that no a priori information about the capsid' s parsing into single proteins is used to identify the domains. Indeed, in principle mechanical domains can cut through proteins, for example when a rather loose loop tightly binds to a different block. The comparison between the mechanical and the proteins boundaries is done a posteriori, providing information on the reliability of the subdivision itself. The -dependence of the miminized geometric strain for all considered capsids is shown in Fig. S5. Besides calculating the geometric strain, the genuine quasi-rigid character of a given decomposition into domains is more intuitively assessed by computing the fraction of overall capsid motion that can be ascribed to the relative rigid-like movements of the domains (i. e. by neglecting intra-domain fluctuations as if the domains were strictly rigid). This quantity is calculated by considering that each normalised mode, , can be decomposed as a sum of two contributions: one consisting of pure rigid rotations and translations of the domains, , and one describing intra-domain fluctuations, , i. e. . Because these two components are orthogonal [33] one has that. The fraction of the capsid' s mean square structural fluctuations that can be ascribed to the relative rigid displacement of the domains is accordingly: The profile of the fraction of motion captured by the domain decomposition of all considered capsids is shown in Fig. S2. The algorithm for the subdivision into domains was applied to the viral capsid several times, varying between 2 and the total number of proteins in the capsid. After establishing the quasi-rigid character of the putative subdivisions by monitoring the strain and the above-mentioned fraction of captured motion, the identification of the optimal value (s) of corresponding to a subdivision into viable, basic mechanical units was performed by monitoring two physical quantities: protein integrity and the number of inequivalent capsomere types. They respectively account for the compatibility of the subdivision with the natural elementary units represented by the single proteins and for the structural similarity of the tiles, which results in a low number of different tiles. Given a subdivision into domains, an integrity parameter was defined for each protein. For a general subdivision, the amino acids of a protein can be assigned to a number of different domains. However, a good subdivision should preserve the integrity of the protein, i. e. almost the whole protein should belong to a single domain. We thus defined the integrity score for a protein as the largest fraction of its amino acids assigned to a single domain. This quantity was then averaged for all the proteins, providing a score for the capsid subdivision. We also computed the number of similar tiles identified by our subdivision by size inspection. Specifically, we defined the size of the domain as the number of amino-acids belonging to the domain itself; we then assigned domains to a tile type if their size is the same within ca. 3% of the average size. Viable subdivisions into basic mechanical units were identified by maxima in the integrity score corresponding to a small number of tile types. To detect possible intertwinings between quasi-rigid units (e. g. due to swapped tails or subdomains of the parent proteins) we computed the interlocking parameter. Specifically, we considered separately the two termini of each protein in the capsid, namely the first and last twenty amino acids, and counted the number of amino acids assigned to a rigid domain different from the dominant one (i. e. the domain to which most of the protein' s amino acids belong). This calculation returned the number of interlocked amino acids for each terminus of each protein in the capsid. The numbers relative to the N and C terminals were averaged separately, and the largest of the two averages was taken as a measure of the interlocking of the quasi-rigid domains. In other words, if a quasi-rigid domain subdivision has interlocking number equal to 10, it means that on average one protein has 10 terminal residues assigned to a different domain than its core. Clearly, this also implies that the other terminus has less than 10 interlocked amino acids.

The genetic material of viruses is packaged inside capsids constituted from a few tens to thousands of proteins. The latter can organize in multimers that serve as fundamental blocks for the viral shell assembly or that control the capsid conformational transitions and response to mechanical stress. In this work, we introduce and apply a computational scheme that identifies the fundamental protein blocks from the structural fluctuations of the capsids in thermal equilibrium. These can be derived from phenomenological elastic network models with minimal computational expenditure. Accordingly, the basic functional protein units of a capsid can be obtained from the sole input of the capsid crystal structure. The method is applied to a heterogeneous set of viruses of various size and geometries. These include well-characterised instances for validation purposes, as well as debated ones for which predictions are formulated.

Abstract Introduction Results/Discussion Methods

2013

Mechanical and Assembly Units of Viral Capsids Identified via Quasi-Rigid Domain Decomposition

HIV-1 is subject to immune pressure exerted by the host, giving variants that escape the immune response an advantage. Virus released from activated latent cells competes against variants that have continually evolved and adapted to host immune pressure. Nevertheless, there is increasing evidence that virus displaying a signal of latency survives in patient plasma despite having reduced fitness due to long-term immune memory. We investigated the survival of virus with latent envelope genomic fragments by simulating within-host HIV-1 sequence evolution and the cycling of viral lineages in and out of the latent reservoir. Our model incorporates a detailed mutation process including nucleotide substitution, recombination, latent reservoir dynamics, diversifying selection pressure driven by the immune response, and purifying selection pressure asserted by deleterious mutations. We evaluated the ability of our model to capture sequence evolution in vivo by comparing our simulated sequences to HIV-1 envelope sequence data from 16 HIV-infected untreated patients. Empirical sequence divergence and diversity measures were qualitatively and quantitatively similar to those of our simulated HIV-1 populations, suggesting that our model invokes realistic trends of HIV-1 genetic evolution. Moreover, reconstructed phylogenies of simulated and patient HIV-1 populations showed similar topological structures. Our simulation results suggest that recombination is a key mechanism facilitating the persistence of virus with latent envelope genomic fragments in the productively infected cell population. Recombination increased the survival probability of latent virus forms approximately 13-fold. Prevalence of virus with latent fragments in productively infected cells was observed in only 2% of simulations when we ignored recombination, while the proportion increased to 27% of simulations when we allowed recombination. We also found that the selection pressures exerted by different fitness landscapes influenced the shape of phylogenies, diversity trends, and survival of virus with latent genomic fragments. Our model predicts that the persistence of latent genomic fragments from multiple different ancestral origins increases sequence diversity in plasma for reasonable fitness landscapes. Patients infected with HIV-1 require lifelong highly active antiretroviral therapy (HAART) to suppress infection. Treatment cessation typically leads to HIV viral rebound to pre-therapy levels; the resurgence is thought to be associated with activation of long-lived, latently HIV-infected cells. A cure for HIV therefore requires either clearance of all cells harboring latent virus, or prevention of virus release from the reservoirs after discontinuation of treatment. Increasing evidence suggests that latency plays an integral role throughout the life cycle of the virus. We recently observed that the majority of HIV-1 plasma sequences in two untreated chronically infected patients had accumulated significantly less mutations than expected, suggesting a period of latency during which no replication occurred in the history of these lineages [1]. Moreover, viral variants with reduced evolution consistent with periods of latency are frequently involved in transmission events [2–4]. While recent advances have shed light on the mechanisms leading to the establishment and maintenance of latent reservoirs [5,6], the prevalence of viral sequences displaying a signal of latency in the replicating population remains enigmatic, especially in the absence of antiviral treatment. HIV-1 is subject to selection pressure exerted by the immune system; strains that can avoid the immune response have an advantage within a host. The neutralizing antibody response in patient sera is much stronger against virus circulating in infection earlier than contemporaneous variants, with immunological memory persisting for years [7–9]. Virus from activated latent cells is therefore less fit due to long-term immunological memory than variants that have continually evolved in response to the host immune pressure. Yet, we detect a signal of latency in the replicating population, suggesting that virus from activated latent cells persists despite competing against better-adapted contemporaneous variants. In order to survive, activated latent forms must either have a replicative advantage outweighing immune selection, or quickly acquire escape mutations to evade the immune response. While mutations associated with immune escape decrease the replicative capacity of the virus between 0–24% [10], compensatory mutations can completely restore fitness in some variants [11]. It is thus unknown whether virus from activated latent cells is able to compete against better adapted contemporaneous virus due to higher replicative fitness. To persist in plasma after activation, the virus must adapt to immune pressure before being outcompeted. The virus may adapt by sequentially accumulating escape mutations at recognized epitopes, or by recombining with contemporaneous virus and simultaneously gaining multiple epitopes not recognized by the immune system. Recombination provides a mechanism to preserve genomic fragments originating from activated latent cells within contemporaneous virus backbones adapted to the immune response. Further recombination events may propagate latent genomic fragments through the replicating plasma population, enhancing the adaptability potential of the virus. In this paper, we investigate the persistence of latent envelope genomic fragments in the replicating plasma population by developing a detailed model of HIV-1 sequence evolution and the cycling of lineages in and out of the latent reservoir. There has been an increasing focus on modeling the dynamics of the latent reservoir, particularly in the context of delineating the effect of latency on viral blips during treatment, and the re-emergence of infection after treatment failure/interruption [12–16]. A common feature of these models is the cycling of infected cells through a latent compartment, in which they may divide or die before becoming activated and re-joining the productively infected cell population. Our agent-based model integrates the dynamics of the latent reservoir proposed in [12–14], with a sequence evolution framework first introduced in [17,18], which follows how individual viruses mutate and recombine. Vijay et al. developed an HIV sequence evolution model of infected patients that incorporates mutation, infection of cells by multiple virions, recombination, fitness selection, and epistatic interactions between multiple loci [17]. Their model, which describes quantitatively the evolution of HIV-1 diversity and divergence in patients, has been applied to a wide variety of questions, including the effect of recombination on sequence diversification [17], the effective population size of HIV-1 [18], the genetic structure of HIV-1 quasispecies and its error threshold [19], and the fraction of progeny viruses that must incorporate a drug treatment target for suppression of productive infection [20]. We extend the sequence evolution model by incorporating a latent compartment in which no evolution takes place, and by explicitly simulating how the virus population interacts with and adapts to the immune response. To evaluate whether our model captures sequence evolution in vivo, we compare our simulated sequences to sequences from 16 untreated or unsuccessfully treated patients followed longitudinally from seroconversion [9,21,22] in terms of sequence divergence (i. e. average distance of a population of sequences from a founder sequence), diversity (i. e. average pairwise distance between all sequences in a population), and phylogenetic tree shape measures. We developed an agent-based model to simulate within-host HIV-1 sequence evolution after primary HIV-1 infection (PHI), tracking how long each virus lineage has spent in the latent reservoir throughout its history (Fig 1). The model incorporates a detailed mutation process including recombination, latent reservoir dynamics, diversifying selection pressure driven by the immune response, and purifying selection pressure asserted by deleterious mutations. During each generation, a population of virus infects a population of uninfected cells, with infecting viruses chosen according to their relative fitness. Genetic variability is introduced through recombination and substitution during reverse transcription of viral RNAs to proviral DNA in infected cells. Infected cells have a nonzero probability of moving into the latent reservoir, while latent cells have a nonzero probability of dying, proliferating, or becoming activated. Infected cells and activated latent cells produce progeny virus, forming the population of virus infecting the next generation of cells. The relative fitness of each virus depends on whether it has acquired mutations at invariant sites, which are considered deleterious, and how well it is recognized by the immune system. All virus sequences have multiple epitopes at pre-defined sites, which are presented to the immune system. If an epitope variant reaches a sufficient frequency in the plasma, it elicits an immune response, which remains in memory for the duration of the simulation. The breadth of the immune response is updated each generation by adding any newly recognized epitope variants to immunological memory. While we simulate infections established by a single founder strain, the model can also accommodate multiple founder strains, which are observed in 20% of heterosexual cases [23]. The simulations were implemented using a computer program written in R [24], and the phylogenies were created using the R package ape (Analyses of Phylogenetics and Evolution) [25]. The algorithm is outlined in the supplement (S1 Text). We initialized the simulation by generating a set of NV identical virions, where each virion consists of a string of L nucleotides (A, U, G, C). While HIV contains two RNA molecules, we considered only a single RNA molecule for each virus to simplify computations. Because recombination between two identical parental strains does not lead to new genetic variants, restricting our simulations to one RNA molecule per virus should have a negligible effect on our simulation results. We chose as the starting sequence the consensus of the first time point sequences, sampled close to seroconversion, of Patient A from the Amsterdam cohort [9]. The sequence is from env, and consists of approximately 700 nucleotides. At each generation, NC viruses are chosen to infect NC uninfected cells such that each cell is infected with a single virus. Multiple infection of cells is described below. The probability of selecting a given virus is proportional to its relative fitness, which is a function of the strength of the immune response against it, and the number of deleterious mutations it has acquired. Fitness selection is described in detail below. Upon infection each cell has a small probability η of becoming latent and moving to the latent reservoir. The virus in the infected cells that remain productive undergo reverse transcription, where the viral RNA is copied into proviral DNA. This process includes both recombination and substitution. To simplify computations, we only considered dual infections that lead to recombination. Based on estimates of Josefsson et al. [26], we assumed that 10% of cells are infected with a second virion, ignoring multipe infection with three or more virions. On average half of the progeny virus produced by a cell infected with two virions acquires one RNA molecule from each parent. To simplify computations, we allowed at most one cross-over event per sequence. This simplification is justified because while there are on average 2. 8 cross-overs per each HIV-1 genome of 10,000 nucleotides every life cycle [27], our simulated env sequences were over 10 times shorter. For a template switching rate of 3 × 10−4 per site per generation [27,28], the expected number of cells accommodating a recombination event between two different parental strains is NR = 0. 1 × 0. 5 × 3 × 10−4 × NC L. We incorporated dual infection into the model by sampling NR additional viruses from the virus population with probability proportional to their relative fitness and adding them to randomly chosen infected cells. The dually infected cells then undergo recombination. We assumed that cross-over events are distributed uniformly across the genome, selecting the starting strand and the cross-over position ∈ {2, L − 1} randomly. The sequence of the starting strand is copied until the cross-over position, the strand is then switched, and the second strand is copied to the end of the sequence. The recombinant sequence replaces the two parent strands, so that all infected cells contain a single viral sequence. The sequence is then mutated according to a general-time-reversible substitution model [29]. We used the nucleotide-specific substitution rates estimated from the sequences of Patient A via maximum likelihood (PAUP*, [30]), which were representative of the rates found for all 16 patients. The proviral DNA of each infected cell is transcribed to viral RNAs, which are released as new virions. Each infected cell produces P identical virions, which form the current population of virus from which viruses that infect the next generation of cells are selected. As the latent reservoir is established early in infection [31,32], we initialized the reservoir by infecting each of the NL cells with the same founder virus used to infect the replicating population. During effective antiretroviral treatment, the reservoir’s half-life has been estimated to range from 4 to 44 months [33–36]. Assuming that the size of the latent reservoir, NL, remains constant during untreated chronic infection, we set the probability of an infected cell becoming latent equal to the decay rate of the latent reservoir. Each generation, we randomly select which (if any) infected cells become latent. For simplicity, we assumed homogeneous activation and death rates, randomly selecting which cells in the latent compartment die or become activated each generation. Activated latent cells join the population of productively infected cells and produce progeny virus. We assumed that the latent reservoir is maintained through homeostatic proliferation [12,13,37], which we simulated by randomly duplicating cells to replace those lost to death and activation to keep the size of the latent pool constant. Specifically, if the size of latent reservoir NL after the addition of any newly latent cells and the removal of any activated or dead cells is less than the target size N L *, we randomly duplicate N L * - N L of the remaining cells in the latent compartment. No proliferation takes place if the size of the latent reservoir exceeds N L *. The model tracks how many generations each cell spends in the latent reservoir before being activated, with progeny viruses retaining the information for the duration of the lineage. Because viruses that have been latent for different amounts of time can recombine, we track latency separately at each site in the genome. If a virus that has never been latent recombines with a virus that has been latent for n generations, we consider the recombinant progeny a latent virus form, with the portion of its genome acquired from the non-latent parent recorded as having been latent for zero generations, and the portion acquired from the latent parent as having been latent n generations. At the end of the simulation, we know the position and age of every latent genomic fragment for each virus sequence. We are interested in the persistence of sequences that contain at least one latent genomic fragment consisting of at least one site originally from a latent ancestor. The immune response against antigens with recognized epitopes consists of two major arms; neutralization of a pathogen via antibodies, and killing of infected cells by cell-mediated immune responses. Because we simulated the viral envelope, we focused primarily on the antibody response, assuming that 1) variants with a recognized epitope are less fit than those able to evade the immune response, and 2) once an epitope is recognized, it is retained in memory for all time. The immune response imposes diversifying selection pressure on the virus population, resulting in sequential escape at each epitope. A virus variant not recognized by the immune system has a fitness advantage, and starts to take over the virus population, eventually eliciting an immune response, and thus selection for new escape variants. We chose the location of NE epitopes randomly in the envelope fragment that we are simulating. Epitopes are approximately 30 nucleotides in length, and may be non-contiguous and overlapping [38]. As synonymous mutations at the nucleotide level have no effect at the amino acid level, we ignored third-codon positions, and assumed that the epitopes are 20 nucleotides long and non-overlapping. Serum from HIV-1 infected individuals does not neutralize contemporaneous virus, but rather virus that dominated the population 3 to 6 months earlier [7]. To model this behavior, we imposed an immune response against a new escape variant not when it first appeared but after it had risen in frequency in the plasma, and increased the strength of the subsequent response gradually over time. We began the immune response against the virus after 30 generations. Every generation thereafter, we determined which epitope variants had reached a sufficient frequency, f, in the plasma, and initiated an immune response against them. The epitopes were stored in immunological memory for the duration of the simulation, and all virus variants containing them had a fitness disadvantage. For simplicity, we assumed that to be recognized the nucleotide sequence of a virus epitope must match perfectly with the sequence of a stored epitope. Note that virus may be recognized at multiple epitopes simultaneously. Because antibody responses mature over time, we assumed that the fitness cost to the virus imposed by the immune response to epitope i of variant j increases linearly until it reaches some maximum value c i *. Thus, the fitness cost of epitope i of variant j is given by c i j (t) = m i n { c i *, c i * (t - t i j 0) / d } δ i j, t ≥ t i j 0, where t i j 0 is the time when the immune response against epitope i of variant j is introduced, d is the time it takes to reach full potency, and δij is the Dirac delta function, where δij = 1 when epitope i of variant j is recognized, and δij = 0 when it is not recognized. The strength of the immune response against a particular epitope depends in part on how accessible it is to antibodies/killer T-cells. While escape mutations appear shortly after seroconversion in HIV-1/SIV infection at some epitopes [10,39], it may take years to see evidence of selection at others [40]. We initialized the maximum fitness cost of each epitope in the beginning of the simulation by drawing it from a uniform distribution, ci*∼U [0, cmax]. We assumed that neutralization via antibodies is the primary driver of selection in chronic infection, and that virus is saturated with antibodies. In this scenario, the fitness loss of a virus variant upon antibody recognition is driven by the most potent antibody that binds it, c j i m m = m a x (c i j δ i j). The distribution of fitness costs associated with the epitopes defines the fitness landscape, which influences the evolutionary trajectory of the virus population. Purifying selection conserves sites in the HIV-1 genome where mutations would be deleterious to the virus. We assume that mutations at invariant sites (where no variability is observed across sequences from different time points) are inherently deleterious, and incur a multiplicative fitness cost, c j i n v = 1 - (1 - ψ) k, where ψ is the reduction in fitness per mutation, and k is the number of mutations. At the beginning of each simulation, we randomly distribute Lpinv invariant sites across the sequence, where pinv is the proportion of invariant sites. Note that the positions of the invariant sites vary between simulations, and may occur in both non-epitope and epitope regions. Following Ganusov et al. [41], we define the relative fitness of variant j as f j = (1 - c j i m m) (1 - c j i n v) = (1 - m a x (c i j δ i j) ) (1 - ψ) k. We considered a pool of NC = 15000 infected cells, which is in line with the mean effective population size of HIV-1 in chronic infection [18]. HIV-1 has a large burst size of approximately 50,000 [42], but only a small fraction of one in 1000 to 10,000 virions appear to be infectious [43–45]. Each productively infected cell therefore produces between 5 and 50 infectious virions. For consistency with the sequence evolution models developed by Vijay et al. and Balagam et al. , we set the number of progeny virions produced by each productively infected cell to P = 5 [17,18]. Following again Balagam et al. , we set the generation time to 1. 2 days [18], and the substitution rate to the mean of the mutation rates estimated by Mansky et al. , μ = 3. 5 × 10 − 5 [46]. We set the proportion of invariant sites to 50%, corresponding to the mean proportion of invariant sites estimated from the HIV-1 sequence alignments of patients described in Bunnik et al. and Karlsson et al. [9,22]. We ran the simulations for 3000 generations, corresponding to 10 years. To maintain the ratio of productively infected to latent cells predicted by the homeostatic proliferation model of latency introduced by Kim et al. and Rong et al. , we set the size of the latent reservoir to NL = 100 [12,13]. Assuming a total body load of 2 × 108 productively infected cells [47] and between 2. 2 × 105 to 1. 6 × 107 latent cells with replication competent DNA [48], there are between 0. 0011 to 0. 08 latent cells for every productively infected cell. In our simulations with 15,000 productively infected cells, this corresponds to a latent reservoir size NL of 16 to 1200 cells; our chosen latent reservoir size is close to the geometric mean of this range. We chose a conservative estimate of 44 months for the half life of the latent reservoir [33,36], which we used to calibrate the probability η of an infected cell becoming latent upon infection such that the size of the latent reservoir was maintained. Archin et al. recently estimated that the composite parameter βη is on the order of 10−14 [15], where β is the mass-action infection rate constant in ml−1 day−1. The latter has been estimated to be approximately β = 1. 5 × 10−8 [49,50]. Our value of η = 3. 5 × 10−6 is therefore in line with these estimates. We set the death rate of latent cells to dL = 0. 004 per generation, corresponding to the estimated death rate of memory cells, which make up the bulk of the latent reservoir [13,14]. Following Rong et al. , we varied the activation rate from aL ∈ (0. 001,0. 01) per generation so that between 0. 1 and 1 latent cells were activated each generation [13,14]. Barr et al. estimated that the minimum efficacy of three early neutralizing antibodies at blocking de novo infections ranged from 19. 6% to 35. 2% [10]. To account for the observation of Richman et al. [7] that some antibodies isolated from patient serum show no neutralization activity, we allowed the maximum fitness cost, c i *, due to recognition of a viral epitope to range from 0 to 0. 40, with a mean of 0. 2. The remaining parameters associated with the immune response are not well known. For our default parameter setting, we assumed that there are ten epitopes in the approximately 700 nucleotide region of envelope that we simulate. In subsequent sensitivity analyses, we varied the number of epitopes between 0 and 15. While it is not known how many epitopes there are on the envelope, analysis of escape mutations to SIV in rhesus monkeys suggests that there are several [51]. We assumed that an antibody is created against an epitope variant once it reaches 1% frequency in the plasma. In sensitivity analyses, we varied the antigen frequency required for stimulating a new antibody response between 0. 1% and 10%. Because patient serum does not neutralize contemporaneous virus but rather virus that circulated in the plasma at least 3 months earlier [7], we assumed that it takes 90 generations for a newly introduced immune response to reach its full potency (i. e. impose maximal fitness cost to the virus variant it recognizes, with fitness cost increasing linearly from zero over the 90 generations). The default values of the model parameters are summarized in Table 1. To investigate the robustness of our results, we also considered alternative parameter values proposed in the literature. Recent estimates of the basic reproductive number of HIV-1 during primary infection suggest that an infectious cell generates at least eight new infected cells at the start of infection when target cells are not limiting. Thus, each infected cell produces at least P = 8 infectious progeny virions [52]. Following Pearson et al. , we set the number of progeny virions to P = 10 to account for viral clearance [53]. We set the generation time of HIV-1 in vivo to two days, estimated from the decay dynamics of productively infected cells [54], and the substitution rate to the reverse transcriptase nucleotide substitution rate of approximately 2. 2 × 10−5 [46,55]. We further assumed that 10% of sites in the envelope region of interest are invariant, corresponding to the minimum proportion of invariant sites in the patient data. Because the mechanism of recombination is central to our investigation, we also varied the number of recombination events per generation between 0. 1NR, and 10NR. Estimating sequence divergence and diversity allows us to quantify viral evolution. Divergence is a measure of how far viral genomes have evolved from the founder strain, whereas diversity is a measure of the genomic variation in the viral population at any given time. Every 30 generations, we calculated the divergence and diversity of the HIV-1 sequences in the entire populations of productively infected and latent cells, recording the mean, median, and 5% and 95% quantiles. Furthermore, we randomly sampled 100 productively infected cells and 100 cells from the latent reservoir every 300 generations, storing the HIV-1 sequences for later phylogenetic analysis. Using d (i, j) to denote the mean number of differences per position between sequences i and j, divergence and diversity are defined as follows for a collection of k sequences: d i v e r g e n c e = 1 k ∑ i = 1 k d i, f o u n d e r d i v e r s i t y = ∑ i = 1 k - 1 ∑ j = 1 + 1 k d (i, j) 1 2 k (k - 1). Stochastic effects dominate at small population sizes—the smaller the population, the larger the probability that stochastic fluctuations lead to extinction. We therefore investigated how the size of the simulated population affects the ability of latent genomic fragments to survive in patient plasma, assuming that recombination is necessary for survival. To this end we determined the expected number of dual infections, a precursor for recombination, involving at least one latent form when the sizes of the replicating and latent populations were increased ten-fold. We ran 100 simulations, tracking the progeny of the activated latent cells for ten generations, introduced into a replicating population of size nr at generation zero. The relative fitness of each latent virus was set to 0. 5, while the relative fitness of each non-latent virus was drawn from a uniform distribution between 0. 5 and 1. As in our full model, each generation of the simulation consisted of every infected cell producing five progeny virions, and downsampling the virus population to infect the next generation of nr uninfected cells. The expected number of dual infections involving at least one latent virus was then estimated for each simulation run. Simulation results were compared to HIV-1 DNA sequence data from 16 untreated or unsuccessfully treated patients [9,21,22]. All patients were infected with HIV-1 subtype B. The patients were followed longitudinally from seroconversion with 2–22 sequences sampled at intervals of 1–67 months (mean 13 months). The sequences were derived from the envelope, and were between 532–948 nucleotides long (mean 683 nucleotides). For each patient, we calculated the mean divergence and diversity at each time point the same way as for the simulated data. Note that in [21], sequence divergence and diversity were estimated from mean pair-wise distances determined using either a two-parameter Kimura model or a general time-reversible model with site-to-site variation in substitution rates, while here only simple pairwise Hamming distances were used. We considered the unsuccessfully treated patients as effectively untreated, because their viremia was not suppressed, and their divergence and diversity trends were similar to those of the untreated patients. We also estimated the proportion of invariant sites across all time points for each patient using PhyML [56]. The mean proportion of invariant sites across all patients was 39%. The lowest proportion of invariant sites was 13% for Patient 9 in the Shankarappa cohort, while the mean was greater than 50% for both the Amsterdam and Karlsson patients. To be sure that our model invokes realistic trends of HIV-1 genetic evolution, we compared our simulated patient HIV-1 populations generated by simulations incorporating recombination to HIV-1 DNA sequence data from 16 untreated or unsuccessfully treated patients [9,21,22]. Reassuringly, the empirical HIV-1 diversity and divergence trends over time were qualitatively and quantitatively similar to our simulated patient HIV-1 populations (Fig 2). Divergence increased linearly over time to a mean of 0. 06 substitutions per site at 10 years post-primary HIV-1 infection (post-PHI), while diversity initially increased rapidly, and then saturated to approximately 0. 05 substitutions per site at 10 years post-PHI. As seen in S1 Table in the supplement, there was no significant trend between the activation rate and mean sequence diversity (p = 0. 67, cor = 0. 15; Pearson’s product-moment correlation). Increasing the activation rate slightly decreased sequence divergence at the end of the simulations (p = 0. 04, cor = −0. 65; Pearson’s product-moment correlation). Unsurprisingly, both mean sequence divergence and diversity increased in the latent reservoir at a much slower initial rate than in productively infected cells (Fig 3). After approximately 6 years post-PHI, divergence in the latent reservoir proceeded at nearly the same rate as in productively infected cells. Because diversity in the latent reservoir grew linearly but started to saturate in plasma, by 10 years post-PHI diversity in the latent reservoir almost reached that in productively infected cells. Reconstructed phylogenies of the simulated patient HIV-1 populations, with 20 sequences sampled every 12 months, showed similar topological structures as patient HIV-1 populations. While the serially sampled HIV-1 phylogenies ranged from star- to ladder-like structures, they generally displayed a clear time trend (Fig 4). Both in the simulated and clinical data, the fraction of surviving phylogenetic lineages between samplings ranged from 0. 1 to 1. 0 (S1 Fig). Low survival of lineages corresponds to a ladder-like phylogeny while high survival corresponds to a star-like phylogeny. The tree shapes were largely explained by the fitness landscape, where convex profiles lead to more ladder-like trees, and concave profiles lead to bushier, star-like, trees. Intermediate fitness landscapes (nearly linear profiles) generated trees more typically observed in HIV-1 infected patients. Importantly, these intermediate landscapes also showed typical patient diversity trends (Fig 4). Interestingly, in intermediate and concave profiles we sometimes observed dramatic selective sweeps where the diversity dropped drastically and then recovered. The fitness landscape also influenced the number of latent origins (latent virus of different ages reaching at least 1% frequency in plasma virus population). The mean (s. d.) number of different latent origins was 0. 16 (0. 42) for convex landscapes, 0. 42 (0. 76) for nearly linear landscapes, and 7. 6 (7. 5) for concave ones. Thus, our simulations suggest that the fitness landscape influences general tree shape, diversity trends, and the survival of latent forms. The shape of the landscape is defined by the maximum fitness costs exerted by immune responses against different viral epitopes. The concave landscapes have a higher number of different immune responses exerting similar high fitness costs on the virus than convex or nearly linear landscapes. Escaping the immune response exerting the maximum fitness cost only marginally increases fitness if there are simultaneous immune responses against other epitopes that also impose high fitness costs. Therefore, a more concave fitness landscape exerts a stronger selection pressure on the virus, reducing both the mean and variance of the relative fitness in the virus population. The survival of less adapted latent forms in plasma therefore increases, while the reduction in fitness differences between virus variants in productively infected cells increases diversity. Fig 5 shows that the more genomic fragments with different latent origins persist in the plasma, the higher the sequence diversity (panel A). Compared to simulations with no surviving latent origins, sequence diversity became significantly higher after 9. 5 years post-PHI in simulations where there were between 1 and 5 latent origins (p < 0. 05, Wilcoxon rank-sum test), while in simulations where there were more than 5 latent origins, sequence diversity became significantly higher much earlier, after approximately 5 years post-PHI (p < 0. 05, Wilcoxon rank-sum test). At 10 years post-PHI, mean sequence diversity was 0. 038 substitutions/site when no latent genomic fragments survived, increasing by 16%, when there were between 1 and 5 latent origins, and by 95% when there were more than 15 latent origins. The propagation of genomic fragments that have accumulated fewer mutations than contemporaneous variants should reduce average sequence divergence in the plasma. As expected, when the number of surviving genomic fragments of different latent origins increased, mean sequence divergence decreased (Fig 5, panel B). Our model predicts that multiple introductions of latent forms into plasma are necessary for increased diversity; simulations with only one surviving latent origin did not have significantly higher diversity than those with no latent forms in plasma. Because sequence diversity increased almost linearly for concave fitness landscapes, the introduction of older virus forms into plasma did not further increase diversity (S2 Fig). Furthermore, because the survival of multiple latent origins was rare for convex fitness landscapes, it was not possible to determine whether latency could have had an effect on diversity under such conditions. However, the phylogenies and diversity trends generated by convex or concave fitness landscapes did not resemble those observed in typical patients. Fig 6 shows the effect of recombination on the proportion of virus with latent genomic fragments in productively infected cells in 2000 simulations of untreated patients. Our simulation results suggest that recombination facilitates the survival of latent genomic fragments. In simulations with recombination, 27% of the HIV-1 replicating plasma populations reached a mean of ≥ 10% virus with latent genomic fragments between 5–10 years post-PHI, while only 2. 0% reached ≥ 10% latent forms without recombination. Thus, recombination increased the survival of latent genomic fragments in productively infected cells approximately 13-fold. The proportion of virus with latent genomic fragments was higher when we allowed for recombination than when we ignored it. This distinction is clear from even the first sample time, after 30 generations (p < 0. 05, Wilcoxon rank-sum test). Furthermore, the rate of increase of virus with latent genomic fragments over time in the replicating population was approximately 9 times faster when recombination was involved Fig 6. In most simulations without recombination, virus with latent fragments immediately disappeared from productively infected cells upon introduction, while in some simulations, sharp peaks were observed where the proportion of latent forms increased rapidly to a high fraction but then fell quickly and disappeared completely. In simulations with recombination, the proportion of virus with latent genomic fragments often increased gradually, and stabilized both at intermediate values and the extremes. Most recombinant HIV-1 that include a fragment from a latent virus underwent relatively few recombination events; 96% had fewer than 5 latent fragments (Fig 7A). However, only 0. 7% of sequences with latent sites at 10 years post-PHI had not acquired genomic fragments from non-latent contemporaneous virus through recombination. The most common latent recombinant observed in our simulations had one latent fragment, with a mean (s. d.) of 85 (123) sites (Fig 7A). When such a recombinant proliferates further, additional recombination with contemporaneous virus removes sites from the latent fragment. Overall, the mean (s. d.) number of latent sites at 10 years post-PHI was 99 (121), while the mean (s. d.) fragment length was 66 sites (Fig 7B). When the number of latent fragments in a viral sequence increased, the fragment length decreased (Fig 7B). Interestingly, viruses that had more latent fragments also had more latent origins (Fig 7C). Hence, the most common forms of latent recombinants had few breakpoints, and multiple breakpoints frequently involved latent virus with different origins in time. Increasing the recombination rate increased the survival of virus with latent genomic fragments (S2 Table). As expected, the number of latent fragments per sequence also increased, while the fragment length decreased. Higher proportions of surviving lineages through time imply higher numbers of virus with genomic fragments of different latent origins that have long branches reaching far back into the phylogeny (Fig 8). When the prevalence of latent genomic fragments in productively infected cells was high, they were typically of different latency ages, i. e. , deposited and activated at different times (S3 Fig). These observations are supported by the noted increase in both the proportion of virus with latent genomic fragments and the observed number of different latent origins in productively infected cells when the activation rate was increased (S4 Fig). When a person becomes infected with HIV-1 the initial spike in viral load may deposit many similar copies of the initial virus in the reservoir. We modeled this by initially filling the reservoir with the infecting strain. Thereafter, a relatively slow deposit rate and proliferation rate refill and maintain a diversifying latent virus population. Interestingly, at 10 years post-PHI, our simulations predict that about 27% of the reservoir still consists of virus deposited in the first year of infection (Fig 9A). Conversely, the latent genomic fragments in the plasma population originated mostly from more recently deposited and activated virus (Fig 9B). Overall, the age structure among latent genomic HIV-1 fragments in the plasma of an untreated patient followed that of the latent reservoir with the exception of the initially deposited virus, which is strongly selected against upon activation by the refined immune response. The evolutionary trends of the virus populations simulated under the alternative model parameters (P = 10, μ = 2. 2 × 10−5, inv = 0. 1, τ = 2 days) were qualitatively and quantitatively similar to those seen previously under the default parameters. The mean sequence divergence and diversity of 100 simulations with the alternative model parameters are shown in (S5 Fig). We found that the prevalence of virus with latent genomic fragments in productively infected cells was much higher in simulations with recombination than without (S6 Fig). Our simulation results are therefore robust to the choice of parameters defining the infection and replication dynamics. The sensitivity of the model to the immune system parameters, i. e. , the antigen frequency necessary to elicit a new immune response, and the number of epitopes accessible to the immune system, are described in (S3 and S4 Tables), respectively. Fifty simulation runs were performed for each parameter value. When the antigen frequency required for stimulating a new immune response was increased from 0. 1% to 10%, the survival of virus with latent fragments and sequence diversity decreased (p = 0. 025, cor = -0. 67; p = 0. 013, cor = -0. 71, respectively; Pearson’s product-moment correlation), while sequence divergence increased (p = 0. 003, cor = 0. 79, Pearson’s product-moment correlation). When the immune response to a new escape mutant is delayed, the frequency of virus variants carrying the advantageous escape mutation increases. The increase in the number of virus variants that have evolved away from the founder sequence increases divergence but the simultaneous convergence towards the new escape variant reduces diversity. Because the mean relative fitness of virus in productively infected cells increases, the survival probability of virus from activated latent cells decreases. The survival of virus with latent genomic fragments was low in all simulations without recombination. On the other hand, when the number of epitopes was increased, sequence diversity at 10 years post-PHI initially decreased from one epitope to 4 epitopes and then increased from 4 to 15 epitopes (S4 Table). Sequence divergence initially increased from one to 4 epitopes, and then slightly decreased as the number of epitopes was increased. The survival of virus with latent genomic fragments was more stochastic, but generally displayed a similar trend. Virus with latent genomic fragments rarely survived in simulations without recombination. Therefore, our fundamental model result suggesting that recombination is an essential mechanism facilitating the survival of latent genomic fragments is robust to large perturbations in the largely unknown parameters describing the immune response. When the number of epitopes was zero, no immune pressure was exerted on the virus. This resulted in high diversity (0. 081 subst/site) because all virus variants without mutations at invariant sites had equal probability of surviving, and low divergence (0. 047 subst/site), because rapid accumulation of escape mutations was not selected for. Virus with latent genomic fragments survived in all simulations, because earlier virus forms were not at a fitness disadvantage compared to more recent variants. When the number of epitopes was increased from 1 to 4, it became increasingly more difficult but not impossible for virus to escape the immune response. When two epitopes had high fitness costs associated with them, it took longer for any virus variant in the population to simultaneously escape both responses, and the significant fitness gain and proliferation due to escape resulted in decreased diversity (S7 Fig). Such genetic sweeps were more dramatic when three epitopes had high fitness costs. When four epitopes had high fitness costs, escape at all epitopes to gain fitness was nearly impossible, and happened in only a small fraction of the simulations. Therefore, more concave fitness landscapes with four or more epitopes with high fitness costs imposed selection pressure resulting in high diversity (S7 Fig). The probability that a randomly drawn fitness landscapes had 2 or 3 epitopes with high fitness costs was highest when the total number of epitopes was 4 or 5. When the number of epitopes was increased from 6 to 15, the probability that each fitness landscape had 4 or more epitopes of high fitness costs also increased. Surviving in the replicating population long enough to recombine with a replicating virus forms the bottleneck that largely determines the long-term fate of a latent lineage. When one latent form was introduced into a population of 15,000 cells at generation t = t0, corresponding to the expected number of cells activated each generation from the latent reservoir of 100 cells at the highest activation rate α = 0. 01, the expected number of dual infections involving at least one latent virus was 0. 1 over 10 generations. However, when 10 latent forms were introduced to a population of 150,000 cells, i. e. , at the same proportion but in a ten-fold population, the total expected number of dual infections was 0. 94. Thus, when the simulated population size was increased ten-fold, the probability that at least one latent form ended up in a dually infected cell before disappearing from the replicating population increased nearly ten-fold. The long-term behavior in simulations where virus with latent genomic fragments proliferates is expected to be robust to stochastic effects due to the increased number of such viruses, however, the variance observed in the survival of latent virus forms between simulations is overestimated in a smaller population due to early stochastic extinction. The proportion of simulations with persistence of virus with latent genomic fragments is therefore likely an underestimate. We developed a within-host HIV-1 evolution model that includes point mutation, recombination, immune (positive) selection, negative selection, and latency. This model quantitatively captures previously observed viral sequence divergence and diversity trends and mechanistically explains these patterns, allowing us to model realistic within-host natural evolution of HIV-1. We used this model to investigate when and how latent forms of HIV-1 can survive in productively infected cells despite not being well adapted to the immune response, therefore having reduced relative fitness. Our simulation results suggest that recombination is a key mechanism facilitating the survival of virus forms with latent genomic fragments. We have previously shown that the majority of phylogenetic lineages in HIV-1 populations of untreated patients display a statistically defined signal of latency [1]. The results in this paper suggest that such lineages likely survived because of recombination. Our model further predicts that the survival of latent genomic fragments in plasma depends on the fitness landscape induced by the immune response. By comparing simulations where the surviving genomic fragments originated from different numbers of latent ancestors, we found that latency reduced mean sequence divergence, but increased mean sequence diversity for reasonable fitness landscapes. This effect may explain how HIV-1 can keep a high adaptation potential without mutating too far away from the infecting strain. High adaptation potential is useful for escaping immune and antiviral drug pressures, while less divergent sequences closer to the infecting form are arguably more fit to infect new hosts [2]. The dynamics of our simulated immune response depend on the number of epitopes and the antigen frequency triggering a new immune response, neither of which are well known. To test the sensitivity of our model, we varied these parameters to the extremes where measures of sequence divergence and diversity became unrealistic. Importantly, the trend of much higher survival of virus with latent genomic fragments in simulations with recombination than in simulations without recombination was robust for the full range of immune system parameters. This suggests that our main result is not sensitive to the exact details of how the immune response was implemented in our model. While in vivo HIV-1 viral populations are typically very large, computational considerations limited us to follow instead the relatively small population of 15,000 productively infected cells, which is in line with recent estimates of the effective population size in real HIV-1 populations [18]. The variability in the survival of virus with latent genomic fragments observed in our simulations is likely a result of stochastic fluctuations when latent virus first enters the replicating population. The proportion of simulations in which latent virus forms survive is therefore likely to be an underestimate. However, we expect the mean dynamics of virus with latent genomic fragments to be robust to scaling issues after the initial bottleneck. Previous studies have suggested that escape mutations have a fitness cost [10], and thus virus closer to the infecting strain may have higher replicative fitness due to having accumulated fewer escape mutations to evade immune surveillance. One of the limitations of our model is that we do not consider fitness costs associated with immune escape. However, compensatory mutations may arise in other parts of the genome, and at least partially restore fitness [11]. Furthermore, mutations can hitchhike and eventually affect population fitness. We do not attempt to incorporate epistatic effects here but rather investigate the theoretical mechanisms that may allow the persistence of latent genomic fragments in the plasma population despite virus activated from latent cells having a fitness disadvantage. Assuming that latent virus does not have a replicative advantage may lead us to underestimate the survival of latent genomic fragments in the plasma virus population. Furthermore, if activated virus from latent cells has similar fitness as contemporaneous variants, recombination may not be as important for survival. However, recombination does not merely facilitate survival, but rather allows latent genomic fragments to propagate through the plasma virus population. In the most extreme case where no selection pressure was exerted by the immune response (0 epitopes, S4 Table), approximately 60% of viral sequences contained latent genomic fragments during the last five years of infection in simulations with recombination, while only 6% of sequences were latent in simulations without recombination. While continual recombination with virus circulating in the plasma reduces the size of the genomic fragment inherited from a latent ancestor, even a few latent sites of high replicative fitness may allow the virus to better adapt to different evolutionary pressures. Latency may therefore expand the adaptive potential of HIV-1 besides enabling the virus to hide from immune surveillance and antiretroviral treatment. Because the patient sequences we chose to study were in envelope, we focused on antibody responses and escape from them. While we did not explicitly account for CTL responses, the epitopes in our model are approximately the same length as CTL epitopes, so our model could be interpreted as one in which CTL responses instead of or in addition to antibodies are providing immune pressure. However, CTL responses exert selection pressure on the whole proteome. Since we only simulated sequence evolution in env and ignored the interplay between different genes, our model predictions may not be generalizable to the whole virus. Our future directions include investigating further the balance between replicative fitness and immune escape on the level of the whole virus genome, and adapting our model to simulate HIV-1 evolution and latency under different treatment scenarios.

Increasing evidence suggests that HIV-1 released from activated latent cells survives in productively infected cells in patient plasma despite competition against better adapted virus variants that have evolved in response to the host immune pressure. Long-term survival requires that latent virus forms adapt to the host immune response so that they are not outcompeted. We simulated the dynamics of HIV-1 envelope sequence evolution in response to host immune pressure to investigate how virus from activated latent cells can survive despite having reduced fitness compared to the more evolved virus variants in patient plasma. The evolutionary trends of our simulated virus populations followed closely those observed in HIV-1 sequence data from 16 patients. Our simulation results suggest that recombination facilitates the survival of genomic fragments originating from virus activated from latent cells. Our model further predicts that sequence diversity increases with the number of latent genomic fragments from different origins that persist in plasma.

Abstract Introduction Models Results Discussion

2015

Recombination Enhances HIV-1 Envelope Diversity by Facilitating the Survival of Latent Genomic Fragments in the Plasma Virus Population

The burden of malaria in Vietnam has drastically reduced, prompting the National Malaria Control Program to officially engage in elimination efforts. Plasmodium vivax is becoming increasingly prevalent, remaining a major problem in the country' s central and southern provinces. A better understanding of P. vivax genetic diversity and structure of local parasite populations will provide baseline data for the evaluation and improvement of current efforts for control and elimination. The aim of this study was to examine the population genetics and structure of P. vivax isolates from four communities in Tra Leng commune, Nam Tra My district in Quang Nam, Central Vietnam. P. vivax mono infections collected from 234 individuals between April 2009 and December 2010 were successfully analyzed using a panel of 14 microsatellite markers. Isolates displayed moderate genetic diversity (He = 0. 68), with no significant differences between study communities. Polyclonal infections were frequent (71. 4%) with a mean multiplicity of infection of 1. 91 isolates/person. Low but significant genetic differentiation (FST value from -0. 05 to 0. 18) was observed between the community across the river and the other communities. Strong linkage disequilibrium (IAS = 0. 113, p < 0. 001) was detected across all communities, suggesting gene flow within and among them. Using multiple approaches, 101 haplotypes were grouped into two genetic clusters, while 60. 4% of haplotypes were admixed. In this area of Central Vietnam, where malaria transmission has decreased significantly over the past decade, there was moderate genetic diversity and high occurrence of polyclonal infections. Local human populations have frequent social and economic interactions that facilitate gene flow and inbreeding among parasite populations, while decreasing population structure. Findings provide important information on parasites populations circulating in the study area and are relevant to current malaria elimination efforts. Vietnam has been extremely successful in decreasing the country’s malaria burden, thanks to the large scale implementation of control interventions such as insecticide-treated bed nets, indoor residual spraying, and prompt, free-of-charge diagnosis and treatment; the number of cases fell from 130,000 in 2004 to 27,868 in 2014 [1]. Malaria has been virtually eliminated from Northern and Southern Vietnam [1,2]. In 2014,80% of malaria cases occurred in nine “hot provinces” where annual incidence peaked at 3. 1 cases per 1000, indicating a highly heterogeneous transmission, with hot spots of transmission (mostly in mountainous and forested areas) surrounded by areas of low transmission [1–5]. Vietnam aims at eliminating malaria by 2030 [6]. Such ambitious goal is threatened by P. vivax, whose characteristics (dormant liver forms that relapse weeks or months after clearance of the primary infection and gametocytes production before the occurrence of symptoms) together with the relative high occurrence of sub-patent and asymptomatic infections that remain undetected and thus untreated [3,7–10], make its transmission much more difficult to interrupt than that of P. falciparum. In addition, as already reported, elimination efforts are threatened by the emergence of drug resistance, for P. falciparum to artemisinin derivatives and partner drugs and for P. vivax to chloroquine (CQ) [2,11,12]. Current efforts to eliminate malaria are targeted to districts and communes reporting an increased number of malaria cases over time [2,6]. In this context, understanding parasite genetic diversity and its population structure is relevant for (i) monitoring temporal changes in transmission following control efforts, (ii) elucidating the spatial distribution of parasite populations and predicting outbreaks, population resilience, and the spread of drug-resistant parasites, and (iii) identifying ecological and behavioral risk factors that can inform malaria control and elimination efforts [13–15]. A previous study conducted in Binh Thuan province in central Vietnam reported high levels of genetic diversity (average expected heterozygosity (He) = 0. 86) and all infections being multi-clonal despite low transmission [13] (similar to what has been previously reported in South-East Asia) [16]. The aim of this study was to provide baseline data on the P. vivax parasite populations in four rural communities in the Vietnamese Quang Nam province. Samples were collected from April 2009 to December 2010 in four communities (Fig 1) in the South Tra My district of Quang Nam, Central Vietnam during a prospective cohort study aiming to assess the short- and long-term efficacy of CQ and high-dose piperaquine (PQ) for the treatment of P. vivax mono-infections [12]. Detailed sociodemographic characteristics of the local population have been already reported elsewhere [3]. In 2009, the prevalence of malaria by light microscopy was 7. 8%, while by polymerase chain reaction (PCR) prevalence was estimated at 22. 6% (ranging from 16. 4 to 42. 5%), with a high proportion of P. vivax mono infections (43%). Sub-patent infections accounted for 58. 7% of all infections, evidencing the existence of a substantial hidden human reservoir of malaria [3]. Malaria transmission is seasonal, and peaks during the rainy season (May to November). Based on data from the Provincial Malaria Station, between 2009 and 2013 the mean prevalence of P. falciparum, P. vivax, and mixed malaria cases for all age groups in the study area was 64. 1%, 31. 5%, and 4. 4%, respectively [17–18]. The main malaria vectors in the area are Anopheles dirus sensu stricto and An. minimus, though An. vagus, An. aconitus, and An. philippinensis are also present [17–18]. A finger prick blood sample was collected at day 0 (before treatment) for diagnosis by light microscopy and two blood spots were collected on grade 3 filter paper (Whatman Ltd. , Springfield Mill, Maidstone, United Kingdom) for molecular diagnosis and microsatellite (MS) genotyping. The study was approved by the National Institute of Malariology, Parasitology and Entomology in Hanoi, the Ministry of Health of Vietnam, and the review boards of the Institute of Tropical Medicine and Antwerp University Hospital (UZA) in Antwerp, Belgium. Adult participants (in case of minors one of the parents/guardians) provided written informed consent. Thick and thin film blood slides were stained with a 3% Giemsa solution for 45 minutes, and the number of asexual parasites was calculated following World Health Organization (WHO) guidelines [19]. Parasite density was estimated by dividing the number of asexual parasites for 200 white blood cells (WBCs) counted and expressed as the number of asexual parasites per microliter of blood, assuming 8000 WBC/μL. All blood slides were double-read by two technicians and in case of disagreement, slides were read by a third senior technician. The final results were expressed as the mean of the two closest results. A slide was declared negative if no parasites were found after counting 1,000 WBCs. DNA was extracted from filter paper blood spots cut into 5-mm-diameter disks with the QIAamp DNA Micro Kit following the manufacturer’s recommendations (Qiagen, Hilden Germany). P. vivax mono infections were confirmed by species-specific multiplexed semi-nested PCR, as described by Rubio et al. [20]. All samples were genotyped with 14 MS (MS1-MS10, MS12, MS15, MS20, and PvSal1814) following previously described PCR protocols [13–14]. The PCR products of four MS were pooled and analyzed by capillary electrophoresis in a 3730 XL ABI sequencer (Applied Biosystems) and 1200 Liz was used as the internal size standard. Negative samples were repeated once. Allele calling was performed using GeneMarker version 2. 4. 0. After pooling capillary electrophoresis fsa files from all samples, a standard cut-off value of 500 relative fluorescence units was defined and peaks below this limit were considered background noise. In addition, all samples were double-checked manually to confirm true alleles. At each locus the predominant allele (the one giving the highest peak) and minor alleles within at least two-thirds of the height of the predominant allele were scored [13,15,21]. Only predominant alleles were used to define haplotypes to ensure an unbiased estimate of minor allele frequency in polyclonal infections [15,22–24]. Samples were defined as polyclonal if at least one locus presented more than one allele [13–14]. Polyclonal/locus (%) describes the percentage of samples identified as polyclonal by a given MS out of the total number of samples [13]. Multiplicity of infection (MOI), defined as the minimum number of different clones observed in a sample, was estimated by taking the maximum number of alleles at the two most polymorphic markers [13,25]. Average MOI was defined as the sum of MOIs detected across all samples divided by the total number of samples. Average MOI and the proportion of monoclonal and polyclonal infections were compared with the Kruskal-Wallis and Pearson χ2 test, respectively. A value of p < 0. 05 was considered significant. The predominant alleles in each sample were used to calculate the number of haplotypes by GenAlEx 6. 5 [26]. Haplotypes that appear only once in the population were defined as unique haplotypes. Genetic diversity, defined as the probability of observing different genotypes at a given locus in two unrelated parasites, was assessed by calculating the expected heterozygosity (He) for each community using the formula: He = [n/ (n-1) * (1-∑pi2], where n is the total number of alleles and p is the allele frequency. He ranges between 0 and 1, with values close to 1 indicating high genetic diversity [27]. Allelic richness, defined as the number of alleles per locus independently of sample size, was calculated using FSTAT v2. 9. 3 [28]. He and allelic richness were compared between communities using the Kruskal-Wallis test. The presence of bias due to false assignment of predominant haplotypes was investigated by comparing He in the database containing all the alleles and the database containing the predominant alleles [29]. A standardized index of association (IAs), calculated with LIAN v3. 5, was used to assess the presence of multilocus linkage disequilibrium (LD) in the parasite population [30]. The significance of the IAs estimate was assessed with Monte Carlo simulation using 10,000 random permutations of the data. To differentiate between clonal propagation and epidemic expansion, we compared LD in the predominant allele dataset and the unique haplotype dataset [31]. Pairwise LD was used to evaluate the physical linkage between loci located within the same contig using the G statistic in FSTAT v2. 3. 9 [28,32]. Genetic differentiation between pairs of communities was estimated using a pairwise unbiased estimator of F-statistics FSTAT v2. 3. 9 with no assumption of Hardy-Weinberg equilibrium within samples [28,32]. A matrix of p values corresponding to each pairwise FST was calculated after Bonferroni correction, with a value of p < 0. 05 considered significant. Crude FST values were adjusted for sample size using Recode Data v. 0. 1 [33] and standardized FST estimates were obtained by dividing crude FST values by adjusted FST values. FST estimates ranged from 0 (no genetic differentiation between communities) to 1 (full differentiation). As a complementary approach, population structure was investigated using the software programs STRUCTURE v2. 3. 2 [34], CLUMPP [35], DISTRUCT [36], and GENODIVE [37]. STRUCTURE was used to identify clusters of genetically related samples. The number of clusters (K) was set from 1 to 10 with 10 replications per K, and 150,000 Markov Chain Monte Carlo steps after a burn-in period of 50,000 iterations using the admixture model. The loc-prior model was used for accurate inference of population and individual ancestry. Next, we used STRUCTURE HARVESTER v0. 6. 94 [38] to calculate the most likely number of K clusters. Additional data parsing and formatting of the STRUCTURE output was performed using CLUMPP and DISTRUCT [38–39]. CLUMPP permutes the clusters’ output by performing multiple replicate runs for the selected K. Samples with an average pairwise similarity (H value) of over 85% in one of the K clusters were considered to belong to that particular population; all other samples were considered admixed samples. DISTRUCT performs geographical displays of the aligned cluster assignment. We confirmed the optimal K, i. e. the K with the highest pseudo-F statistic, using AMOVA-based K-means clustering analysis in GENODIVE V2. 0b23 (OS X 10. 6 operating system). This method divides a number of individuals into an a priori assigned number of clusters (K) in such a way that minimizes within-group diversity and maximizes between-group diversity. The pseudo-F statistic was calculated by setting up simulated annealing runs with 150,000 steps and 50 algorithm repetitions to determine optimal clustering (highest pseudo- F-statistic). eBURST v3 was used to identify clusters of closely related haplotypes, or haplogroups (HGs), which were defined as haplotypes sharing at least 9 loci from the 14 MS analyzed [40]. Haplotypes unrelated to any haplogroup (HG) were classified as singletons. The relationship between haplotypes following the defined K clusters was further analyzed by PHYLOViz [41]. Finally, to investigate the relationship between geographic and genetic distances in the study population, we performed principal coordinate analysis with the Mantel test for matrix correspondence in GenAlEx 6. 5 [26]. Allele frequency was calculated in GenAlEx 6. 5 using the predominant allele data set with 14 loci. The existence of a recent population bottleneck was investigated by evaluating the allele frequency distribution in the population (alleles at low frequencies are less abundant in populations with a recent bottleneck) [42–43]. In total, 234 individuals with P. vivax mono infection from the 260 recruited in the original study [12] were successfully genotyped and included in the analysis. Baseline characteristics of study participants are described in Table 1. Successful genotyping, with at least 12 of the 14 MS, was achieved in 194 patients (82. 9%). Allele data were successfully recovered in more than 83% of the samples for all MS except MS20 and Pvsal1814, for which successful amplification was achieved in 66% and 75% of samples respectively. The MS characteristics are described in Table 2. Overall, genetic diversity was moderate, with an average He = 0. 68 (95%CI 0. 58–0. 77) for all MS. MS3 and MS9 were the least polymorphic markers (He = 0. 45 and 0. 31 respectively), while MS10 and Pvsal1814 were the most polymorphic markers (He = 0. 99 and 0. 92 respectively), which were therefore used to calculate MOI. The number of alleles per MS ranged from 3 to 14. All MS had non-significant differences in He values in the database containing all alleles per locus and the predominant allele datasets ruling out bias in the construction of haplotypes from polyclonal infections (p = 0. 68). The average number of alleles per locus was 5. 5 (95%CI 3. 82–7. 17), the average number of alleles detected in a sample by any locus was 1. 14 (95%CI 1. 0–1. 27) and the average allelic richness was 5. 05 (95%CI 4. 11–5. 98). The proportion of polyclonal infections, similar in all communities (p >0. 05), was 71. 4% (167/234) when all 14 MS were used, but 64. 1% (141/220) (N = 220 as 14 samples had missing data for MS10 and Pvsal1814) when only Pvsal1814 and MS10, the most polymorphic markers were used. The same two MS were used to calculate MOI (based on 220 samples with completed data), whose mean in the four communities was 1. 91 (95% CI 1. 81–2. 02), with no significant differences between communities (p = 0. 52), age groups (MOI≤15years = 2. 0 vs MOI>15years = 1. 9, p = 0. 50), gametocyte carriage (MOIgametocytes present = 1. 93 vs MOIgametocytes absent = 1. 80, p = 0. 42), sex (MOImale = 1. 89 vs MOIfemale = 1. 95, p = 0. 55), symptomatic vs asymptomatic (defined as fever at enrolment vs no fever at enrolment, MOI = 1. 98 vs MOI = 1. 80, respectively) (p = 0. 10), season (MOIrainy season = 1. 90 vs MOIdry seaoson = 1. 97, p = 0. 64), and ethnic minority (MOICadong = 1. 89 vs MOIM’nong = 2. 0, p = 0. 46). No significant differences were found between the four communities for either level of He (p = 0. 08) or allelic richness (p = 0. 31). We identified 101 haplotypes from 144 samples of which 84 haplotypes were defined as unique haplotype with complete genotyping data for 13 MS. MS20 was excluded because it had the lowest successful genotyping rate (66%). Of these haplotypes, 25. 7% (26/101) were found in monoclonal infections and 16. 8% (17/101) were found in both monoclonal and polyclonal infections; 6. 93% of haplotypes (7/101) had a frequency of over 2 in 40 samples and the two most frequent haplotypes were detected in 7. 6% (11/144) and 6. 2% (9/144) of samples. One haplotype was shared between the four communities, 4 haplotype found in community 1 were also present in community 2 and one haplotype shared between community 3 and community 4. To evaluate the existence of a recent population bottleneck we analyzed the allele frequency distribution in the population. Fig 2 shows an L-shaped distribution of allele frequencies, as would be expected from neutral evolution. MS20 was also excluded from the LD analysis to maximize sample size and avoid bias due to an imbalanced number of samples between communities. Hence at least 25% of samples per community were included in the analysis. Significant LD was observed in each community (IAs ranged from 0. 10 to 0. 17) and in the overall study population (IAs = 0. 113, p < 0. 001). LD remained significant (IAs = 0. 059, p < 0. 001) when only the unique haplotypes were used. We then examined patterns of LD between pairs of MS (Fig 3). Pairwise LD was observed between loci located within the same contigs (MS4-MS5, MS7-MS8, and MS12-MS15) and also within different contigs. Even though lower pairwise LD was observed in communities 3 and 4, the fact that the overall LD was significant (p = 0. 008) suggests the existence of a clonal parasite population. We first compared the datasets containing only monoclonal infections (n = 67) with the predominant allele (n = 234) and found low genetic differentiation (FST = 0. 05), indicating absence of bias. Then, we calculated FST values for pairwise genetic differentiation between the four communities (Table 3). We observed moderate genetic differentiation between community 4 and the other communities (FST = 0. 15–0. 18) and low differentiation for the other combinations (FST < 0. 1) (n = 144), indicating that the parasite population in community 4 is moderately, although significantly, differentiated from parasite populations in communities 1,2, and 3 (p = 0. 008). Similar FST values were obtained when only unique haplotypes (n = 84) were used. Structure analysis identified the most likely clusters in the population to be (i) K = 7 (ΔK = 9. 4), (ii) K = 2 (ΔK = 5. 2), and (iii) K = 3 (ΔK = 3. 2) (n = 144). The AMOVA-based K-means clustering analysis identified K = 2 as the optimal number of clusters (pseudo-F = 30. 5). We further analyzed the parasite population divided by K = 2 (cluster 1 and cluster 2) with CLUMPP and DISTRUCT (Fig 4), and found that 33. 3% (48/144) of the samples (with complete haplotypes) observed in the study population belonged to cluster 1 and 20. 1% (29/144) belonged to cluster 2 and 46. 6% (67/144) were admixed samples. Community 4 had the highest proportion of admixed samples (62. 1%), followed by community 3 (53. 9%), while community 1 and 2 had similar rates (40. 9% and 41. 3% respectively). The proportion of admixed samples remained high when the number of clusters was set to K = 3 and K = 7 (48. 6% and 38. 9%, respectively). Of note, cluster 1 samples were absent from community 4, which supports a moderate degree of population structure between communities 1–3 and community 4. However, principal coordinate analysis failed to detect geographical clustering (per community) in the population. Then, we investigated genetic relatedness, defined as haplotypes sharing at least 10/13 loci (n = 101) by eBURST. Eight different HGs and 13 singletons were identified. HG1 contained 51. 5% (52/101) of all haplotypes in the four communities, while the other 7 HGs contained between 2. 0% (2/101) to 11. 9% (12/101) haplotypes. However, when relatedness was defined as haplotypes sharing at least 9/13 loci, only 2 HGs and 2 singletons were identified. HG1 included 95. 0% (96/101) of all haplotypes detected. PHYLOVIZ analysis supported the existence of related haplotypes among all study communities (with slightly clustering of community 4 samples) (Fig 5A) and confirmed the absence of cluster 1 haplotypes in community 4 (Fig 5B). We analyzed the genetic diversity and population structure of 234 P. vivax pre-treatment clinical isolates collected in a forested area of Central Vietnam between April 2009 and December 2010 [12]. We observed moderate levels of heterozygosity in all four study communities, with a high proportion of polyclonal infections and significant LD, suggestive of inbreeding across parasite populations circulating in the study communities. Genetic differentiation and population structure between study communities was low but present between villages at each side of the river defining a moderate geographical barrier to gene flow. In this study we used eight MS (MS1, MS4, MS6, MS9, MS10, MS12, MS15, and MS20) with balanced diversity, three (MS2, MS5 and Pvsal1814) with unbalanced diversity, one (MS8) with significant excess diversity, and two (MS3 and MS7) with significant reduced diversity [25]. Mean He in the study population (He = 0. 68) using those 14 MS was non-significantly different to He when only MS with balanced diversity (recommended for measuring population diversity) were used. Therefore, all MS were kept in the analysis to assess both diversity parameters and polyclonal infections, but only the two most polymorphic markers (MS10 and Pvsal1814) were kept to investigate MOI. The mean He in our study population (He = 0. 68) was similar to figures seen in areas of north-west Brazil with similar transmission intensities (He = 0. 74 and He = 0. 68) [43–44]; higher than those observed in South Korea (He = 0. 43) [45] and the Loreto district, Peru (He = 0. 37) [16], and lower than those seen in Sri Lanka (He = 0. 89) [46], Pursat, Cambodia (He = 0. 84) [16], and Binh Thuan, Central Vietnam (He = 0. 88) [13]. In our study MS9, MS3, and MS7 displayed the lowest number of alleles per locus (n = 3) and MS9 and MS3 had the lowest He values (HeMS9 = 0. 31 and HeMS3 = 0. 45). MS3, MS7, and MS9 would therefore appear to be poorly informative markers in the study area and their use in future studies is not recommended. Polyclonal infections were frequent (71. 4%) when the results from 14 MS were combined and moderately lower (64. 1%) when just Pvsal1814 and MS10 were used [47]. Mean MOI was 1. 91, with similar MOI observed in symptomatic and asymptomatic study participants, possibly because of the high parasite density (mean 3,919/μL; 95%CI 2,852–4,986) detected in asymptomatic participants at day 0. In the literature, the proportion of polyclonal infections vary considerably depending on the MS markers used [16,48], highlighting the need for a standardized methodology that allows comparison between studies and geographical regions. High proportions of polyclonal infections have also been reported in hypo-endemic areas in Sri Lanka (60%) [49], Colombia (60–80%) [48], the Amazon Basin in Brazil (50%) [43], and more recently, in a pre-elimination context in Sri Lanka (69%) [46]. It is noteworthy in a study carried out (1999–2000) in Binh Thuan province, central-south Vietnam, where the entomological inoculation rate was estimated at 1 infective bite/person/year, 100% of vivax infections were polyclonal with a mean MOI of 3. 7 [13]. The high levels of genetic diversity and polyclonal infections in low transmission areas [13,23,48] can be, at least partially, explained by the unique biology of P. vivax which result in (i) a high prevalence of asymptomatic and low parasite density infections (which last longer because are difficult to detect, increasing the likelihood of repeated infections with divergent clones, resulting in increased polyclonality) and (ii) relapse from dormant liver stages (the reactivation of heterologous clones increases the likelihood of peripheral superinfections). Since a high proportion of study participants were asymptomatic at recruitment (59. 0%) and poor adherence to PQ radical cure is known in the study area [3], the high proportion of polyclonal infections found in this study may reflect peripheral superinfection fed by heterologous clones from both relapses and reinfections. Despite those high rates of polyclonal infections, we observed a significant LD (IAs = 0. 113, p < 0. 001) in the overall study population. Asexual clones present in one infection produce gametocytes that, taken by the vector, recombine during meiosis and generate new haplotypes in a process known as outcrossing. Consequently, the breakdown of pre-existing associations between unlinked loci would reduce LD to low levels [50] as opposed to recombination between gametes from the same parasite [51]. As transmission decreases, fewer parasite types will be present in the population and recombination will often occur between related parasites, increasing the level of inbreeding in the population. This is supported by the fact that 53. 8% of all polyclonal infections were identified by multiple alleles at just one locus. Indeed, LD remained significant in the analysis using only unique haplotypes, indicating that it is a result of inbreeding rather than expansion of few haplotypes due to outbreaks or epidemics [31]. Closely related parasites in hypoendemic areas have been previously reported [52,53]. In addition, inbreeding was further supported by overall significant pairwise LD [21,31]. LD combined with high levels of polyclonality has been reported in rural Amazonia [54] and more recently in Sri Lanka [46]. The authors of these studies offered two alternative interpretations for this phenomenon. First, the MS may not be strictly neutral (10/14 MS mapping to loci encoding either hypothetical or annotated proteins may be subject to natural selection) [22–23]. And second, replication-slippage events during mitotic (asexual) replication could result in the generation of new alleles due to the addition or deletion of repeats [49,55]. If the replication-slippage rate is higher than that of effective recombination (the probability of producing a recombinant genome), the clones generated would increase polyclonality, without altering LD. It has been previously reported that replication-slippage events (and therefore number of alleles per locus and He) correlate positively with increasing repeat length and non-perfect repeats motifs, i. e. interrupted or compound motifs [25,56]. Pvsal1814 MS used in this study, which had an (AGA) 44 motif structure with an interrupted/compound motif, He = 0. 91 and 14 different alleles with frequencies ranging from 1. 3% to 16%, identified 53. 8% of all polyclonal samples in the study population. Indeed, inherent mutability in this MS has been described to produce excess diversity, which in turn is recommended to identify MOI [25]. We identified 101 haplotypes, of which 84 appeared only once in the population. Ninety percent of them were grouped in a single haplogroup (HG1), defined by identical alleles in at least 9/13 loci, indicating a high degree of relatedness among parasites across the communities. These results support the view that despite a high level of polyclonality, inbreeding among highly related haplotypes maintains LD. The adjusted genetic differentiation was low between communities 1,2, and 3 (FST < 0. 05) and moderate when community 4 was included (FST = 0. 15–0. 18), indicating limited geographical boundaries between neighboring communities 1–3 but higher differentiation with the community across the river. In concordance with the FST values, the STRUCTURE analysis detected two main parasite populations. Two clusters of haplotypes, with a high proportion of mixture haplotypes (60. 4%) were observed in all four communities. The fact that a majority of haplotypes found in community 4 belonged to cluster 2, which was the minor cluster in the other 3 communities, supports a certain degree of differentiation between communities 1–3 and 4. Moderate population differentiation between these communities can be explained by geographical proximity and socioeconomic relationships between the communities’ inhabitants as previously described [3]. Inhabitants of community 1–3 (located at one side of the river) belong to the Cadong ethnic group and therefore share some degree of kinship, facilitating social exchange. Conversely, community 4, whose inhabitants belong to the M’nong ethnicity, is located at the other side of the river with limited access during the rainy season. Malaria incidence in the Quang Nam province has dropped by 78. 0% over the last decade thanks to the implementation of efficient control strategies [1,17,57]. At the time of the study, malaria prevalence in the study area was 7. 8% as assessed by light microscopy and 23. 6% as estimated by PCR [3]. Therefore, the moderate-to-high levels of genetic diversity detected, together with the high polyclonality and low population structure are consistent with an epidemiological context of transition from moderate to low endemicity [58–59]. Future studies aiming at identifying changes in genetic diversity and population structure to support the development or improvement of control and elimination interventions should include isolates collected at several time points from all areas where malaria is prevalent (or has been recently eliminated). Ideally, a molecular surveillance system should be implemented within the existing network of sentinel sites for drug resistance across the country to support evaluation of interventions and improve response strategies at the provincial level. Parasite populations with strong LD and the presence of gene flow could fuel the spread of resistant parasites in the event of the emergence of drug resistance, threatening current treatment efforts and achievements towards malaria elimination in Central Vietnam. Temporal analysis to investigate haplotype persistence and the risk of clonal expansion is urgently needed in order to inform decision makers.

In Vietnam, Plasmodium vivax (P. vivax) is the second most frequent human malaria parasite and a major obstacle to countrywide malaria elimination. Knowing the local parasite structure is useful for elimination efforts. Therefore, we analyzed, with a panel of 14 microsatellite markers, 234 P. vivax mono infections in blood samples collected from 4 communities in central Vietnam. Genetic diversity in the population was moderate; a high occurrence of polyclonal infections and significant linkage disequilibrium were detected, suggesting inbreeding or recombination between highly related haplotypes. In addition, both genetic differentiation and population structure was low and only detected between communities at each side of the river. Those results suggest gene flow between study communities with the river defining a moderate geographical barrier. Future studies should determine how this genetic variation is maintained in an area of extremely low transmission.

Abstract Introduction Materials and Methods Results Discussion

medicine and health sciences parasite groups ecology and environmental sciences rivers plasmodium population genetics tropical diseases geographical locations parasitic diseases parasitic protozoans parasitology apicomplexa protozoans aquatic environments bodies of water population biology malarial parasites marine and aquatic sciences vietnam genetic loci people and places haplotypes freshwater environments asia earth sciences genetics biology and life sciences malaria evolutionary biology organisms

2016

Population Genetics of Plasmodium vivax in Four Rural Communities in Central Vietnam

Although ICP4 is the only essential transcription activator of herpes simplex virus 1 (HSV-1), its mechanisms of action are still only partially understood. We and others propose a model in which HSV-1 genomes are chromatinized as a cellular defense to inhibit HSV-1 transcription. To counteract silencing, HSV-1 would have evolved proteins that prevent or destabilize chromatinization to activate transcription. These proteins should act as HSV-1 transcription activators. We have shown that HSV-1 genomes are organized in highly dynamic nucleosomes and that histone dynamics increase in cells infected with wild type HSV-1. We now show that whereas HSV-1 mutants encoding no functional ICP0 or VP16 partially enhanced histone dynamics, mutants encoding no functional ICP4 did so only minimally. Transient expression of ICP4 was sufficient to enhance histone dynamics in the absence of other HSV-1 proteins or HSV-1 DNA. The dynamics of H3. 1 were increased in cells expressing ICP4 to a greater extent than those of H3. 3. The dynamics of H2B were increased in cells expressing ICP4, whereas those of canonical H2A were not. ICP4 preferentially targets silencing H3. 1 and may also target the silencing H2A variants. In infected cells, histone dynamics were increased in the viral replication compartments, where ICP4 localizes. These results suggest a mechanism whereby ICP4 activates transcription by disrupting, or preventing the formation of, stable silencing nucleosomes on HSV-1 genomes. The genes of the nuclear-replicating double stranded (ds) DNA virus herpes simplex virus 1 (HSV-1) are expressed in a coordinate manner. VP16, a virion protein, first activates expression of the five immediate early (IE) genes, in part through the recruitment of the histone demethylase LSD1 and histone acetyltransferase CBP/p300 to IE promoters [1–5]. Two IE proteins, ICPO and ICP4, then activate transcription of the early (E) genes, which encode proteins required for HSV-1 DNA replication and several other functions [6]. Late (L) genes are transcribed after DNA replication. Both ICP0 and ICP4 also contribute to the activation of L gene expression. The mechanisms whereby VP16 activates IE gene transcription are well characterized [1,3, 5,7–12]. In contrast, the mechanisms whereby ICP0 and ICP4 then activate transcription of E and L genes remain only partially understood. ICP4 binds to specific DNA sequences to inhibit transcription of IE genes [13]. However, it does not bind to any specific sequences to activate transcription of E or L genes [14]. Over 141 proteins that interact with ICP4 at 6 h post infection (hpi) were identified by mass spectrometry analyses, including the chromatin remodeling complexes SWI/SNF, Ino80, and NuRD [15]. The histone acetyltransferase CLOCK was identified as another ICP4 interactor by coimmunoprecipitation [16]. ICP4 also interacts with many components of the mediator complex and may activate transcription by a gene looping mechanism [15], promoting the recycling of RNA polymerase II from the 3’ end of a gene back to the transcription start sites. Whereas HSV-1 genomes are regularly chromatinized in latent infection, HSV-1 genomes are in particularly dynamic chromatin in lytic infections [17]. The basic unit of chromatin is the nucleosome, which consists of two dimers of each of the core histones H2A-H2B and H3-H4 wrapped by 146 base pairs of double stranded DNA. Linker histone H1 further binds DNA at the entry and exit sites of the core nucleosome. Chromatin is dynamic, nucleosomes disassemble and then the released histones diffuse through the nucleus bound to chaperones and re-assemble nucleosomes at different sites. Linker histones are more dynamic than core histones, with their exchanges occurring in minutes or hours, respectively [18–20]. The dynamics of cellular nucleosomes are altered through post-translational modifications to histones and the incorporation of histone variants instead of the canonical ones, among other factors [21–30]. Acetylation of histone tails by histone acetyltransferases generally destabilizes nucleosomes, whereas their methylation by histone methyltransferases destabilizes or stabilizes nucleosomes, depending on the site and the degree of methylation [21–29]. Nucleosomes containing H3. 3 are more dynamic than those containing H3. 1 [30]. Canonical histone H3. 1 is assembled in chromatin with newly synthesized DNA by the histone chaperone CAF-1, whereas variant H3. 3, which differs by only 5 amino acid residues, is assembled in the chromatin of transcribed genes or telomeres by HIRA or DAXX, respectively, independently of DNA replication [31–36]. H3. 3 is typically post-translationally modified with more markers of active transcription than H3. 1, such as K4 and K79 methylation and K9, K14 and K23 acetylation [37]. We had found that histone dynamics increase during infection with wild type HSV-1 [38–40]. Histone dynamics still increased in infected cells treated with phosphonoacetic acid, indicating that neither HSV-1 DNA replication nor L gene expression are required, whereas they were largely unaffected by UV-inactivated HSV-1, indicating that virion attachment or entry are not sufficient. Therefore, IE or E proteins most likely affect histone dynamics. We and others propose a model in which the chromatinization of HSV-1 DNA is a cellular defense mechanism to silence HSV-1 gene expression. To counteract this mechanism, HSV-1 would have evolved proteins that prevent or disrupt the stable chromatinization of HSV-1 genomes. This nucleosome destabilization process would increase histone dynamics and promote transcription. Under this model, one or more of the three HSV-1 transcription activators would be expected to enhance histone dynamics. Here we report that HSV-1 mutants encoding no functional VP16, ICP0 or ICP4 still enhance histone dynamics, but to a much lesser extent than wild type HSV-1. We further show that an HSV-1 mutant encoding no functional ICP4 is the most deficient in enhancing histone dynamics. Transient expression of ICP4 was sufficient to enhance histone dynamics in the absence of any other HSV-1 protein or DNA. ICP4 may moreover preferentially target silencing histone variants, such as H3. 1. The dynamics of canonical H2A were not enhanced in cells expressing ICP4, suggesting that other H2A variants may be targeted by ICP4. During lytic infections, histones were more dynamic in the replication compartments, where ICP4 localizes, than in the cellular chromatin. Together, our results suggest a novel mechanism of transcription activation by ICP4, in which ICP4 prevents the formation of stable nucleosomes on HSV-1 genomes, or destabilizes preformed ones, to promote transcription by allowing access of the RNA polymerase II complex to the HSV-1 genes. IE or E proteins enhance linker and core histone dynamics during HSV-1 infection [38–40]. To test whether the enhanced dynamics required the expression of ICP4 or any E protein, we used HSV-1 strain n12, which expresses a transactivation incompetent truncated ICP4 [41]. Consequently, IE proteins other than ICP4 are expressed to high levels in the absence of any E or L protein expression or DNA replication. The levels of green-fluorescent protein (GFP) -histone fusion proteins in the free pools, and the initial rates of fluorescence recovery after photobleaching (core histones), or time to recover 50% of the relative fluorescence in the photobleached region (T50; for linker histone H1. 2), were evaluated to analyze histone dynamics [38–40]. The fluorescence recovery of histones is biphasic [18,20,42]. The initial, faster, phase of fluorescence recovery, analyzed by the slope of the fluorescence recovery between the first two times, reflects histones assembled in the most dynamic chromatin, such as those in rapidly transcribed genes. The later, slower, phase of fluorescence recovery, analyzed by the slope of the fluorescence recovery between 25 and 100 seconds for core histones, reflects the histones assembled in less dynamic chromatin. The relative fluorescence intensity immediately after photobleaching reflects the “free pool” of histones, as only histones not in chromatin diffuse in and out the bleached region during photobleaching. The global dynamics of linker histones are described by the T50, which is the most sensitive parameter. The levels of all free histones had unimodal normal frequency distributions throughout the population of n12 infected U2OS cells (Fig 1A). n12 infection of U2OS cells was not sufficient to increase the free pools of any core histone, whereas those of H1. 2 were only increased to a basal degree at early times after infection (Fig 1A; P<0. 05). The levels of all free histones also had unimodal normal frequency distributions throughout the population of n12 infected Vero cells (Fig 1B). The free pools of H3. 1, H3. 3, and H4 were also increased at 4 and 7 hpi in n12 infected Vero cells, although less than in KOS infected cells [39,40]. We had previously shown that the enhancement of histone dynamics in Vero cells infected with an HSV-1 mutant in ICP0 was partly impaired, such that the enhanced late increase of histone dynamics ultimately occurs (Fig 2A and 2B, n212 [38–40]). The pools of some free histones were increased to an even larger degree at 7 hpi in the absence of ICP0 (Fig 2A, n212 [38–40]). The double ICP0 and VP16 HSV-1 mutant, which expresses little ICP4 in Vero cells [43], enhanced histone dynamics to almost only a basal level in these cells (Fig 2A and 2B, KM110 [38–40]). Vero cells infected with ICP4 mutant n12 only had a basal increase in the levels of free linker and core histones, which was not further enhanced at later times after infection (Figs 1B and 2). Expression of ICP0, ICP22, ICP27, or ICP47 in the absence of functional ICP4 (and E proteins) is thus not sufficient to increase the pools of free core or linker histones to the same degree as infection with wild-type or ICP0 or VP16 mutant strains of HSV-1 (Fig 2 [38–40]). To test whether the inability of n12 to enhance histone dynamics above the basal degree was due to unknown mutations within this strain, histone dynamics were re-evaluated in a complementary Vero-derived cell line (n-33) that expresses HSV-2 ICP4 upon infection [44]. The dynamics of core and linker histones were enhanced to approximately the same degree in n-33 cells infected with n12 or wild-type KOS (Fig 3). Histones are thus minimally mobilized in U2OS or Vero cells infected with an HSV-1 mutant encoding no functional ICP4 (Figs 1–3). ICP4 may induce the increase in histone dynamics by itself. Alternatively, the protein product of an E gene may increase histone dynamics (DNA replication or L proteins are not required [39,40]), as the expression of E genes requires ICP4. To test these possibilities, we analyzed the effects of ectopically expressed ICP4 in histone dynamics. H4 and H2B have no major variants, and they therefore represent the entire population of H3-H4 and H2A-H2B dimers, respectively. To evaluate the dynamics of H4 and H2B in cells transiently expressing ICP4, we optimized the co-transfection of GFP-H2B or GFP-H4 with free red fluorescent protein (RFP) or RFP-ICP4 such that approximately half of the cells expressing detectable levels of GFP also expressed detectable levels of the RFP fusion proteins (or free RFP). This approach allows us to analyze histone dynamics in cells expressing detectable or undetectable levels of the test proteins in otherwise identical conditions. GFP fluorescence within the bleached region was normalized to total nuclear fluorescence to account for differences in GFP expression. The relative fluorescence within the bleached region at each time was then normalized to the initial relative fluorescence within the same region prior to photobleaching. The results are therefore independent of the GFP-histone expression levels [38–40]. The free pools of GFP-H4 or -H2B were 22 or 12% greater, respectively, in cells expressing detectable than undetectable levels of RFP-ICP4 (p<0. 01) (Fig 4A, 4B, 4D and 4E). As expected, the free pools of GFP-H4 or -H2B were not significantly higher in cells expressing detectable than undetectable levels of free RFP (Fig 4B and 4E). The slow exchange rate of GFP-H2B, which evaluates the dynamics of the H2B molecules in low turnover nucleosomes, was 57% greater in cells expressing detectable than undetectable levels of RFP-ICP4 (p<0. 05) (Fig 4C). While the slow exchange rate of GFP-H4 tended to be faster in cells expressing detectable than undetectable levels of RFP-ICP4, it was not significantly so (Fig 4F). The slow exchange rates of GFP-H2B or -H4 were not significantly changed in cells expressing detectable levels of free RFP (Fig 4C and 4F). H3. 3 is initially detected in the nucleosomes assembled with HSV-1 genomes, whereas H3. 1 is detected in HSV-1 nucleosomes only after the onset of HSV-1 DNA replication [45]. Consistently, the dynamics of H3. 1 and H3. 3 are differentially affected in cells infected with wild type HSV-1 [40]. Their free pools decrease between 4 and 7 hpi in Vero cells, but that of H3. 1 decreases to a much greater extent (Fig 2). The free pools of H3. 1 also decrease between 4 and 7 hpi in U2OS cells, whereas those of H3. 3 do not (Fig 2). Whereas the free pool of H3. 3 at 7 hpi is not affected by HSV-1 DNA replication, moreover, that of H3. 1 is two-fold greater when HSV-1 DNA replication is inhibited [40]. The dynamics of H4 were enhanced in cells transiently expressing ICP4 (Fig 4D and 4E). We therefore expected the dynamics of H3. 1 or H3. 3, which form dimers with H4, to also be enhanced. The free pool of GFP-H3. 3 was 14% greater in Vero cells expressing detectable than undetectable levels of RFP-ICP4 (p<0. 05) (Fig 5A and 5B). The unimodal frequency distribution of the GFP-H3. 3 free pools had its peak shifted to the right, indicating a larger free pool, in cells expressing detectable ICP4 compared to cells expressing undetectable ICP4 (Fig 5E). In contrast, the frequency distribution of the GFP-H3. 3 free pools was not altered in cells expressing detectable or undetectable RFP (Fig 5B and 5F). GFP-H3. 1 was mobilized to a much greater extent (Fig 5C, 5D and 5G). The average free pool of GFP-H3. 1 was 248% greater in cells expressing detectable than undetectable levels of RFP-ICP4 (Fig 5D). The frequency distribution curves of the GFP-H3. 1 free pools showed moreover that the cells expressing undetectable levels of ICP4 had free pools distributed normally around 20%, whereas the cells expressing detectable ICP4 had a skewed distribution peaking at twice as large (Fig 5G). Cells expressing detectable RFP or not had equally distributed free pools (Fig 5H). The increased dynamics of GFP-H3. 1 were also reflected by its nuclear distribution (Fig 5I). GFP-H3. 1 had the punctuated localization characteristic of chromatin in cells expressing undetectable levels of RFP-ICP4. In contrast, GFP-H3. 1 was diffusely distributed through the nucleus in cells expressing detectable levels of RFP-ICP4, distribution which is consistent with soluble proteins (i. e. , free H3. 1). The free pools of GFP-H3. 1 or -H3. 3 were not affected in cells expressing detectable levels of free RFP (Fig 5). The free pools of GFP-H3. 3 or -H3. 1 were also 22% or 40% greater, respectively, in U2OS cells expressing detectable than undetectable levels of RFP-ICP4 (p<0. 01) (Fig 6A–6D). Cells expressing undetectable levels of RFP-ICP4 or RFP had equally normally distributed free pools of GFP-H3. 1 or H3. 3 (Fig 6E, 6F, 6G and 6H). In contrast, cells expressing detectable levels of ICP4 had free pools of GFP-H3. 1 with a skewed distribution with a clear shoulder peaking at 40%. The increased dynamics of GFP-H3. 1 were also reflected by its nuclear distribution in U2OS cells (Fig 6I). H2B was mobilized in cells expressing ICP4, albeit its free pool increased the least of all core histones (Fig 4). H2B forms dimers with canonical H2A or any of its multiple variants. No H2A variant has been shown to interact with HSV-1 genomes, whereas canonical H2A has. We thus co-transfected cells with plasmids expressing GFP-H2A and RFP-ICP4. Surprisingly, the dynamics of canonical GFP-H2A were not significantly affected in cells expressing detectable levels of RFP-ICP4 (or free RFP) (Fig 7A and 7B). All linker histones are mobilized in cells infected with wild type HSV-1 [38]. Variant H1. 2 was mobilized the most, with a T50 in infected cells 60% of that in mock infected cells. H1. 2 is synthesized independently of the cell cycle stage and in all cell types that HSV-1 infects. We therefore focused on the mobilization of H1. 2 in cells expressing RFP-ICP4. The T50 of GFP-H1. 2 in cells expressing detectable levels of RFP-ICP4 was 76% of that in cells expressing undetectable levels of RFP-ICP4 (p<0. 01) (Fig 7C and 7E), and the free pools were 17% greater (p<0. 01) (Fig 7D). As expected, GFP-H1. 2 T50 or its free pools were not significantly different in cells expressing detectable or undetectable levels of free RFP (Fig 7D and 7E). HSV-1 n12 encodes only the amino-terminal 251 amino acid residues of ICP4. This mutant is unable to activate early or late gene expression [41]. The HSV-1 n12 mutant virus barely enhanced the dynamics of any histone in Vero or U2OS cells (Figs 1 and 2). The mutant form of ICP4 was therefore not expected to alter histone dynamics. To test this model, a plasmid encoding the n12 form of ICP4 fused in frame with red fluorescent protein was constructed (RFP-n12). Mobilization of core and linker histones was analyzed in cells expressing RFP-n12. The dynamics of no histones were altered in Vero cells expressing detectable or undetectable levels of RFP-n12 (Fig 8). Free pools, fast and slow exchange, or T50 of no histone were affected by expression of RFP-n12. GFP-H3. 1 had the expected punctuated localization in cells expressing detectable levels of RFP-n12, or of RFP (S1 Fig). HSV-1 DNA and ICP4 localize to the HSV-1 replication compartments, where they also co-localize with a small pool of histones (Fig 9). There was less fluorescence in the replication compartments than in the cellular chromatin (Fig 9), which may indicate fewer histones in the replication compartments or that the histones within the replication compartments are more dynamic and spend less time in them than in the cellular chromatin. A fluorescent micrograph cannot distinguish between 80% fewer histones or the same amount of histones having an 80% shorter residency time in the replication compartments. We therefore characterized next the histone dynamics in the replication compartments and the cellular chromatin of the same cell (Fig 10A). The free pools of core histones H2A, -H2B, -H3. 1, -H3. 3, and -H4, and that of linker histone GFP-H1. 2, all increased preferentially within the HSV-1 replication compartments (Fig 10). GFP-H4 and -H3. 1 had the largest average relative free pools in the replication compartments, 73 or 67% greater than those in the cellular chromatin, respectively (Fig 10B and 10C; p<0. 01). The free pools of GFP-H2A, -H2B, and -H3. 3 were 50%-56% larger in the replication compartments than in the cellular chromatin, whereas that of linker histone GFP-H1. 2 had the smallest difference, 41% larger in the replication compartments than in the cellular chromatin (Fig 10B and 10C; p<0. 01). The free pools were consistently higher in the replication compartments than in the cellular chromatin in all cells (Fig 10C). The average slow exchange rates of H3. 3 or H2B were 67 or 128% faster (P<0. 01), respectively, in the replication compartments than in the cellular chromatin, whereas those of other histones were not statistically different. It has recently been largely agreed that HSV-1 genomes are chromatinized during lytic infections [46–53], albeit the viral chromatin is far more dynamic than the cellular one [17]. Cellular chromatin containing transcribed genes is more dynamic than that containing silenced genes [21,54–56]. The dynamics of the viral chromatin are therefore consistent with the high rate of transcription of the viral genomes during lytic infection. A balance between cellular and viral effects may determine the unusual dynamics of HSV-1 chromatin. The assembly of nucleosomes with HSV-1 DNA may be a cellular response to inhibit HSV-1 transcription by assembling the viral genome in silenced chromatin. To counteract such silencing, HSV-1 would have evolved proteins to destabilize nucleosomes or to mobilize them away from its genome. Either mechanism would result in increased access by RNA polymerase II to the viral DNA, activating transcription. These HSV-1 proteins would thus be expected to act as transcription activators, without actually binding to specific promoter sequences. ICP4 is one of the three HSV-1 transcription activators, and the only one required for HSV-1 replication. Though it binds to specific DNA sequences to inhibit transcription, it does not do likewise to activate it [13,14]. Its mechanism of transcription activation remains only partially understood. Here we show that ICP4 is both necessary and sufficient to increase histone dynamics. Consistently with these findings, HSV-1 genomes are less accessible to nuclease digestion in the absence of IE proteins [57], and H3 association with HSV-1 DNA increases in the absence of functional ICP4 [58] (the changes were not statistically significant, perhaps due to the variability in the degree of the increased association for the ICP4 mutant virus). We selected the histones to be evaluated based on the following criteria. H1. 2 is expressed in all cell types that HSV-1 infects, and is mobilized the most of all linker histones [38]. H4 and H2B have no variants, and therefore represent the two core histone dimers, whereas H3 and H2A have several variants. H3. 1 and H3. 3 bind to HSV-1 genomes, via DNA-replication dependent or independent mechanisms, respectively [45], and their dynamics are differentially affected in HSV-1 infected cells [40]. Canonical H2A is the most prevalent H2A in nucleosomes, and no H2A variant has yet been reported to interact with HSV-1 chromatin. We analyzed histone dynamics at 4 or 7 hpi, and never beyond 8 hpi. At later times, chromatin shearing or marginalization [59] are likely to indirectly affect histone dynamics. The HSV-1 n212 ICP0 mutant induced increases in the free pools of all histones except H4 in U2OS cells larger than those induced by the wild type virus, suggesting that ICP0 may induce the degradation of the histones in the free pools. Nonetheless, the HSV-1 mutant encoding a truncated non-functional ICP4 n12 was the most defective. This mutant either failed to enhance histone dynamics (in U2OS cells) or only enhanced them to a basal level (in Vero cells), even though it overexpresses all other IE proteins. ICP4 may therefore modulate histone dynamics by itself. Alternatively, ICP4 could indirectly affect histone dynamics through any E protein, as the expression of all E proteins requires ICP4 (DNA replication or L proteins are not required [38–40]). To test these possibilities, we constructed plasmids expressing full length or truncated forms of ICP4 fused in frame with RFP. The dynamics of all core histones except H2A increased in cells transiently expressing ICP4 but not in cells expressing the non-functional truncated n12 mutant form of ICP4. H3. 3 is initially assembled in nucleosomes with HSV-1 genomes, whereas H3. 1 starts to be assembled in HSV-1 nucleosomes concomitantly with HSV-1 DNA replication [45]. Nucleosomes containing H3. 3 are more dynamic than those containing H3. 1 [30]. Therefore, H3. 1-containing nucleosomes would be expected to be less prone to support transcription than those containing H3. 3. The free pools of GFP-H3. 3 increased by only 15 or 22% in Vero or U2OS cells expressing ICP4, whereas those of GFP-H3. 1 increased by 248 or 40% in Vero or U2OS cells, respectively. ICP4 may thus preferentially prevent the assembly of HSV-1 genomes in the more stable H3. 1 containing nucleosomes. The free pool or slow exchange rate of GFP-H2B increased by 12% or 57%, respectively, in cells expressing ICP4. H2B forms dimers with canonical H2A or any one of its many variants. No H2A variant has yet been reported to bind to HSV-1 genomes. H2A was therefore expected to be mobilized in cells expressing ICP4. Surprisingly, it was not. It is thus most likely some other H2A variants are targeted by ICP4. Whereas H2A and H2A. X associate with both transcribed and silenced genes, for example, macroH2A preferentially associates with silenced ones [60]. Nucleosomes containing macroH2A are less dynamic than those containing canonical H2A [61,62]. Like its differential effects on H3. 1 and H3. 3, ICP4 could also preferentially mobilize particular H2A variants such as macroH2A away from HSV-1 genomes. If ICP4 itself enhanced histone dynamics, then one would expect histone dynamics to increase preferentially in the replication compartments, where ICP4 accumulates. Indeed, we found that the dynamics of all histones were faster in the replication compartments than in the cellular chromatin of the same nuclei. Though the free pools of all histones were greater in replication compartments, the slow exchange rates of only GFP-H2B and GFP-H3. 3 were significantly greater, and that of GFP-H2B nearly twice as much as that of GFP-H3. 3. Consistently, the slow exchange rate of only GFP-H2B was also significantly greater in cells expressing detectable levels of ICP4. HSV-1, and ICP4 in particular, may preferentially affect the less dynamic nucleosomes, which affect the slow exchange rate the most, over the more dynamic ones. All herpesviruses appear to encode proteins that regulate chromatin dynamics. These proteins are either tegument proteins, and therefore introduced into the cell with the infected virions, or expressed immediately upon nuclear entry of the viral genome. Either way, they are all available to remodel chromatin before the activation of generalized viral gene expression. The genomes of human cytomegalovirus (HCMV) are in much less dynamic chromatin in the absence of immediate early protein 1, for example, and the Epstein-Barr virus (EBV) major tegument protein BNRF1 binds to the H3. 3 chaperone Daxx, which physiologically assembles silencing H3. 3-containing nucleosomes in telomeres, thus preventing silencing H3. 3 incorporation in EBV chromatin [63,64]. Nonetheless, the genomes of beta- or gamma- herpesviruses are assembled in far less dynamic chromatin than those of the alpha-herpesviruses. The genomes of HCMV and EBV are less accessible to MCN digestion than those of HSV-1 [63,65–67], which is consistent with them being assembled in less dynamic chromatin. ChIP also co-immunoprecipitates relatively more EBV or HCMV than HSV-1 DNA, also consistent with the EBV and HCMV chromatin being less dynamic than that of HSV-1 [68–70]. Nucleosomes are also more uniformly assembled with EBV or HCMV than HSV-1 genomes [70–72], again consistent with less dynamic chromatin for the former. Alpha-herpesviruses also have much shorter replication cycles (~18 hours for HSV-1) than beta- or gamma- herpesviruses (~3 days for HCMV, ~4–5 days for EBV). ICP4 is conserved only among all alpha-herpesviruses, and not in beta- or gamma- herpesviruses. It is tempting to speculate that ICP4 may induce the particular dynamics of the alpha-herpesvirus chromatin, which would in turn result in the increased rate of transcription and consequently shorter replication cycle. HSV-1 genes are transcribed by the cellular RNA polymerase II complex, which is enriched on HSV-1 genes while depleted from cellular genes in lytic infections [73,74]. Nucleosomes impair accessibility of the RNA polymerase II complex to promoters DNA [21,54–56], and the HSV-1 chromatin is far more dynamic and accessible than the cellular one [17]. ICP4 may maintain the HSV-1 genomes in this dynamic and highly accessible chromatin, resulting in the RNA polymerase II complexes being sequestered away from the cellular genome and to the HSV-1 genomes [74], thus leading to the activation of HSV-1 transcription and inhibition of cellular transcription. In summary, we show that the HSV-1 transcription activator ICP4 is sufficient and necessary to enhance histone dynamics. ICP4 preferentially affects the silencing histone H3. 1 over the activating variant H3. 3, and it does not affect canonical H2A. ICP4 may therefore target silencing histones, preventing them from assembling silencing nucleosomes with HSV-1 genomes, or mobilizing them away from HSV-1 nucleosomes, to activate HSV-1 gene transcription. This mobilization may function to counteract a cellular defense mechanism against dsDNA viruses involving chromatin silencing. African green monkey Vero cells and their HSV-2 ICP4 expressing derivative n-33 cell line (a generous gift from the late Dr. P. Schaffer; [44]) were maintained in Dulbecco’s modified minimum Eagle’s medium (DMEM) supplemented with 5% fetal bovine serum (FBS) at 37°C in 5% CO2. Human osteocarcinoma U2OS cells (a generous gift from Dr. J. Smiley, University of Alberta) were maintained in Dulbecco’s modified minimum Eagle’s medium (DMEM) supplemented with 10% FBS at 37°C in 5% CO2. Wild-type HSV-1 strain KOS and mutant strain n12 (generous gifts from the late Dr. P. Schaffer) are described [75,76]. KOS viral stocks were prepared and titres were determined by standard plaque assay as described [38–40]. Preparation of n12 viral stocks and determination of n12 titres was as for KOS except that n-33 cells were used instead of Vero cells. Parallel n12 titrations of Vero cells ensured that the genetic defect in ICP4 was not rescued through recombination with HSV-2 ICP4 during viral stock preparation. The green fluorescent protein (GFP) -histone fusion plasmids were described previously [38,39,40,42]. pEGFP-H3. 3 was a generous gift from Dr. John Th’ng (Northern Ontario School of Medicine). The amino-terminal 2300 base pairs of ICP4 were amplified from HSV-1 DNA using primers 5’ AGA TCT CCG GAG GAT CGC CCC GCA TCG and 5’ CGT CCG AGC CGG GGG CGT CCG and the carboxy-terminal 1800 base pairs using primers 5’ CGG CGG CCC GCG ACC CCC and 5’ TCT AGA TCA CAA GCG CCC CGC CCC. The ICP4 PCR fragments were digested with SapI, and the amino- and carboxy- fragments were ligated with T4 ligase (Invitrogen). Full length ICP4 was then cloned into the BglII and XbaI sites of the pmCherry-C1 vector (Clontech). U2OS, Vero, and n-33 cells were transfected with Lipofectamine 2000, seeded onto coverslips, and infected for FRAP basically as described [38–40]. For co-transfections, 2. 2–2. 4x105 Vero or U2OS cells were transfected with 14 or 4 μl, respectively, of Lipofectamine 2000 reagent (Invitrogen), 0. 2 or 1 μg, respectively, of GFP-histone plasmid and 1. 8 or 1 μg, respectively, of RFP-ICP4, RFP-n12 or RFP-expressing plasmids. Following 6. 5 h incubation with the transfection mix, 1 ml of 37°C DMEM medium supplemented with 10% FBS was added to the cells. Cells were seeded onto coverslips at least 4 h later. Cells were incubated at 37°C for at least 12 (GFP-H2A, -H2B, -H3. 3, or -H1. 2) or 24 (GFP-H3. 1 or -H4) h after transfection before subsequent infection. Transfected cells were seeded onto coverslips and infected for FRAP as described [38–40]. Histone mobilization was evaluated by FRAP 24–48 h after transfection essentially as described [38,39]. A 1. 5 μm-wide region spanning the nucleus was photobleached, with 30 to 45 iterations at 95% intensity. Fifteen or more cells from at least three independent experiments were evaluated per treatment. Statistical significance was tested using one-tailed Student’s T test (for two-way comparisons) or ANOVA (for multiple comparisons). ICP4 localization was visualized by immunofluorescence as described [38]. Equal volumes of replication compartments or cellular chromatin in the same infected cells (Fig 10A) were photobleached at 7 hours post infection (hpi) with 10 PFU/cell of HSV-1, strain KOS. The fluorescence recovery within the photobleached regions was monitored once per second for the first 20 seconds and once every two seconds for the next 40 seconds. Background fluorescence was subtracted and fluorescence was normalized to that of the entire nucleus. Fluorescence at any given time is expressed as a percentage of the normalized fluorescence intensity within the same photobleached region (replication compartment or cellular chromatin) immediately before photobleaching.

The nuclear-replicating DNA viruses of the family herpesviridae cause a variety of diseases. Eight herpesviruses infect humans. Three of them, including herpes simplex virus 1 (HSV-1), belong to the alpha-herpesvirus sub-family. Viruses in this family have the fastest replication cycles of all herpesviruses, producing acute symptoms. During lytic infection, the genomes of HSV-1 associate with histones in more dynamic chromatin than those of the beta- and gamma- herpesviruses. The transcription activator ICP4 is conserved only among alpha-herpesviruses. Although ICP4 is essential, relatively little is known about its mechanisms of action. We have shown that histone dynamics are enhanced in HSV-1 lytically infected cells. Here we show that HSV-1 mutants in ICP4 are deficient in their ability to enhance histone dynamics. ICP4 was sufficient to enhance histone dynamics in the absence of other HSV-1 proteins or DNA. The dynamics of histones were greater in the viral replication compartments, where ICP4 localizes, than in the cellular chromatin. ICP4 may thus mobilize histones away from HSV-1 genomes to activate transcription. Such a mechanism of transcription activation would result in the highly dynamic nature of the viral chromatin and the fast replication cycles, and the acute pathologies, of the alpha-herpesviruses.

Abstract Introduction Results Discussion Materials and Methods

herpes simplex virus vero cells medicine and health sciences pathology and laboratory medicine pathogens biological cultures dna-binding proteins microbiology dna transcription viruses dna replication dna viruses epigenetics dna chromatin herpesviruses herpes simplex virus-1 research and analysis methods chromosome biology proteins medical microbiology gene expression microbial pathogens histones cell lines nucleosomes biochemistry cell biology nucleic acids viral pathogens genetics biology and life sciences organisms

2016

An Essential Viral Transcription Activator Modulates Chromatin Dynamics

Synchronized oscillation is very commonly observed in many neuronal systems and might play an important role in the response properties of the system. We have studied how the spontaneous oscillatory activity affects the responsiveness of a neuronal network, using a neural network model of the visual cortex built from Hodgkin-Huxley type excitatory (E-) and inhibitory (I-) neurons. When the isotropic local E-I and I-E synaptic connections were sufficiently strong, the network commonly generated gamma frequency oscillatory firing patterns in response to random feed-forward (FF) input spikes. This spontaneous oscillatory network activity injects a periodic local current that could amplify a weak synaptic input and enhance the network' s responsiveness. When E-E connections were added, we found that the strength of oscillation can be modulated by varying the FF input strength without any changes in single neuron properties or interneuron connectivity. The response modulation is proportional to the oscillation strength, which leads to self-regulation such that the cortical network selectively amplifies various FF inputs according to its strength, without requiring any adaptation mechanism. We show that this selective cortical amplification is controlled by E-E cell interactions. We also found that this response amplification is spatially localized, which suggests that the responsiveness modulation may also be spatially selective. This suggests a generalized mechanism by which neural oscillatory activity can enhance the selectivity of a neural network to FF inputs. Understanding the responsiveness of a cortical neural network is a fundamental requirement for any study of sensory information processing in the brain. Several experiments show that various factors can affect the neuronal response property and information flow in nervous systems: In the primary visual cortex, spiking responses of neurons can be enhanced by slow cortical oscillation [1]. The spike transfer function of thalamo-cortical neurons is modulated by noisy synaptic background activity [2]. Gain of neuronal responses is modulated by background synaptic input [3]. Even at the single cell level, cellular responsiveness is significantly influenced by the presence of voltage fluctuations [4]. It was shown recently that neuronal oscillations can increase response gain and decrease reaction time as a mechanism of attention selection [5]. Cortical neurons commonly show synchronous or oscillatory patterns of activity, which is thought to be important for cortical functions of information flow [6]. In particular, synchronous gamma frequency oscillations (30∼70 Hz) have been observed in various neural circuits [7], [8], [9], and they are thought to provide a temporal structure for information processing in the brain [10]. This gamma-band synchronization can be generated within local networks by coupling between GABAergic I- (inhibitory) interneurons and E- (excitatory) neurons [11], [12], [13], [14], and is related to cognitive functions [15], [16], and information delivery [15]. This population activity also has been studied in numerical simulations and mathematical models [12], [17], [18], [19]. Previous analyses have shown that cortical oscillations are generated in networks with appropriate connectivity and can be correlated with the firing phases of E- and I- neurons, but the effect of these oscillations on the neural network responsiveness to external inputs remains elusive. In this research, using a large network model of Hodgkin-Huxley type E- and I- neurons, we study how spontaneous cortical oscillation - particularly in the gamma frequency band - modulates the response property of a neural network. We examine the cortical responsiveness to external FF inputs at the single-spike level because the input-output response function for a single input spike is a fundamental feature of neural networks for information processing. A recent study emphasizing the importance of single spike level analysis showed that a significant amount of visual information can be delivered by the very first spike emitted by a neural population [20]. We found that spontaneous cortical oscillation activity noticeably changed the cortical input-output response function. For example, weak inputs that are normally missed in the responses of single neurons, were significantly enhanced by cortical oscillations in the network. This response modulation was similar to the observed effect of the membrane potential oscillation reported in a previous experimental study [21]. More importantly, we found that this cortical response modulation by the oscillation activity was controlled by external feed-forward (FF) input strength variation. This means that the cortical network can self-regulate by differentially amplifying its FF inputs according to their strength, without physiological changes in single neuron properties or structural modulation of interneuron connectivity. We show that this ‘differential’ amplification results from the modulation of gamma oscillation by cortical E-E neuron interaction. We suggest that this is an important example in which the modulation of gamma oscillation by cortical E-E interactions [12] can significantly change the population responsiveness. We also found that this cortical amplification effect was restricted spatially to an ‘oscillation active’ region, which enables the spatially-selective tuning of responsiveness to given FF input. We find that this spatial localization is determined by the range of anatomical interneuron connectivity. This is consistent with recent experimental findings concerning the effects of cortical oscillations [1], [5], [22], and points to aspects of this oscillation which effectively enhance the response selectivity of a neural network to FF inputs. We examined a variety of neural network activity patterns produced under different conditions. We used a cortical network model in which E- and I- neurons interact with each other via lateral synaptic connections. We constructed isotropic local cortical connections, using physiological and earlier modeling data [23], [24]. We varied the strength of each type of cortical interaction (E-E, E-I, I-E and I-I) in order to study different cortical connectivity conditions. Artificially generated random spikes were injected into the cells in the center area (diameter∼450 µm, ∼500 neurons: E- 377, I- 113) of this network model (1 mm by 1 mm, consisting of ∼3300 neurons: E- 2500, I- 841) to simulate localized FF spike input to the network. The actual spike pattern for each neuron was generated by a Poisson process with constant mean firing rate, and FF input strength (The amount of intracellular conductance fluctuation caused by a single FF input spike) was varied within the range 5∼100 µS/cm2, as a control parameter. By performing many simulations of different cortical parameters with FF input rates in the range 5∼40 spike/s, we observed several different types of cortical activity patterns. Gamma oscillation can be generated by interactions among E- and I- cells: The E- cells synchronize the I- cells, and the gamma-modulated I- cells drive E- cells to generate gamma frequency rhythms in the network [11], [14], [25]. Our simulations agreed with earlier studies that gamma oscillations are generated by applying E-I and I-E cortical connections; I- cells were synchronized by E-I connections first (Figure 1A), then I-E connections generated gamma rhythm in E-cells via periodic inhibitory activity (Figure 1B) as in the PING model [11], [14]. I- cells fire at higher rate than E- cells, just as fast-spiking cells fire at higher rates than regular-spiking cells [10]. The relative firing phase of E- and I- cell (Figure 1A and 1B) also showed a similar phase relation as reported in the previous experiments [25], [26], [27]. E- cells fire 3∼5 ms before I- cells fire, with an approximately 70-degree phase difference in a gamma cycle. We extend the previous studies by also explicitly considering E-E and I-I cortical connections. In ref. [25], Morita et al. showed that gamma- modulated cortical excitation increases the firing rate of E- cells and decreases the E-I firing phase difference. Based on these observations, they predicted that the gamma- modulated E-E cell interaction will suppress the cortical oscillation. In this study, we found that the gamma- modulated E-E coupling does not extinguish the cortical oscillation. When E-E connections were turned on, the firing rate of E- cells was increased and the E-I phase difference was diminished (Figure 1D and 1E), as shown in ref. 25. But these changes did not actually suppress the cortical oscillation. Instead, they caused a transition of operating ‘mode’ such that the oscillation frequency dropped to a low gamma or near beta rhythm [28]. This transition could not be observed using the methods reported in ref 25, because they have only ‘static’ data of pre- and post- synaptic activities to examine ‘static’ equilibrium conditions. The approximately zero E-I firing phase difference is an important feature of E-E coupling, and is distinguishable from that of the case involving no or little E-E coupling (∼70-degrees is usually observed in experiments [26], [27]), so we call the former situation ‘E-E interaction modulated’ E-I phase, in contrast to the ‘normal’ gamma oscillation E-I phase. We will analyze below how this phase modulation results from E-E coupling. The difference between the firing rates of E- and I- cells is also diminished to some degree by E-E interactions, and the population oscillation frequency is also lowered [28], [29]. We maintain this non-zero E-E interaction throughout the following simulations. We also allow I-I connections, which reduce the firing rate of I- cells to some extent (Figure 1C and 1D). In this study, we do not examine the role of I-I connection in detail, and the I-I connection strength was not varied. When the cortical interactions (E-E, E-I, I-E and I-I) are completely turned off and each neuron is driven by FF inputs only, there is no correlated behavior in the network and the average network firing rate simply follows the instantaneous input firing rate (Figure 2A, FF). On the other hand, when cortical interactions (E-E, E-I, I-E and I-I) are introduced, the network exhibits some synchronized patterns depending on FF input and cortical connectivity parameters. For example, with a moderate FF input strength (∼35 µS/cm2) and rate (∼40 spikes/s), the neural population shows an oscillation pattern in its firing rate for a broad range of cortical connectivity parameters. The cross-correlogram among cortical spikes shows a clear oscillatory pattern for both E- and I- cells in this case (Figure 2B, OA). Generally, the amplitude of the instantaneous output firing rate of the neural population depends on the FF input firing rate. The frequency of the oscillation is somewhat affected by the FF input strength and cortical connectivity parameters, but the oscillation frequencies are mostly within the gamma band range (25∼50 Hz), as in earlier experimental observations and theoretical models [8], [9], [30], [31]. For some parameter range of cortical connections, this gamma oscillation can be very small. When the FF input spike rate was low (∼10 spikes/s) with the other parameters unchanged, the oscillating firing pattern in E- cells became barely detectable even though the oscillation in I- cells was maintained to some extent (Figure 2C, OI). We examined the phase of the firing pattern of E-I cells in three cases (FF, OA, and OI). The E-I spike correlogram in the OA network showed that the effect of interactions among E-E cells is significant because the E- and I- cells fire with little phase difference (Figure 2B (iii) ), similar to Figure 1D and 1E (‘E-E interaction modulated’ E-I phase), while the other two cases (FF and OI) showed no clear phase relation (Figure 2A (iii) and 2C (iii) ). Since OA and OI networks have the same parameters except for the FF input firing rate, we conclude that FF input firing rate is a crucial factor in determining E-E interaction strength. When E-E cortical connections are very strong, extremely sharp cortical spike synchronization is generated, resulting in spatially propagating waveform patterns [32], [33], [34]. The amplitude of the instantaneous output firing rate was almost constant, independent of FF input rate. In this case, the neural response was not controlled very much by FF inputs at each moment but mostly by the spatial cortical bursting pattern. We observed that the frequency of this periodically propagating pattern is in the beta oscillation range induced by E-E cortical interactions [28]. We do not examine the generation or the effects of periodically propagating patterns any further here. In the following simulations, we chose cortical conditions such that the system didn' t enter this phase for the FF input strengths tested. To examine the responsiveness of the neural network, we chose a range of parameters that provided moderate and stable oscillatory behavior on application of FF input spikes. We tested this condition using sinusoidal time-varying input rates with a peak amplitude range of 0∼60 spikes/s and peak frequency of 5∼10 Hz. The network rapidly and reliably restored its oscillating state whenever the FF input firing rate became greater than some level (∼20 spikes/s), and the oscillations diminished significantly and very rapidly when the input rate fell below that level. Throughout this research, we did not change any individual neuronal property (e. g. ionic channel parameters). We compared the neural responsiveness for the following three states of network activity: (i) network with no cortical connectivity (and no spontaneous network activity) (FF), (ii) laterally connected cortical network with spontaneous oscillation activity (OA) and (iii) the same network as in (ii) but with little or no oscillation activity (OI). We generated random FF spike inputs by a Poisson process and provided this input to the center area (diameter∼450 µm) of the 1 mm by 1 mm network. All the response properties were measured within this small center area in order to avoid finite size effects from the network boundary. Neurons outside this area received no FF input. FF input strength was varied from 5 to 100 µS/cm2, and FF spike rate was kept constant at 40 spikes/s for (i) and (ii). For (iii) the oscillation inactive case, the FF input rate was lowered to 10 spikes/s in order to maintain minimal oscillations while still providing enough input spikes to allow measurement of the network responsiveness. Figure 3A shows FF input and cortical output spike trains, the membrane voltage, and the internal current fluctuation of a sample neuron with and without network oscillation activity. Each FF input spike induces a synaptic conductance fluctuation in a cortical neuron and the corresponding intracellular current fluctuation. When cortical interactions are turned off, the response of each neuron depends only on the direct FF input (FF network, Figure 3A (i) ). When the FF input strength is weak (25 µS/cm2), a single spike input could not produce a large enough current fluctuation to generate an output spike. Only when two or more inputs are temporally paired within a short time interval (<∼10 ms), can the conductance fluctuations from each spike overlap to generate an output spike (Figure 3A (i) ***), as found in the measurements of correlated inputs in a previously reported experiment [35]. In this case, there is a ‘threshold FF input strength (Sthresh) ’ that determines whether each ‘single’ FF input spike can generate an output spike or not. We calculated the responsiveness of network neurons, using a cross-correlation method (Figure 3B) [35], [36]. The responsiveness of the cortical network was defined as (net peak integral) / (number of FF input spikes), where net peak integral is the total area of the maximum response peak above the background activity in each cross-correlogram. Measuring this quantity at each FF input strength produces the generalized response function of the network (Figure 4A). Since any subthreshold single input produces no response when cortical connections are turned off (Figure 3B unpaired), the response to a single FF input spike is like a step function, similar to the measured thalamo-cortical transfer function in the absence of noisy background activity [2], providing the cell with a simple spike relay capability with limited encoding abilities. When the cortical connections are turned on, each neuron receives cortical synaptic inputs from other neurons as well as FF inputs. In the presence of the spontaneous network activity of synchronized oscillating patterns, the cortical spike inputs that each neuron receives are also oscillating (Figure 3A (ii) ). The current fluctuation due to cortical interactions is added to that by FF inputs, and as a result, a single-spike sub-threshold FF input can produce an output spike response with the help of this additional cortical activity (Figure 3A (ii) *). This input amplification depends on the phase of the cortical activity. When FF input timing is out of phase with the cortical oscillation, it fails to produce an output spike for lack of additional cortical current (Figure 3A (ii) **) just as in the FF network. The general response function of the network to a single ‘unpaired’ input spike is plotted in (Figure 4A), as a function of FF input strength. Different from the step-like FF case, the slope of the response function changes more gradually, with a plateau near the FF threshold input strength (Sthresh∼30 µS/cm2). This broader and more gradual change of response function can provide richer encoding capabilities [2]. Next, we examine how responsiveness changes when the oscillation is inactive while the connectivity of neural population is kept the same. We lower the FF input rate to 10 spike/s, so that the spontaneous oscillation is almost absent. All the other parameters including cortical connectivity are kept the same so that each synaptic interaction via spike delivery in the network can give the same contribution as before. This time, neurons do not exhibit enhanced responsiveness for weak inputs; each neuron still experiences some conductance change by cortical interaction, but its contribution is negligible. The network response character is similar to that of the FF network (Figure 3A). Any ‘unpaired’ weak inputs cannot generate a cortical response, losing its information. We found the response function of neurons is almost the same as that of the FF network (Figure 4A) when the cortical connections are turned off. In Figure 4A, the absolute difference in responsiveness between the oscillatory network and the FF network is large when FF input strength is weak (near the FF response threshold, Sinput∼30 µS/cm2). This difference becomes smaller as the FF input strength increases. Finally the two response functions become equal at very strong inputs (Sinput>80 µS/cm2). In other words, the cortical amplification due to oscillation activity is large for weak FF inputs, and becomes insignificant as inputs become strong. We examined how this cortical response modulation is related to the change of oscillation strength. Figure 5A shows values of the network oscillation as a function of FF input strength. For weak FF inputs (Sinput = 25,35 µS/cm2), the cortical oscillation is strong but it becomes weaker for a medium input strength (Sinput = 50 µS/cm2), and almost disappears for a strong input (Sinput = 80 µS/cm2). This change in the oscillation strength is almost proportional to the extent of the cortical response modulation which can be roughly defined as the difference between the response function and that of the FF network (Figure 5B). As shown by examining the E-I firing phase pattern (Figure 5A (iii) ), the oscillation modulation is significantly affected by the E-E cell interaction. When the FF input is strong, the E-I spike correlogram shows ‘normal’ E-I phase difference of the gamma oscillation (Figure 5A ***). On the other hand, when the FF input is weak, the relative E-I firing phase is near zero (‘E-E interaction modulated’), showing that the network is affected by strong E-E interaction (Figure 5A (iii), Sinput = 25,35 µS/cm2). For some range of FF input strength, two different peaks coexist in the E-I correlerogram (Figure 5A * and **). Whenever the E-I firing phase is significantly affected by the E-E interaction (∼70-degrees→∼0-degrees), the cortical oscillation becomes strong and the responsiveness of the network to weak FF input is enhanced. When the E-E connections were turned off (Figure 6), there was no E-I firing phase modulation (Figure 6 (iii) ), and the cortical oscillation was far less affected by the FF input strength (Figure 6 (i) ). For weak FF input, the cortical oscillation almost disappeared (Figure 6 (i) input strength 25 µS/cm2) and consequently the responsiveness did not exhibit any enhancement. We examined how the E-I firing phase is modulated from ∼70-degrees to ∼0-degrees by the excitatory interactions in the E-E couplings. We turned on only excitatory cortical connections (E-E and E-I) and measured the firing patterns of E- and I- cells (Figure 7). In this condition, the network generated the periodically propagating waves near beta rhythm that we observed in the earlier part of this study (when E-E connections were relatively stronger than other types of connections). In each wave cycle, a small number of E- cells fired due to the FF inputs (Figure 7A i). Then they excited nearby E- and I- cells through the E-E and the E-I connections. The stimulated E- and I- cells fired almost simultaneously, or I- cells fired slightly before E- cells fire (Figure 7A ii and iii) because E-I connections are stronger than E-E connections. The firing of the E- cells continued for a while due to the ‘chain reaction’ of E-E couplings and the I- cells occasionally produced ‘spike doublets’ by this sustained excitation (Figure 7A iv and 7B). These inhibitory spike doublets are frequently observed during long range synchronization processes in neural networks [9], usually along with the ‘delayed’ excitation from distant cells. In our study the E→E→I route could provide the ‘delayed’ or ‘sustained’ excitation. As a result of these E- and I- firing patterns, the E-I firing phase was ∼0-degrees on average. When the average excitation to each I- cell is strong enough to produce I- spike doublets, the E-I phase correlogram has two peaks around 0-degrees (Figure 5A * and **). If the excitation to I- cells is not enough, or the inhibition of E- cells is strong so that the sustained E- cell activity cannot drive I- cells to produce the second spike of spike doublets, the E-I phase correlogram does not show clear peak separation (Figure 5A (iii) input strength 25 µS/cm2). In any case, the average E-I phase difference is close to 0-degrees, clearly different from ∼70-degrees ‘normal’ gamma phase distribution. The cortical response modification caused by this oscillation can be explained as an effect of membrane potential oscillations [21]. In addition, the modulation of the E-E interaction by the FF input strength can be understood as follows: as shown in Figure 3A, when the FF input strength is weak, each E- neuron can respond to individual FF input only with the help of the collective cortical activity of E- cells whose contribution is intrinsically periodic (oscillatory). The probability of a cortical response to an FF input spike strongly depends on the phase of oscillation of the E- cells (Figure 3A * and **), and this dependence is significantly strengthened by the E-E coupling loop. In the beginning of a gamma oscillation cycle, only a small number of E- cells fire together, but they trigger an E-E coupling loop which drives more E- cells to fire simultaneously. As a result, the peak firing rate of E- cells in a gamma cycle is quite high, causing much higher spiking probability near the oscillation peak. Therefore, (i) the cortical responsiveness is dependent on the oscillation phase, and (ii) the gain or the cortical amplification is proportional to the oscillation strength. When an FF input is strong enough to produce an output spike in each E- cell, there is no significant dependence of the E- cell response on the periodic cortical activity and E-E couplings, and each individual E- cell responds directly to its FF input pattern, independently of the network activity. Since the FF input spike train was generated by a random Poisson process, the cortical response pattern is also random and not oscillatory. Although the average output firing rate generally increases with increasing FF input strength, the cortical response modulation actually decreases. The oscillations are depressed by increasing the FF input strength while keeping the FF input rate constant. The average number of FF input spikes does not decrease but the oscillation is depressed, and the response function converges to the FF result. Although cortical oscillation is initially established by the E-I and the I-E interactions, the E-E cell interaction is crucial for responsiveness modulation because its strength strongly depends on the FF input strength. Since the cortical response modulation is controlled by the FF input strength in the system via the self modification of network oscillation, this seems to be a very effective automatic gain control system that does not require any synaptic adaptation or learning mechanism [37]. Next, we examine how the oscillation activity affects the response delay of the network. Generally, the response time (time interval between an FF input spike and a corresponding output spike) is relatively long (∼10 ms) for weak inputs and becomes shorter as input strength increased for all cortical states (Figure 4B). But there is a significant difference in the average response time depending on the network activity state. Figure 4B shows that the average response delay is shorter during spontaneous oscillations than for the other two cortical activity states considered, especially when the FF input is weak. As input strength increases, the response of the oscillating network becomes similar to that of the oscillation inactive network, but still faster than that of the FF network. For a sufficiently strong FF input, the response time delay curve of the oscillating network and the oscillation depressed network were almost the same, with a delay of about 4 ms, agreeing with the experimentally known value for monosynaptic connections [35]. This is still faster than that measured for the FF network. Cortical interaction adds some positive current to each neuron, and this additional current causes the membrane voltage to reach the action potential threshold faster. In the presence of cortical oscillation, the amount of cortical current added is larger in the positive phase of the oscillation and on average, the larger the oscillation, the greater the average positive current that is added to the cell. That leads to faster responses than observed for the simple FF network. As the FF input strength increases above 35 µS/cm2, the oscillation amplitude decreases (Figure 5A), and the net average positive current added is less. The response time difference between oscillatory and FF networks decreases accordingly (Figure 5B). E-E cell interactions control the cortical oscillation strength, and the extent of response delay modulation and the response modulation of the network are in turn proportional to the cortical oscillation strength. To examine the spatial localization of the cortical gain resulting from the spontaneous oscillation of network activity we used a network model four times the size (2 mm by 2 mm, 13000 neurons) of that used in the studies described above. In the center region (diameter∼450 µm, including ∼500 neurons, Figure 8A G1), the cells were activated as before with a FF input rate of 40 spikes/s, a rate that was also used to set up the spontaneous network activity oscillations for the simulations described above. The surrounding neurons (Figure 8A G2∼G11) received a signal with an FF input rate of 10 spikes/s, a rate at which spontaneous oscillations are hardly evident previously. The cortical neural connectivity is the same everywhere so that the center and the surrounding neurons could interact with each other. There are differences in the responsiveness of the network between the central region that shows strong spontaneous oscillations and the surrounding regions with little to no oscillations. In other words, the responsiveness modification by the spontaneous oscillation can be localized. A control simulation where all cortical neurons received an FF input of 10 spike/s in both the center and the surrounding regions is an approximation of the neuronal property at infinite distance from the center region (region G∞ in Figure 8A). For measurement purposes the network is divided into circular annuli (Figure 8A G2∼G11). Each annulus contained 400 to 500 neurons. Figure 8A shows the network response function and the response delay time for single unpaired FF spike inputs in each region. Figure 8B shows the measured values in all the regions for FF input strengths of 25 and 35 µS/cm2 which are slightly smaller and slightly larger, respectively, than the FF response threshold value (Sthresh = 30 µS/cm2, Figure 4A). When the FF input strength was weaker than Sthresh, the responsiveness and the response delay pattern gradually moved from the oscillation-active center region (G1) to approach the control oscillation suppressed case (G∞). Subsets of the Figure 8A graphs show that the quantities measured in the surrounding regions G2∼G6 are continuously distributed in the interval between the values obtained for G1 and G∞. Values measured in surrounding regions G7∼G11 showed negligible differences from those measured for G∞, and are not plotted in the figure. The effect of oscillatory network activity is restricted to a central local region about 500 µm in radius. When the FF input strength is larger than Sthresh but not very large (30 µS/cm2<Sinput<50 µS/cm2), the spatial attenuation of the cortical activity effect is more apparent (Figure 5B, Sinput = 35 µS/cm2). In this case, all surrounding regions, even including the nearest region G2, are clearly separated from G1, and are close to G∞. In this case, the cortical oscillation activity effect is restricted to the central activated region (G1). For very strong FF inputs (50 µS/cm2<Sinput), all response properties converge to control group (G∞) behavior as expected since under that condition oscillatory behavior is barely evident even in the central region. It seems reasonable to expect that the spatial localization of gamma oscillation is dependent upon the range of single neuron synaptic connectivity so that the shorter synaptic connection range, the smaller the surrounding area that is affected by the oscillation in the center. To verify this expectation, we reduced the excitatory and the inhibitory synaptic interaction range from the initial value (radius of 200 µm for E- cells, 100 µm for I- cells), keeping the ratio of E- and I- range the same. In Figure 8C, the range of oscillation effect is proportional to the single cell synaptic connection range, as expected. This suggests that the effective range of gamma oscillations is strongly dependent upon the details of the anatomical connectivity of neurons in experimental observations. Thus the effect of cortical oscillation is fairly well localized for weak and moderate FF input strengths. Neural response properties are modified only within or near the area in which spontaneous oscillation is activated. This suggests that spatially selective cortical response modification is possible. Spontaneous cortical oscillations can be locally induced by spatially localized thalamic inputs, and the cortical response character can also be selectively tuned within a limited region. It was previously thought that cortical E-E activity interaction is not essential for gamma rhythm generation but can modify the oscillation frequency and the phase of cell firing pattern [11], [28], [29]. A recent experimental study showed that gamma rhythms in E-E cells activity can change the E- and I- cells firing phase distribution, and suggested that it may suppress the gamma oscillation [25]. We have shown that recurrent E-E interaction significantly modulates the oscillation frequency, the firing phase distribution of E- and I- cells, and the oscillation strength, without extinguishing the cortical oscillation. Moderate levels of E-E interaction generally strengthen the oscillation, causing the ∼0-degrees E-I firing phase and the lower oscillation frequency (near beta range). As a result, it modulates the cortical response function that is clearly relevant to encoding/decoding of information. The fact that the effect of E-E interaction is self regulatory for FF input strength variation suggests a useful mechanism for the cortical gain control, without neuronal feedback loops from the visual cortex to earlier visual stages. This suggests a general mechanism by which the same types of neurons in different cortical layers can have different properties due to the different coupling within each layer. In previous studies, it was reported that the upper and the lower layers of the cortex can have different oscillation characters and phase response properties [38], [39], [40]. Our observations about the phase and the frequency modulation of cortical oscillation by E-E coupling, suggest that different neuronal properties in different cortical layers may originate from the different lateral connectivity (especially E-E coupling) strength in each layer. For example, neurons in the different hippocampus regions (CA1 and CA3) show noticeably different firing phase distribution and correlation activity patterns in gamma oscillations [27]. Considering that the E-E couplings are significant in CA3 [41] while they are sparse in CA1 [42], this may be a situation in which the E-E coupling property plays an important role in the modulation of neural activity, as suggested above. Previous experimental work has shown that (i) the visual responsiveness of the cortical network is significantly enhanced by slow cortical oscillation [1] and (ii) the phase of slow theta rhythm (4∼8 Hz) oscillation modulates the high frequency gamma (80∼150 Hz) band oscillation power [22]. Here we provide a clue to a mechanism for modification of neural response properties by various types of synchronized cortical network activities. As shown above in the results section, when the gamma frequency oscillation is generated, it enhances the neuronal responsiveness to weak FF input. If this gamma power is modulated by slower (theta or lower frequency) rhythm, then the network responsiveness could depend on the phase of this slow oscillation. This suggests a simple and consistent basis for the modulation of high frequency oscillation power by lower frequency activity. Some previous experimental work has shown that the power and the frequency of gamma oscillations can be modulated by various factors such as the operation of fast spiking interneurons and some neuromodulators [10]. In our simulation, the strength of the gamma oscillations can be significantly modified by changing the strength or rate of the FF input, with the help of E-E interactions but without changing any individual neuronal properties or network connectivity features. In addition, our results on the gamma oscillation modulation mechanism do not require modifying the FF input correlation pattern, learning/adaptation behaviors [37] in cortical synapses, or cortico-thalamic feedbacks [43]. If the slow frequency oscillations affect the FF input strength or the input rate locally within the network, the gamma oscillations will be readily modulated. The consequent modulation of the responsiveness will depend on the phase of the low frequency oscillation [1]. This proposed responsiveness tuning mechanism does not require any accompanied background activity control. Therefore it is simpler than those gain control models suggested in the previous reports [2], [3], [4] that are mostly dependent on the modulation of the background activity. An important advantage of the present model is that the response modification can be ‘dynamically selective’ according to the FF input strength and the FF input projection range variation. Since the cortical gain is modulated by FF input strength, the cortical amplification is selective to FF input strength. The system effectively determines the ‘optimized’ gain via modulations of oscillation power spontaneously, and can avoid unnecessary adjustments even without any delayed feedbacks to thalamus or thalamo-cortical neurons. Moreover, the tuning is spatially localized to distances of less than about 500 µm for weak FF inputs, and less than 50 µm for strong inputs. This is comparable or even smaller than the size of the receptive field of single neuron in the mammalian primary visual cortex [44]. Therefore responsiveness modulation can be spatially selective, and this is more effective than mechanisms proposed in previous studies [2], [3], [4] where the cortical modulations were generally global and the gain optimization could not be achieved easily. We also suggest that this mechanism might be applicable to the functional modulation of the relative weight between thalamic inputs versus cortical inputs to the visual cortex neurons [45]. In some experiments with monkeys, when attention is directed, visual sensitivity increased due to the increased synchronization between the visually evoked potentials and the stimulus [46]. In another report, neurons activated by the attended stimulus showed increased gamma frequency synchronization [47]. Considering the response enhancement by gamma oscillation in our model together with these experimental measurements, spontaneous gamma band activity seems to play an important role for regulating the information flow in the visual nervous system, as suggested previously [6], [16]. Our findings support these experimental observations, and may suggest a new mechanism for attention modulation that is different from that of other models [48], [49], [50]. This neural network model consists of a two-dimensional layer of coupled neurons, each modeled as a Hodgkin-Huxley neuron with Na+, K+ and Cl− ion channels and E- and I- synaptic conductance channels. 75% of the neurons are E- and 25% are I- as suggested by experimental data [51], and adopted in other publications [24]. We used networks of two sizes in this research: (i) 1 mm by 1 mm, including ∼3300 neurons for responsiveness simulation and (ii) 2 mm by 2 mm, including ∼13400 neurons for locality simulations. The membrane potential of the jth neuron, , is determined bywhere σ is E or I depending upon whether the neuron is E- or I-, respectively, C is the membrane capacitance, and gL is the leakage conductance. gjσE is the synaptic conductance of the jth neuron, E- or I- as specified by σ, providing the cortical inputs from the neurons in its spatial neighborhood, and gjσI is the synaptic conductance of the jth neuron providing the I- input from the neurons in its spatial neighborhood. We used the commonly accepted biophysical parameters [52], [53]: the capacitance C = 10−6 Fcm−2, the leakage reversal potential VL = −70 mV, the Na+ equilibrium potential VNa = 55 mV, the K+ equilibrium potential VK = −80 mV, the E- reversal potential Vexc = 0 mV, the I- reversal potential Vinh = −80 mV, the leakage conductance gL = 50*10−6 Scm−2, and Hodgkin-Huxley Na+ and K+ conductances gNa = 120*10−3 Scm−2, gK = 36 *10−3 Scm−2. The Hodgkin-Huxley ion channel conductance takes the form [54]: where m, h and n denote the channel activation variables. The rate constants take empirically known forms [55]: A jth neuron' s synaptic conductance is given by: , and are the spatial, the E- temporal, and the I- temporal kernels of cortical interaction, respectively, which describe the contribution of lth spike from kth neuron to the jth neuron. For cortico-cortical connection, the spatial kernel in the synaptic conductance equation takes the form: where and are the jth and kth neurons' spatial positions respectively. The decay constant, is 200 µm (for E- connections) and is 100 µm (for I-). The temporal kernel in the equation is set to beand the time constants in milliseconds were chosen as (3,1) for E- and (7,1) for I- synapses where Csσ′ and Ctσ′ are normalization constants chosen so that that the sum of the contributions of the two kernels would sum to unity. We assume spatially isotropic local connections with a range of 200 µm in radius for E- and 100 µm in radius for I- synapses. Wσσ′ are strengths of synaptic connections for the neuron pair of type (σ, σ′). If all Wσσ′ = 0, the network is then equivalent to the simple FF model. When the cortical synaptic connections were turned on, these values ratios were set to (WEE, WIE, WEI, WII) = (0. 03: 0. 06: 0. 12: 0. 12) throughout the simulation. This condition was achieved from the parameter search simulations shown in the first part of results section. The contribution to the E- conductance by the FF input spikes was given by: Sinput is the weighting factor for FF input synaptic strength and gjinput was varied within 5∼100 µS/cm2, throughout the simulations reported here. The temporal kernel has the same form as the E- temporal cortical kernel given above. The spike timings, , of input were generated by Poisson processes. All of our simulations were coded using the GENESIS 2. 3 environment [55], and performed with a Pentium IV PC system. Simulation outputs were analyzed using Matlab R2006b scripts.

In the nervous system, information is delivered and processed digitally via voltage spikes transmitted between cells. A neural system is characterized by its input/output spike signal patterns. Generally, a network of neurons shows a very different response pattern than that of a single neuron. In some cases, a neural network generates interesting population activities, such as synchronized oscillations, which are thought to modulate the response properties of the network. However, the exact role of these neural oscillations is unknown. We investigated the relationship between the oscillatory activity and the response modulation in neural networks using computational simulation modeling. We found that the response of the system is significantly modified by the oscillations in the network. In particular, the responsiveness to weak inputs is remarkably enhanced. This suggests that the oscillation can differentially amplify sensory information depending on the input signal conditions. We conclude that a neural network can dynamically modify its response properties by the selective amplification of sensory signals due to oscillation activity, which may explain some experimental observations and help us to better understand neural systems.

Abstract Introduction Results Discussion Methods

computational biology/systems biology biophysics/theory and simulation neuroscience/theoretical neuroscience

2009

Spontaneous Local Gamma Oscillation Selectively Enhances Neural Network Responsiveness

The Mitogen-Activated Protein Kinase (MAPK) network consists of tightly interconnected signalling pathways involved in diverse cellular processes, such as cell cycle, survival, apoptosis and differentiation. Although several studies reported the involvement of these signalling cascades in cancer deregulations, the precise mechanisms underlying their influence on the balance between cell proliferation and cell death (cell fate decision) in pathological circumstances remain elusive. Based on an extensive analysis of published data, we have built a comprehensive and generic reaction map for the MAPK signalling network, using CellDesigner software. In order to explore the MAPK responses to different stimuli and better understand their contributions to cell fate decision, we have considered the most crucial components and interactions and encoded them into a logical model, using the software GINsim. Our logical model analysis particularly focuses on urinary bladder cancer, where MAPK network deregulations have often been associated with specific phenotypes. To cope with the combinatorial explosion of the number of states, we have applied novel algorithms for model reduction and for the compression of state transition graphs, both implemented into the software GINsim. The results of systematic simulations for different signal combinations and network perturbations were found globally coherent with published data. In silico experiments further enabled us to delineate the roles of specific components, cross-talks and regulatory feedbacks in cell fate decision. Finally, tentative proliferative or anti-proliferative mechanisms can be connected with established bladder cancer deregulations, namely Epidermal Growth Factor Receptor (EGFR) over-expression and Fibroblast Growth Factor Receptor 3 (FGFR3) activating mutations. Mitogen-activated protein kinase (MAPK) cascades can be activated by a wide variety of stimuli, such as growth factors and environmental stresses. They affect diverse cellular activities, including gene expression, cell cycle machinery, apoptosis and differentiation. A recurrent feature of a MAPK cascade is a central three-tiered core signalling module, consisting of a set of sequentially acting kinases. MAPK kinase kinases (MAPKKKs) are activated following upstream signals. For instance, they can be phosphorylated by small G-proteins belonging to the Ras/Rho family in response to extracellular stimuli. Their activation leads to double phosphorylation and activation of downstream MAPK kinases (MAPKKs), which in turn double phosphorylate MAPKs. Once activated, MAPKs act on their target substrates, which include other kinases and transcription factors [1]. To date, three main cascades have been extensively studied, named after their specific MAPK components: Extracellular Regulated Kinases (ERK), Jun NH2 Terminal Kinases (JNK), and p38 Kinases (p38). These cascades are strongly interconnected, forming a complex molecular network [1]–[4]. MAPK phosphorylation level is regulated by the opposing actions of phosphatases. As the effects of MAPK signalling have been shown to depend on the magnitude and duration of kinase activation, phosphatase action might play an important functional role [5]. Moreover, scaffold proteins bring together the components of a MAPK cascade and protect them from activation by irrelevant stimuli, as well as from negative regulators (such as phosphatases) [6]. The involvement of MAPK cascades in major cellular processes has been widely documented [1], [7], [8]. However, the wide range of stimuli and the large number of processes regulated, coupled with the complexity of the network, raises the fundamental and debated question of how MAPK signalling specificity is achieved [9]. Several interrelated mechanisms have been proposed: opposing action of phosphatases; presence of multiple components with different roles at each level of the cascade (e. g. different isoforms of a protein); interaction with scaffold proteins; distinct sub-cellular localisations of cascade components and/or targets; feedback mechanisms; great variety of molecular signals, as well as distinct durations and strengths; cross-talks among signalling cascades that are activated simultaneously [4], [10]. All these factors contribute to the complexity of the MAPK network and presumably act together to determine signalling specificity. Deregulations of the MAPK cascades are often observed in cancer [11], [12]. Several components of the network have already been proposed as targets in cancer therapy, such as p38, JNK, ERK, MEK, RAF, RAS, and DUSP1 [12]–[23], but the intricacy of the underlying mechanisms still hinders the conception of effective drugs [24]. A deeper understanding of the regulatory mechanisms involved is crucial to clarify the roles of MAPKs in cancer onset and development, as well as to delineate therapeutic strategies. During the last decades, mathematical modelling has been widely used to study different aspects of the MAPK cascades [25] (Table S1). Quantitative models (based on Ordinary Differential Equations – ODE) have been proposed to explain the main dynamical properties of the MAPK cascades in relation with their particular structural features (double phosphorylation events, phosphatase effects, feedback loops, role of scaffold proteins, etc.) [26]–[32]. Other modelling studies investigated the behaviour of specific cascades (mainly ERK) leading to MAPK activation in response to external stimuli [33]–[42]. More recently, comprehensive qualitative dynamical models have been developed. Samaga et al. [43] built a large logical model of the signalling network (including MAPKs) responding to Epidermal Growth Factor Receptor (EGFR) stimuli, which was largely derived from the reaction map published by Oda et al. [44]. This model accurately covers the early response of the MAPK cascades to signalling stimuli, with a particular reference to primary and transformed hepatocytes. Also focusing on cancer (in particular, epithelial tumours), the logical model proposed by Poltz and Naumann recapitulates the response of a cell to DNA damaging agents (DNA repair versus apoptotic cell death), and was used to identify candidate target molecules to design novel therapies [45]. In this study, we aimed at better understanding how MAPK signalling deregulations can interfere with tissue homeostasis, leading to imbalance between cell proliferation, on the one hand, and cell growth arrest, possibly followed by apoptotic cell death, on the other hand. The choice between these phenotypes (cell fate decision) is of vital importance in cancer progression: transformed cells receive external and/or autocrine growth stimuli pushing towards cell proliferation (i. e. tumour growth); but they also receive external and/or internal anti-proliferative signals, which coupled with apoptotic stimuli trigger transformed cell clearance from the organism [46]. Our goal was to build a predictive dynamical model able (i) to recapitulate the response of the entire MAPK network to selected stimuli, along with its specific contribution to cell fate decision, and (ii) to assess novel hypotheses about poorly documented mechanisms involved in specific cancer cell types. We focused on urinary bladder cancer, where MAPK network deregulations have been associated with specific phenotypes. Bladder cancer is the fourth most common cancer among men in Europe and America. Two main types of early stage bladder carcinoma have been delineated so far: (i) non-invasive papillary carcinomas (Ta) are usually mildly aggressive and rarely progress towards higher stages, whereas (ii) carcinomas in situ (Cis) often develop into invasive tumours (T1 to T4 stages) [47]. Activating mutations of Fibroblast Growth Factor Receptor 3 (FGFR3) have been found in 70–75% of Ta tumours, but they are absent in Cis and less frequent (15–20%) in invasive tumours [47], [48]. Oncogenic FGFR3 gene fusions have also been recently identified in a small percentage of invasive bladder tumours [49]. In contrast, over-expression of EGFR has been recurrently associated with higher probability of tumour progression [50]. The mechanisms underlying the cellular response of cancer cells to these signalling stimuli are still poorly understood. Alterations of p53 and retinoblastoma (RB) pathways are presumably involved in tumour progression [51]. These pathways are major controllers of the cell cycle, and the MAPK network presumably regulates their activation by responding to growth factor stimuli. For instance, ERK phosphorylation leads to MYC activation, which can inhibit cell cycle progression through the p14/p53 pathway [52], or activate Cyclin/CDK complexes leading to RB phosphorylation and subsequent E2F-dependent gene transcription [51]. Both EGFR and FGFR3 pathways can activate the MAPK cascades. Although the two signalling pathways largely overlap, the presence of specificity factors might contribute to tune the final cellular response. To tackle these questions, we first compiled published data to build a comprehensive generic reaction map using CellDesigner software [53]–[55]. This map takes into account signals propagating from major stimuli, such as growth factors, cytokines, stress, leading to the activation of MAPKs and their downstream targets, and consequently to cell fate decision. We considered three main cell fates: proliferation, apoptosis, growth arrest. Next, we used the resulting reaction map to design a qualitative dynamical model with GINsim software [56], [57], which relies on a logical formalism [58]–[60]. To cope with the relatively high number of components, we applied a semi-automatic model reduction procedure to lower the computational cost of dynamical analyses, without losing the main dynamical properties of the system. We then performed logical simulations to check the behaviour of the model in specific documented situations, as well as to predict the behaviour in novel situations. We further investigated the role of positive and negative regulatory circuits in cell fate decision. Altogether, these analyses provided novel insights into the mechanisms differentiating the response of urinary bladder cancer cells to EGFR versus FGFR3 stimuli. We built our dynamical model using the logical formalism originally proposed by Thomas [58], [59]. Implemented in GINsim, our logical modelling approach initially requires the delineation of a regulatory graph, where each vertex (node) represents a model component and each arc (signed, directed edge) represents an interaction (activation or inhibition) between two components. Here, all components are associated with Boolean variables, meaning that they can take two possible levels, 0 or 1, denoting the absence/inactivation or the presence/activition of the modelled entities (protein activation level, gene expression level, activation of a cellular process, etc.). The model definition is completed by assigning a logical rule to each component. This logical rule specifies the target value of the component depending on the presence/absence of its regulators, using the classical Boolean operators AND, OR and NOT. The dynamical behaviour of the model can be computed starting from any initial state, step by step, updating the current state according to the logical formulae (logical simulations) [60]. To deal with large models, GINsim enables their reduction by “hiding” selected components [63]. In practice, the user selects the components to hide, and the software hides each of them iteratively, while recomputing the logical rules of their targets. Provided that no functional regulatory circuit is eliminated in the process, this reduction preserves all attractors. For example, the stable states are all conserved, in the sense that each stable state of the reduced model is the projection (on the reduced state space) of a stable state of the original model [63]. This tool is particularly useful when the high dimensionality impedes the computation of the full STG. The analysis of the paths from initial states to attractors can be done directly on the STG when it is small (tens of states), but becomes quickly intractable as the size of STG increases. To cope with this difficulty, we use a novel feature of GINsim, which iteratively clusters the state transition graph into groups of states (components or hyper-nodes) sharing the same set of successors [64]. The resulting hierarchical state transition graph (HTG) displays all the reachable attractors (components at the bottom of the graph), while the other clusters of states represent their basins of attractions (including strict basins with a single outgoing arc targeting a specific attractor, or non-strict basins grouping states that can reach a specific set of HTG components). HTG computation is done on the fly, i. e. without having to store the whole STG, which often leads to strong memory and CPU usage shrinking. Furthermore, this functionality eases the identification of the key commutations (changes of component levels) underlying irreversible choices between the different reachable attractors. Altogether, the HTG representation is very compact (often much more compact than the more classical graph of strongly connected components, as HTG further compacts linear/non circular pathways) and very informative regarding the organisation of the original STG. René Thomas proposed generic rules linking the presence of regulatory circuits (simple oriented regulatory cycles) in biological networks with dynamical properties. The first rule states that the existence of a positive circuit (involving an even number of negative regulatory interactions) is a necessary condition for multi-stationarity. The second rule states that the existence of a negative regulatory circuit (involving an odd number of negative regulatory interactions) is a necessary condition for the generation of sustained oscillations [65]. More recently, Remy et al. [66] translated these rules into theorems in the case of asynchronous Boolean dynamical systems (which is the case of our MAPK model). However, when embedded in a more complex network, specific constraints on the values of the external components acting on circuit components have to be fulfilled in order to allow a regulatory circuit to produce the expected behaviour (“circuit functionality constraints”) [67]. The concept of circuit functionality has been formalised for logical models and implemented into GINsim [62]. GINsim allows to compute all the functional positive and negative circuits of a model. For each of them, the software also provides the corresponding functionality context, defined as a set of constraints on the values of its external regulators. This tool enables the identification of the regulatory circuits playing key dynamical roles within a complex network. Based on published data, we have built and annotated a comprehensive reaction map using CellDesigner (supplementary Dataset S1). This map encompasses 232 species (proteins, genes, complexes, other molecules) and 167 reactions involved in the three most extensively documented MAPK cascades (ERK, JNK, p38). The CellDesigner version of the map (XML format) is provided as supplementary Dataset S2. The MAPK map has been further integrated into the Atlas of Cancer Signalling Networks developed by the group of Emmanuel Barillot at Institut Curie in Paris (https: //acsn. curie. fr), where it can be explored using a web browser. Our reaction map takes into account signals propagating from different major stimuli, such as growth factors, cytokines, stress, which lead to the activation of MAPKs and their downstream targets. It covers feedbacks and cross-talks among the MAPK cascades, as well as the roles of the best documented phosphatases and scaffold proteins. The main cellular compartments are also represented (plasma membrane, cytoplasm, nucleus, mitochondria, endosomes, etc.), showing the localisation of reactions within the cell. When compartmentalisation has not been fully characterised yet, reactions have been provisionally assigned to the cytoplasm. Proteins are coloured to emphasise relevant families. Figure 1 shows a map excerpt reporting two different mechanisms of ERK activation (see map annotations for more details). We considered two compartments named “Genes” and “Phenotypes” at the bottom of the map, which include representative genes activated by the MAPK cascades, as well as phenotypes (proliferation, apoptosis, growth arrest) enabled by selected readouts. We considered information concerning different human and mouse cell types, implying that the MAPK map should be considered as generic. Indeed, at this stage, information is lacking to build a detailed map based exclusively on data for a specific cell type. However, we selected biological events explicitly considered to be cell type independent. When applicable, information concerning cell types is provided through links to relevant database entries (mainly PubMed). Because the precision of the information retrieved from the literature varies, our map represents different pathways with different levels of details. For instance, we could find detailed information about the scaffold proteins intervening in the ERK cascade and on the sub-cellular localisation of the corresponding protein complexes; in contrast, such information is still largely lacking for the JNK and p38 cascades. This is why the map currently reaches its greatest level of detail for the ERK cascade. Furthermore, the level of detail represented could also be dictated by readability considerations. For instance, the RTK (receptor tyrosine kinase) component in the map represents several different receptors (e. g. EGFR, FGFR, VEGFR, etc.): all these receptors share common features that are related to MAPK activation. However, their mechanisms of action may differ in some subtle ways, which are not fundamental for our purpose here. The detailed representation of all these pathways would have made the map very difficult to read, and we thus decided to simplify the graphical representation, while keeping track of relevant variations in the annotations of the corresponding species or reactions. The resulting CellDesigner map constitutes a comprehensive and integrated source of information concerning the roles of the MAPK network in cell fate decision, taking into account specificity factors. This map can be directly used by biologists and modellers to get information about the reported phenomena. It can also be used for visualisation of high-throughput data (e. g. by automatically colouring components based on expression levels) derived from different cell conditions, for example in order to identify differentially expressed components. This can also give insights into cell type-dependent mechanisms. Hereafter, we focus on the impact of the MAPK network in urinary bladder cancer, with particular emphasis on the differential behaviour between EGFR over-expression and FGFR3 activating mutation. The logical rules assigned to model components were inferred from information about a broad range of experiments and cellular conditions. To check the coherence of the global behaviour of the resulting model with current biological knowledge, we systematically compared its dynamical properties with published data concerning different tumoural cell types, with particular attention to bladder cancer. More specifically, we first assessed the dynamical behaviour of the model under well established perturbations observed in the bladder cancer subtypes of interest. We further checked the coherence of the model behaviour with an additional list of biological facts, not necessarily involved in bladder cancer. These analyses were carried out by performing asynchronous simulations for selected initial conditions (initial states, input signals, potentially in the presence of perturbations), and observing the attractors reached by the system. In practice, the entire process from reaction map construction to model simulations is iterative, requiring several rounds of literature searches and in silico experiments. Whenever the model disagreed with established facts, we went back to the literature to seek complementary information and refined our modelling hypotheses. The reaction map and the logical model where systematically and consistently completed with relevant information during this process. Having shown that our MAPK model is consistent with published data, we designed additional simulations to explore novel mechanistic hypotheses. We have presented a bottom-up modelling approach to gain insights into the influence of the complex MAPK signalling network on cancer cell fate decision. We started by collecting pieces of information from the literature and assembling them into a detailed reaction map, which served as source of information for further dynamical modelling. The resulting map is generic, meaning that it was drawn by using information coming from different experimental models. Based on specific biological questions, our dynamical logical modelling involved the abstraction of relevant information from the map and the drawing of a qualitative influence network (regulatory graph). Next, we assigned consistent logical rules to each component to enable logical simulations. In order to perform detailed analyses at reasonable computational costs, we derived reduced model versions preserving the main dynamical properties of the original model. The reduced versions can be considered as further abstractions of the MAPK network, explaining its qualitative behaviour in terms of selected molecular actors. Despite the fact that we made no use of quantitative data, and that we finally represented an extremely complex signalling network through a limited number of Boolean components, we were able to recapitulate its behaviour for diverse documented biological conditions. These results set the background to investigate the roles of poorly documented regulatory mechanisms. In this modelling study, we particularly focused on bladder cancer. Importantly, our simulations qualitatively recapitulated salient phenotypic differences observed in invasive versus non-invasive carcinomas, and allowed us to formulate reasonable hypotheses concerning the mechanisms determining such differences. These hypotheses are readily amenable to experimental validation. Our MAPK network model can be considered as a module for the assembly of more comprehensive cancer-related network. From this point of view, it will be interesting to merge our model with other logical models implementing other aspects of cell fate decision, in particular the model proposed by Calzone et al. [78], which covers the interplays between NFκB pro-survival pathway, RIP1-dependent necrosis, and extrinsic/intrinsic apoptosis pathways. In the Introduction, we highlighted the importance of specificity factors in determining signal specificity of the MAPK network and took this into consideration in the construction of the reaction map. However, given the heterogeneity of available information among the different MAPK cascades, we could not include all these factors in our logical model. Nonetheless, we considered some of them, including several feedbacks, cross-talks, phosphatases and input stimuli. These allowed us to focus on interesting aspects and identify mechanisms potentially underlying the different responses of bladder cancer cells to different growth factor receptors (EGFR versus FGFR3). The role of SPRY-dependent down-regulation of FGFR3 signalling seems to be determinant for the decision between proliferative and non-proliferative response. Moreover, the model also suggests that the presence of PI3K/AKT, but not ERK, positively correlates with the presence of a proliferative phenotype. Nevertheless, ERK-related mechanisms (fastness/strength of ERK activation and activation of SPRY) seem to be determinant for driving the switch. Such different responses provide a striking example of how signals transduced by largely overlapping pathways can produce opposite effects. To explain this behaviour, we analysed the roles of specific model circuits, which are presumably extremely important in the phenotype choice. Our data further highlight the contribution of cross-talks among the MAPK cascades to cell fate decision. Other specificity factors, including scaffold proteins and sub-cellular localisation, should also be taken into consideration in the near future, as novel data on these factors will be gathered. This will require a regular updating of our MAPK reaction map, by including new findings related to cell fate decision. We interpreted the p53-independent response of the MAPK network to FGFR3 stimulus as a sort of balance between proliferative and non-proliferative phenotype. A decreased rate of cell proliferation might indeed explain the less aggressive phenotypes frequently observed in FGFR3-mutated bladder carcinomas, in comparison with EGFR over-expression cases. Interestingly, this behaviour can be further related with opposite effects of FGFR3 activation in other cell types. In particular, activating FGFR3 mutations have been associated with growth arrest in chondrocytes, whereas they enhance proliferation and/or transformation in several cancer types and skin disorders (e. g. bladder cancer, multiple myeloma, seborrheic keratosis, etc.) [79]. Tentatively, proper modifications (e. g. concerning the introduction of STAT-dependent pathways and tuning of AKT response to growth factors [80]) may enable our MAPK model to account for these observations. Finally, we are currently assessing a potential proliferative role of p38 in FGFR3-mutated bladder carcinomas (unpublished preliminary data), which might lead to further model refinement. To wrap up, the present study demonstrates how Boolean modelling can recapitulate salient dynamical properties of an extremely complex biological network. As further details are uncovered, our logical model could be refined and eventually translated into a more quantitative framework (e. g. using ODEs or stochastic equations). In a first step towards more quantitative simulations, a continuous time Boolean framework could be used to explicitly represent time dependencies [81]. Tentatively, this approach would allow us to recapitulate more precisely the differential effects of transient versus sustained ERK activation [33], [37], [82]. Combining the delineation of a detailed reaction map and that of a predictive logical model, this study can serve as a basis to develop (semi-) automatic tools to derive logical models from reaction maps. Indeed, the manual derivation of a logical model from a complex reaction map presents risks of misinterpretations of either map symbols or map annotations. Errors are particularly likely to happen when the model is not built by the author of the map. In this respect, recent rule-based languages used to derive more quantitative models could be used to systematically derive predictive logical models, although potentially at the cost of additional efforts to build reaction maps in a more rigorous fashion [83]–[85].

Depending on environmental conditions, strongly intertwined cellular signalling pathways are activated, involving activation/inactivation of proteins and genes in response to external and/or internal stimuli. Alterations of some components of these pathways can lead to wrong cell behaviours. For instance, cancer-related deregulations lead to high proliferation of malignant cells enabling sustained tumour growth. Understanding the precise mechanisms underlying these pathways is necessary to delineate efficient therapeutical approaches for each specific tumour type. We particularly focused on the Mitogen-Activated Protein Kinase (MAPK) signalling network, whose involvement in cancer is well established, although the precise conditions leading to its positive or negative influence on cell proliferation are still poorly understood. We tackled this problem by first collecting sparse published biological information into a comprehensive map describing the MAPK network in terms of stylised chemical reactions. This information source was then used to build a dynamical Boolean model recapitulating network responses to characteristic stimuli observed in selected bladder cancers. Systematic model simulations further allowed us to link specific network components and interactions with proliferative/anti-proliferative cell responses.

Abstract Introduction Methods Results Discussion

2013

Integrative Modelling of the Influence of MAPK Network on Cancer Cell Fate Decision

During meiotic prophase, a structure called the synaptonemal complex (SC) assembles at the interface between aligned pairs of homologous chromosomes, and crossover recombination events occur between their DNA molecules. Here we investigate the inter-relationships between these two hallmark features of the meiotic program in the nematode C. elegans, revealing dynamic properties of the SC that are modulated by recombination. We demonstrate that the SC incorporates new subunits and switches from a more highly dynamic/labile state to a more stable state as germ cells progress through the pachytene stage of meiotic prophase. We further show that the more dynamic state of the SC is prolonged in mutants where meiotic recombination is impaired. Moreover, in meiotic mutants where recombination intermediates are present in limiting numbers, SC central region subunits become preferentially stabilized on the subset of chromosome pairs that harbor a site where pro-crossover factors COSA-1 and MutSγ are concentrated. Polo-like kinase PLK-2 becomes preferentially localized to the SCs of chromosome pairs harboring recombination sites prior to the enrichment of SC central region proteins on such chromosomes, and PLK-2 is required for this enrichment to occur. Further, late pachytene nuclei in a plk-2 mutant exhibit the more highly dynamic SC state. Together our data demonstrate that crossover recombination events elicit chromosome-autonomous stabilizing effects on the SC and implicate PLK-2 in this process. We discuss how this recombination-triggered modulation of SC state might contribute to regulatory mechanisms that operate during meiosis to ensure the formation of crossovers while at the same time limiting their numbers. Sexual reproduction depends on the specialized cell division program of meiosis, which allows diploid organisms to form haploid gametes. Reduction in chromosome number from the diploid to the haploid state occurs during the first meiotic division, when homologous chromosomes segregate to opposite poles of the meiosis I spindle. Reliable segregation of homologs in turn depends on creation of temporary attachments between the homologs during an extended prophase that precedes the meiotic divisions. In most organisms, meiotic prophase is characterized by two hallmark features: 1) assembly of a highly-ordered structure known as the synaptonemal complex (SC) between the aligned homologs, and 2) formation of crossover (CO) recombination events between their DNA molecules [1]. The occurrence of crossing over between homologs during meiosis was recognized more the 100 years ago [2], long before DNA was identified as the genetic material. We now know that COs are the products of a meiosis-specific recombination program, consisting of i) the controlled introduction of DNA double strand breaks (DSBs) by the transesterase SPO-11 [3–5], ii) engagement of the homologous chromosome as a template for recombinational repair [6], and iii) formation and resolution of intermediates at a subset of recombination sites in a manner that yields CO products. These inter-homolog COs, in conjunction with sister chromatid cohesion, form the basis of connections that allow the homologs to orient and segregate toward opposite spindle poles at meiosis I. Meiotic CO formation is promoted by conserved meiosis-specific factors that include the MutSγ (MSH4-MSH5) complex [7–10], and in animals, the cyclin-related protein COSA-1/CNTD1 [11,12]. Further, meiotic recombination is tightly regulated in a manner that limits that number of COs formed, yet simultaneously guarantees that every homolog pair receives at least one CO. The SC has also long been recognized as a canonical feature of the meiotic program [13,14]. In EM images, and more recently through use of super-resolution fluorescence microscopy, the SC is observed as a tripartite structure assembled at the interface between pairwise aligned homologous chromosomes [15–18]. Axial structures that form along the length of each homolog comprise the two lateral elements of the SC, paralleling each other at an approximate distance of 100–200 nm. These lateral elements are linked together by the SC central region, which contains an ordered array of transverse filaments that spans the distance between the two axes, resulting in a zipper- or railroad track-like appearance of the SC in EM images. Meiotic chromosome axes in most organisms contain meiosis-specific cohesin complexes and/or meiotic HORMA-domain-containing proteins (however, the numbers of parologous complexes present in different organisms is highly variable and their primary sequences are substantially diverged). SC central region proteins from diverged organisms cannot necessarily be recognized as true homologs, but typically contain predicted coiled-coil domains and can collectively bridge the distance between the homolog axes [19]. CO recombination events are completed in the context of assembled SCs, yet we are only beginning to understand the complex interrelationships between the SC and COs. One conserved function of SC central region proteins is to promote normal levels of COs. However, organisms vary widely in the degree to which meiotic CO formation depends on SC central region proteins [20–23]. Further, the pro-CO function (s) of these proteins may be exerted locally at CO sites, not necessarily requiring assembly of extended stretches of SC [24,25]. Conversely, SC central region proteins have also been implicated in antagonizing the formation of excess COs. This was first demonstrated in C. elegans, where essentially all COs are dependent on its SC central region proteins (SYP-1-4) and COs are normally limited to one per homolog pair [22,26–28]; whereas lack of SYP proteins eliminates interhomolog COs, altering SC composition by partial depletion of SYP proteins increases the number of COs and attenuates CO interference [29,30]. Recent work suggests that the SC central region may also play a role in limiting CO formation in S. cerevisiae, as CO numbers are similarly increased in mutants where the transverse filament protein Zip1p is present at recombination sites but cannot localize along the full length of a bivalent due to either an altered N-terminus or lack of other central region components [24]. The observation that SC central region proteins can both promote and limit COs suggests the possibility that formation of CO intermediates would lead to a change in some assayable propert (ies) of the SC. This proposition would be difficult to test in experimental systems such as mouse or yeast, where assembly of the SC is coupled to and dependent on the formation of recombination intermediates. However, assembly of full length SCs between paired homologs in C. elegans is not dependent on recombination [26,31], making this an ideal experimental system for investigating how formation of CO recombination intermediates might alter the state of the SC. In this work, we investigate how SCs change during meiotic prophase progression in C. elegans. Our findings contribute to a growing body of evidence that the SC is a much more dynamic structure than is suggested by its highly-ordered appearance in EM images[32–35]. We show that the SC incorporates new subunits and switches from a more dynamic to a more stable state during progression from the early pachytene to late pachytene stage. Moreover, we demonstrate that meiotic recombination events can indeed alter the state of the SC in a chromosome-autonomous manner. The pachytene stage of meiotic prophase is defined by the presence of synaptonemal complex along the full interface between lengthwise aligned pairs of homologous chromosomes. However, composite fluorescence images of C. elegans gonads generated using an automated stitching algorithm ([36] which preserves relative fluorescence intensities of different fluorophores across a set of tiled images) reveal that nuclei that are in the pachytene stage based on this definition change in their appearance during the time that they spend in this stage. In immunofluorescence (IF) images of whole-mount gonads (Fig 1A), the IF signal intensity for SC central region protein SYP-1 gradually increases while the signal for chromosome axis protein HTP-3 remains relatively constant as nuclei progress through the pachytene stage. Further, in live worms expressing an EmGFP-tagged version of SC central region protein SYP-3 (Fig 1B), the EmGFP: : SYP-3 fluorescence intensity in germ cell nuclei increases approximately twofold (2. 1 ± 0. 3,2 gonads analyzed) over the course of pachytene progression. A similar increase in intensity also occurs in mutants lacking meiotic recombination (2. 2 ± 0. 1,3 gonads analyzed; S1 Fig) and presumably reflects the ongoing addition of SC central region subunits to existing “full length” SCs. Together, these observations indicate that the relative subunit composition and/or structure of SCs are not constant throughout the pachytene stage, but instead vary during pachytene progression. Paralleling our observations in live worms and whole-mount dissected gonads, we also find that early and late pachytene nuclei differ in the sensitivity of their SCs to experimental perturbation by detergent-based lysis and partial spreading procedures (S2–S4 Figs). These observations reinforce the conclusion that the SC changes during progression through the pachytene stage of meiotic prophase. We investigated the potential dynamic nature of the SC central region using Fluorescence Recovery After Photobleaching (FRAP) in live worms expressing GFP-tagged versions of SC central region component SYP-3 (Fig 2). We used two distinct experimental systems and two independently-generated GFP: : SYP-3 transgenic lines. For both transgenes, experiments were conducted using strains that also expressed untagged SYP-3 from the endogenous locus, as neither transgene fully rescued a syp-3 (null) mutant (see Materials and methods). Both of these approaches led to the same conclusions, namely that the SC is indeed a dynamic structure, and that the SC central region changes from a more dynamic to a less dynamic state during meiotic progression. The experiments depicted in Fig 2A–2C were carried out using a Leica SP2 confocal microscope in which the FRAP module was configured to perform a Z-stack bleach. In these experiments, a region corresponding to roughly 30–50% of each selected nucleus was subjected to stepwise photobleaching throughout a Z-stack (0. 24 μm step size) encompassing the entire depth of the nucleus, and images were acquired at multiple subsequent time points (Fig 2A). For each nucleus analyzed, the extent of recovery at post-bleach time points was quantified by calculating the ratio of fluorescence signal within a Region of Interest (ROI) confined to the bleached portion of the nucleus relative to the fluorescence signal for that whole nucleus, normalized to the same ratio observed prior to the bleach (Fig 2C, S5A Fig; Materials and methods). Analyzed nuclei were grouped into “early”, “mid” and “late” pachytene stages based on their relative positions along the distal/proximal axis of the gonad. Data plotted in Fig 2C show that substantial recovery of fluorescence occurred, predominantly over a period of 30 minutes, with the majority of recovery achieved during the first 15–20 minutes. Moreover, extent of recovery was strongly correlated with the relative position of the nucleus within the pachytene region of the germ line (R2 = 0. 71; S5A Fig), with the median recovery level observed for late pachytene nuclei being 51% lower than that observed for early pachytene nuclei (p <0. 0001; Mann Whitney test). These data indicate that the SC central region becomes less dynamic as nuclei progress through the pachytene stage. For Fig 2D–2G, experiments were carried out using a Deltavision OMX Blaze wide-field deconvolution microscopy system, in which photobleaching is focused at a single selected focal plane (rather than bleaching stepwise throughout a Z-stack; see Materials and methods). The images presented are partial projections showing the half of the nucleus containing the bleached portions of the SCs. For the experiments depicted in Fig 2D and 2E, small ROIs (1–3 μm2) were deliberately selected in order to bleach limited segments of the individual SCs within the nuclei; this approach allowed assessment of recovery in a context where the majority of the initial fluorescence in a nucleus was retained and potential photo-damage was minimized. Quantitation of these experiments (using a scoring system devised to accommodate movement of nuclei and chromosomes; see Materials and methods) demonstrated that recovery was strongest in early pachytene nuclei, then diminished as nuclei progressed to the late pachytene stage (Fig 2E). Thus, FRAP analyses conducted using two very different experimental set-ups yielded the same findings, together providing strong support for the conclusion that SC transitions from a more highly dynamic state to a less dynamic state during pachytene progression. Two additional conclusions can be made regarding the nature of the recovery observed in our experiments. First, several lines of evidence indicate that recovery occurs predominantly by redistribution of existing GFP: : SYP-3 protein within a nucleus. Recovery was minimal in fully bleached nuclei, indicating that import of new protein from the cytoplasm is insufficient to account for the recovery observed during the time frame of the experiments (Fig 2A). (This is consistent with the that fact that the total amount of new fluorescent protein per nucleus is estimated to increase by less than 5% during the duration of a FRAP experiment, based on the observation that fluorescence increases by only two-fold during the entire course of the pachytene stage; Fig 1B.). Further, in our confocal experiments (where 30–60% of the initial fluorescence was eliminated by bleaching) the normalized total fluorescence within such a partially-bleached nucleus (relative to a neighboring unbleached reference nucleus) remained almost constant during the recovery time course (S5B Fig). Finally, in experiments using large ROIs, the unbleached portions of SCs within partially-bleached nuclei diminished in fluorescence relative to unbleached neighboring nuclei in the same field (Fig 2G). A second conclusion can be drawn from images of nuclei in which an individual SC was bleached throughout its entire length (e. g. Fig 2B and 2F, S5C Fig). Such SCs can nevertheless exhibit substantial recovery of fluorescence. This observation indicates that while recovery likely involves redistribution of subunits within individual SCs, it also involves redistribution of subunits among the different SCs within a nucleus, presumably through exchange with a nucleoplasmic protein pool. The observed transition in the dynamic state of the SC occurs in parallel with the process of meiotic recombination. Further, previous studies have revealed that cytological features characteristic of early pachytene are prolonged in mutants that are impaired in meiotic recombination [37–39]. Thus, we conducted FRAP experiment to evaluate the dynamic state of the SC in spo-11 mutants, which are defective in forming the DSBs that serve as the initiating events of recombination [31], and in cosa-1 mutants, which are defective in converting resected DSBs into interhomolog crossovers (COs) [12] (Fig 3). Both the wide-field small ROI FRAP assay (Fig 3A and 3C) and the confocal Z-stack bleach FRAP assay (Fig 3B) yielded the same overall findings. First, the substantial decline in recovery observed during progression from early to late pachytene in the wild type was not observed in spo-11 and cosa-1 mutants (albeit a modest 9% decline was detected in the confocal assay for the cosa-1 mutant). Second, the degree of fluorescence recovery in late pachytene nuclei in both the spo-11 and cosa-1 mutants was significantly higher than in wild type. Thus, an inability to form CO recombination intermediates is associated with persistence of a more dynamic state of the SC. This suggests that formation of CO recombination intermediates may serve as a trigger for the SC to switch to a less dynamic state. In the course of analyzing immunofluorescence images of whole-mount gonads from spo-11 mutant worms, which lack the SPO-11 enzyme responsible for making meiotic DSBs, we noticed a difference compared to wild-type in the appearance of the SC immunostaining in the second half of the pachytene region (Fig 4): whereas SYP-1 immunostaining appeared uniformly distributed among the individual SCs in most wild-type nuclei, a subset of nuclei in spo-11 mutant worms (17 ± 7% for spo-11 (me44), 20 ± 3% for spo-11 (ok79) ) exhibited an uneven distribution of SYP-1 among the SCs, and within the last quarter of the pachytene region, it was clear that SYP-1 staining was preferentially enriched on a single SC relative to the other SCs in that same nucleus. Late pachytene nuclei with uneven SYP-1 distribution were likewise observed in the dsb-1 mutant (20 ± 3%), which is also impaired in initiation of meiotic recombination [38], indicating that the presence of such nuclei is a shared property of mutants that are proficient for synapsis but defective in the formation of SPO-11-dependent DSBs. Both the incidence of late pachytene nuclei exhibiting uneven SYP-1 distribution (p<0. 0001, Fisher’s exact test) and the number of SYP-1-enriched SCs within such nuclei (p<0. 0001, Mann Whitney test) increased following low dose IR treatment of the spo-11 mutant, suggesting that in this context, DNA breaks can trigger enrichment of SYP-1 proteins on a subset of chromosomes (S6 Fig). Further, the number of SYP-1-enriched SCs in late pachytene nuclei was significantly lower (p<0. 0001, Mann Whitney test) in IR-treated spo-11 worms heterozygous for a reciprocal translocation (spo-11; szT1 (I; X) /+), in which approximately 33% of the genome is engaged in heterosynapsis, than in IR-treated spo-11 worms (S6 Fig). This further suggests that the ability of IR-induced DNA breaks to trigger SYP enrichment depends on the ability of such breaks to engage in recombination-based interactions with homologous DNA sequences. Our finding that IR-induced DNA breaks produced in limiting numbers can elicit uneven distribution of SYP-1 within nuclei seems at odds with the fact that nuclei with uneven SYP-1 distribution occur even in the absence of irradiation in the meiotic-DSB-defective mutants. These results can be reconciled by hypothesizing that spontaneous SPO-11-independent DNA lesions (of unknown structure) occur at a low frequency in mutants lacking meiotic DSBs. While such spontaneous DNA lesions do not appear to yield inter-homolog COs [31], they are nevertheless capable of recruiting meiotic DNA repair proteins when the normal preferred substrates for these proteins are absent (see below). Two lines of evidence support the conclusion that the uneven distribution of SYP-1 protein within nuclei in the context of limiting DNA breaks reflects the formation of recombination/repair intermediates that recruit meiotic CO factors. First, the presence of nuclei exhibiting an uneven distribution of SYP-1 requires the presence pro-CO factors COSA-1 and MutSγ (MSH-4–MSH-5) (Fig 4), as such nuclei are not detected in cosa-1 or msh-5 single mutants, which are proficient for DSB formation but defective in repairing DSBs as COs. Further, such nuclei are not detected in spo-11; cosa-1 double mutants, indicating that COSA-1 is required to trigger the uneven SYP-1 distribution observed in a spo-11 mutant background. Second, when DSBs are limiting, SYP-1 is preferentially concentrated on the subset of SCs harboring a focus where pro-CO factors are concentrated (Fig 5A–5D). Late pachytene nuclei in wild-type worms contain 6 bright GFP: : COSA-1 foci marking the site of the single CO on each homolog pair [12]. Most late pachytene nuclei in spo-11 worms expressing GFP: : COSA-1 completely lack COSA-1 foci, consistent with an absence of DSBs and CO recombination intermediates [12]. However, a subset of late pachytene nuclei in these worms harbor a single COSA-1 focus, suggesting the infrequent occurrence of a SPO-11-independent DNA lesion that is capable of recruiting meiotic recombination factors during the late pachytene stage. Further, whereas late pachytene nuclei lacking COSA-1 foci exhibited a uniform distribution of SYP-1, nuclei with a COSA-1 focus exhibited uneven SYP-1 distribution, with SYP-1 becoming preferentially enriched on the SC harboring the COSA-1 focus (Fig 5A). Experiments using MSH-5 as a recombination marker gave identical results (Fig 5C). These experiments further showed that when a single SYP-1-enriched SC harboring a recombination focus is detected, it can be associated either with the X chromosomes or with an autosome pair (Fig 5C). However, we note that association with the X chromosomes is disproportionate (34% observed vs. 17% expected; n = 38, p = 0. 0018), perhaps reflecting known differences in replication timing [40] and/or chromatin structure [41] between the X and autosomes that might affect the incidence of the triggering DNA lesions. Analysis of late pachytene nuclei in a dsb-2 mutant, in which SPO-11-dependent DSBs are substantially reduced but not eliminated [37], strongly corroborated the correspondence between the presence of COSA-1 foci in late pachytene nuclei and uneven distribution of SYP proteins (Fig 5B and 5D). Likewise, within such nuclei, SYP-1 was enriched on the subset of SCs harboring the COSA-1 foci. Taken together, our data indicate that when DSBs are limiting, concentration of CO proteins at a putative DNA repair site triggers a change in status of SC components, resulting in enrichment of SYP proteins in cis on the chromosome harboring such an event. We conclude that COSA-1-marked recombination intermediates, and/or concentration of pro-CO factors at such sites, cause chromosome-autonomous stabilization of the SCs. To better understand how formation of a COSA-1 / MutSγ dependent recombination intermediate (and/or recruitment of these pro-CO factors) triggers a change in state of the SC with which it is associated, we investigated the potential involvement of polo-like kinase PLK-2. Whereas PLK-2 is initially concentrated at pairing centers (nuclear envelope-associated chromosome sites that coordinate chromosome movement, homolog recognition and SC assembly) during early prophase, it exhibits dynamic localization during meiosis and is subsequently detected on the SCs later during the pachytene stage [42,43]. We first re-examined PLK-2 localization during meiotic prophase progression in germ cell nuclei from otherwise wild-type worms expressing PLK-2: : HA from the endogenous plk-2 locus, prepared using mild detergent-based lysis conditions (Fig 6, top). Our images refine the previously reported dynamic localization pattern of PLK-2, showing in addition that: 1) localization of PLK-2 on SCs can be detected very early in pachytene, and 2) preferential concentration of PLK-2 on limited SC subdomains (defined by recombination sites) can be detected right after the early-to-late pachytene transition, substantially earlier than the SYP proteins exhibit a similar pattern of CO-triggered relocalization. This suggests that PLK-2 localization during pachytene progression may not simply track with the SYP proteins, but may predict the behavior of SYP proteins during late pachytene. Analysis of PLK-2 localization in the dsb-2 mutant clearly shows that PLK-2 becomes highly enriched on the subset of chromosomes harboring a MutSγ-marked recombination site (Fig 6, bottom). Moreover, strong preferential localization of PLK-2 on MutSγ-marked chromosomes is clearly detected much earlier than preferential concentration of SYP-1 is detected on such chromosomes. These findings are consistent with a model in which COSA-1/ MutSγ- marked recombination sites trigger relocalization of PLK-2, which in turn affects the state of the SC. To complement these localization experiments showing that PLK-2 is present at the right time and place to mediate changes in SC state, we conducted experiments using the null mutation plk-2 (ok1336) to test the impact of loss of PLK-2 function on the state of the SC during the late pachytene stage (Fig 7). The early meiotic roles of PLK-2 in promoting homolog pairing and synapsis are substantially supplanted in the plk-2 (ok1336) mutant by its close paralog PLK-1; thus, while SC assembly is delayed in this mutant, most chromosome regions are synapsed by late pachytene and most chromosome pairs recruit COSA-1 foci and form chiasmata [42,43] (Fig 7C). Fig 7A and 7B show the results of FRAP experiments comparing GFP: : SYP-3 dynamics in late pachytene nuclei of wild-type and plk-2 mutant worms. Strong fluorescence recovery was consistently detected in all late pachytene nuclei in the plk-2 mutant background, reflecting a more highly dynamic state of the SC central region in the plk-2 mutant than in wild-type controls. Similarly, immunofluorescence analysis of late pachytene nuclei in the plk-2; dsb-2 double mutant revealed that in contrast to the dsb-2 single mutant (Fig 5), SYP-1 does not become preferentially concentrated on the subset of chromosomes that harbor a COSA-1-marked recombination site (Fig 7C). Together, our data show that 1) PLK-2 becomes localized on chromosomes harboring MutSγ-marked recombination intermediates in a manner that precedes and predicts the later localization of SYP proteins, and 2) PLK-2 is required to elicit changes in state of the SC that are normally triggered by such intermediates. We have presented two independent lines of evidence that converge on the conclusion that formation of CO-eligible recombination intermediates triggers a change in state of the SC. First, our FRAP analyses revealed that a more highly dynamic status of the SC central region characteristic of early pachynema in wild-type meiosis is retained in late pachytene nuclei both in the spo-11 mutant, which fails to initiate recombination [31], and in the cosa-1 mutant, which fails to accumulate pro-CO factors at recombination sites and can’t form COs [12]. These data indicate that proper execution of meiotic recombination is responsible for at least part of the change in dynamics observed during prophase progression. Second, whereas FRAP analysis indicated a link between recombination and SC stabilization, IF analyses of SCs under conditions where meiotic DSBs are limiting helped to clarify this relationship. These IF experiments revealed that when chromosomes with and without a COSA-1-marked (or MutSγ –marked) recombination intermediate co-exist within the same nucleus, SYP proteins become preferentially associated with the chromosome pair (s) that harbor such a recombination intermediate, apparently at the expense of non-recombinant chromosomes. In light of our observations in the FRAP analyses, this latter set of findings supports the view that the effects of recombination on SYP dynamics are not nucleus-wide but are instead exerted at the level of individual chromosome pairs; i. e. , we infer that CO recombination events alter the SC central region in chromosome-autonomous manner. Together our data allow us to conclude that SC central region stabilization is triggered either by a recombination intermediate that is created and/or stabilized by MSH-5 and COSA-1, or by the physical presence of these pro-CO proteins at recombination sites. We further reasoned that homologous engagement is necessary, but not sufficient, for these changes in the properties of the SC: The need for homolog engagement can be inferred from our experiments showing that the incidence of SYP-enriched chromosomes detected following induction of DNA breaks by IR in spo-11 mutant meiocytes is significantly reduced when one-third of the genome is engaged in heterologous (rather than homologous) synapsis. Conversely, the insufficiency of homolog engagement is implied by the fact that pro-CO factors are required for SC central region stabilization despite the fact that msh-5 mutants are competent to engage in homologous recombination to repair DSBs induced during meiosis as inter-homolog non-COs [48]. How can a nascent CO event at one site on a chromosome pair trigger a chromosome-wide change in the properties of the SC? Our experiments implicate the polo-like kinase PLK-2 as a key player in this process. Polo-like kinases are known to play multiple roles during mitosis and meiosis. For example, the POLO homolog Cdc5 is required both to resolve meiotic CO intermediates and to disassemble the SC at pachytene exit during budding yeast meiosis [49]. In C. elegans, previous work has implicated PLK-2 in multiple meiotic functions, including: 1) promoting early prophase chromosome movements coupled to homolog pairing and SC assembly, 2) checkpoint responses to unsynapsed chromosomes, and 3) normal restructuring of bivalents during late meiotic prophase [42,43]. Based on our current data, we propose a new role, in coupling formation of CO-eligible recombination intermediates to stabilization of the SC central region. Our thinking is shaped by integration of findings from functional assays and from PLK-2 protein localization. FRAP analysis showed that late pachytene nuclei in a plk-2 mutant exhibit a more highly dynamic SC state than in wild type germ lines, and IF analysis showed that PLK-2 is required to achieve preferential association of SYP-1 on chromosomes harboring CO sites (when recombination intermediates are present in limiting numbers). On their own, these findings are consistent either with a) PLK-2 being required for the SC central region to mature to a state where CO intermediates are capable of triggering a change in SC state, or with b) PLK-2 playing a role in eliciting the state change. Our analysis of PLK-2 localization under conditions where DSBs (and consequent CO intermediates) are limiting leads us to favor the latter interpretation. Concentration of PLK-2 along the length of chromosomes harboring CO sites is detected prior to the eventual enrichment of SYP proteins on such chromosomes. This observation, together with the requirement for PLK-2 to achieve this preferential SYP association, are best explained by a model in which nascent CO sites serve to recruit PLK-2 to the SCs that harbor them. We hypothesize that PLK-2 localization and activity then spread along the lengths of these SCs via a self-reinforcing feedback loop that in turn results in stabilization of the SC central region along the full length of the SC. This model is reinforced by the recent finding of S. Nadarajan and M. Colaiacovo [33] that SC central region protein SYP-4 is phosphorylated in a PLK-1/2-dependent manner. What function (s) might be served by coupling formation of CO-eligible recombination intermediates to a change in state of the SC? We can envision several possible mechanisms by which such a coupling could help to promote a robust outcome of meiosis: First, we propose that CO-intermediate-triggered alteration of SCs may play a key role in a regulatory/surveillance network that regulates meiotic progression and promotes CO assurance [37–39]. This surveillance network monitors the formation of CO-eligible recombination intermediates and makes timely progression from early pachytene to late pachytene contingent on their occurrence. Progression to late pachytene is characterized by a major coordinated transition affecting multiple distinct aspects of the meiotic program, including cessation of DSB formation, maturation of CO sites, and loss of the ability to engage the homolog as a template for DNA repair. A key feature of this surveillance network is that germ cells can detect a single chromosome pair lacking a CO intermediate, prolonging the early pachytene stage in response. However, it was difficult to envision how a chromosome pair lacking a relevant recombination intermediate might be detected. Based on our current findings, we propose that the state of the SC may in fact be the monitored feature, and that CO-triggered modulation of SC state could serve as a means to extinguish a signal that maintains the early pachytene stage. According to this model, formation of the required CO-eligible intermediates on all chromosome pairs would lead to all six SCs becoming stabilized, which would in turn extinguish the “maintain early pachytene” signal and thereby enable nuclei to progress to the late pachytene stage. The model proposed here is distinct in several ways from the model recently proposed by Machinova and colleagues [32]. Machinova et al. similarly found that under conditions where DSBs are limiting, SC central region proteins are preferentially retained or stabilized on chromosomes undergoing recombination events. Further, using genetic analyses complementary to those reported here, they likewise implicated COSA-1, MutSγ and PLK-2 in this process. In their model, Machinova et al. asserted that nuclei respond to detection of synapsed chromosome pairs that lack CO intermediates by actively triggering desynapsis/destabilization of the SC in late pachytene to provide another chance to form COs. Given our demonstration that the SC central region exhibits a more highly dynamic state in early pachytene that is shut down during progression to late pachytene, we argue that it is not necessary to invoke active desynapsis to explain the data. Rather, we suggest that the reduction of SYP proteins on late pachytene bivalents lacking CO sites may be a secondary, pathological, consequence of SYPs becoming actively stabilized on CO-associated bivalents in the same nucleus, which could result in redistribution of nucleoplasmic subunits and lowering of local protein concentrations below a threshold needed to reliably maintain SYP association on bivalents where stabilization has not occurred. Further, our analysis of SYP dynamics can likewise explain the even distribution of SYP proteins in nuclei where all chromosomes lack recombination sites. To reconcile this observation with a model wherein chromosomes lacking CO intermediates actively trigger desynapsis, Machinova et al. postulated that DSBs are required to activate the proposed surveillance mechanism. However, the observation is readily explained (without the need to invoke DSB-dependent activation of the surveillance system) by persistence of the more highly dynamic early pachytene state for all SCs, which would result in the SYP proteins remaining well distributed among them despite the fact that the SCs are not stabilized. Second, we propose that CO-intermediate-triggered SC central region stabilization may be an integral feature of the CO regulation (aka “CO control”) system that operates to promote and ensure the formation of COs while at the same time limiting their numbers. Prior work showed that the C. elegans SYP proteins function both in promoting CO formation and inhibiting excess COs [22,29,30], leading Libuda et al. to propose that meiotic CO regulation may function as a self-limiting system in which the SC or its components initially create an environment that promotes the formation of CO-eligible recombination intermediates, which in turn triggers a change in the state of the SC that antagonizes the formation of additional COs. We have now demonstrated that formation of CO intermediates is indeed coupled to a stabilization of the SC central region. We speculate that the stabilized SC central region may help both to a) facilitate correct maturation of CO intermediates at the sites that had been selected to become COs, and b) create a barrier to additional CO, e. g. by excluding DNA ends from engaging the homolog as a repair template and/or by promoting eviction of pro-CO proteins from recombination sites not selected to become COs. Finally, we suggest that CO-intermediate-triggered SC stabilization likely plays a role in solidifying pairwise partnerships between homologous chromosomes. While the SC is required to stabilize associations between homologs during C. elegans meiosis [22], it is well established that formation of SCs between homologs in C. elegans does not depend on recombination [31]. Indeed, the Caenorhabditis genus has lost an entire functional module of genes encoding conserved proteins needed to promote inter-homolog DNA strand invasion (DMC1/MND1/HOP2) in organisms where SC assembly is recombination-dependent [50]. However, our recent work using polyploidy as a means to perturb the meiotic program revealed that recombination can nevertheless contribute to formation of stable pairwise interactions between homologs under conditions where three partners compete for synapsis [47]. Our findings led us to propose a two-phase model for C. elegans synapsis involving a) an early phase in which synapsis interactions are driven by a recombination-independent homology assessment mechanism, and b) a late phase in which recombination promotes mature synapsis. The current work strongly reinforces this view. Although the SCs formed in the absence of recombination may appear “normal” [31], the assays used in the current work reveal that they do in fact differ from normal SCs. We speculate that the early, recombination-independent form of the C. elegans SC may function to mediate/maintain close homolog juxtaposition to facilitate productive engagement of the homologous chromosome by processed DSBs, whereas the late CO-intermediate-stabilized form may be analogous to the recombination-dependent SCs of budding yeast, Arabidopsis, and mammals. Unless otherwise specified, worms were cultured at 20°C under standard conditions [51]. The following C. elegans strains were used: A strain expressing PLK-2 with a C-terminal HA tag was generated by CRISPR/CAS9 genome engineering of the endogenous plk-2 locus using the protocol of [52]. A BamHI site was introduced as a spacer/screening sequence between the HA-tag and the last amino acid of PLK-2, resulting in following modification at the plk-2 locus: (plk-2 to codon 632) GGATCCTACCCATACGACGTCCCAGACTACGCCTAA (plk-2 3’UTR). All analyses were performed on 20-24h post-L4 adults. For immunofluorescence experiments on whole-mount gonads, dissection of gonads, fixation, immunostaining and DAPI counterstaining were performed as in [53]. The following primary antibodies were used: chicken anti-HTP-3 (1: 250 [46]), rabbit anti-GFP [12]), guinea-pig anti-SYP-1 (1: 200 [22], rabbit anti-MSH-5 [12]). Secondary antibodies were Alexa Fluor 488,555 and 647-conjugated goat antibodies directed against the appropriate species (1: 400, Life Technologies). Images in Figs 1,4, 5 and 7C were collected as Z-stacks (at 0. 2μm intervals) using a 100x NA 1. 40 objective on a DeltaVison OMX Blaze microscopy system (except for the 7C plk-2 single mutant, which was acquired using a 63X NA 1. 40 objective), deconvolved and corrected for registration using SoftWoRx. Multiple overlapping fields covering the whole length of the gonad were acquired for each specimen, and gonads were assembled using the “Grid/Collection” plugin [36] on Fiji. Final assembly of 2D maximum or sum intensity projections was performed using Fiji [54], with adjustments of brightness and/or contrast made in Fiji or Adobe Photoshop. Worms were prepared and imaged using the OMX microscope as described below for FRAP analysis (except that serotonin was not included); slides were imaged within 15 minutes after mounting. For Figs 1B and S1 (increase of SYP-3 during pachytene progression), we generated a sum intensity projection of all the stacks containing the nuclei of the top layer of the gonad; for each nucleus fully contained within this projection, we calculated the average fluorescence intensity of that nucleus as the total fluorescence in the smallest circle containing the nucleus, divided by the area of the circle. For Fig 4, pachytene nuclei were manually segmented on maximum intensity projections and scored for the presence of an uneven distribution of SYP-1 among their SCs. Nuclei in the second half of the pachytene region were assessed for presence or absence of this feature; the pachytene region was defined as spanning the distance between the most distal nucleus with six SC stretches and unclustered chromosomes to the last row containing multiple nuclei. For S6 Fig, we quantified the number of chromosome pairs that exhibited SYP-1 enrichment (within each scored nucleus); for this feature, we scored nuclei in the last quarter of the pachytene region, where the distinction between SYP-1-enriched chromosomes and SYP-1-depeleted chromosomes was most pronounced. The protocol used for detergent lysis and spreading of C. elegans germ cell nuclei (S2–S4 Figs, Figs 5 and 6) is an adapted version of the protocol used in [55] for S. cerevisiae. The gonads of 20–100 adult worms were dissected in 5μl dissection solution (see note A) on a 22x40mm coverslip (thickness #1. 5, as required for the OMX microscope). 50μl of spreading solution (see note B) was added and gonads were immediately distributed over the whole coverslip using a glass rod or a pipette tip. Coverslips were left to dry overnight at room temperature or for two hours at 37°C, washed for 20 minutes in methanol at -20°C and rehydrated by washing 3 times for 5 minutes in PBS-T. A 20 minute blocking in 1% w/v BSA in PBS-T at room temperature was followed by overnight incubation with primary antibodies at 4°C (antibodies diluted in: 1% w/v BSA in PBS-T supplied with 0. 05% w/v Sodium azide). Coverslips were washed 3 times for 5 minutes in PBS-T before secondary antibody incubation for 2 hours at room temperature. After PBS-T washes, the nuclei were immersed in Vectashield and the coverslip was mounted on a slide and sealed with nail polish. The following primary antibodies were used: chicken anti-HTP-3 (1: 500 [46]), rabbit anti-GFP (1: 750 [12]), guinea-pig anti-SYP-1 (1: 200 [22]), rabbit anti-MSH-5 (1: 10000 (SDIX) ) and mouse anti-HA (Covance)

Reliable chromosome inheritance during sexual reproduction depends on the formation of temporary connections between homologous chromosomes that enable them to segregate toward opposite spindle poles at the meiosis I division. These connections are established during an extended meiotic prophase characterized by two prominent features: a highly-ordered structure called the synaptonemal complex (SC) that assembles at the interface between aligned pairs of chromosomes, and crossover recombination events between their DNA molecules that are completed in the context of the SC. In the current work, we investigate the inter-relationships between these two hallmark features of the meiotic program in the nematode C. elegans. Our work reveals the C. elegans SC as a much more dynamic structure than is suggested by its highly-ordered appearance in EM images, and further demonstrates that the SC switches from a more highly dynamic/labile state to a more stable state as germ cells progress through meiotic prophase. Moreover, we show that formation of crossover recombination intermediates can trigger stabilization of the SC in a chromosome autonomous manner. We speculate that recombination-triggered SC stabilization may provide a means for germ cells to monitor whether chromosome pairs have acquired the prerequisite crossover intermediate needed to ensure correct homolog segregation.

Abstract Introduction Results Discussion Materials and methods

fluorescence imaging invertebrates meiosis medicine and health sciences reproductive system gonads caenorhabditis cell cycle and cell division cell processes light microscopy animals animal models caenorhabditis elegans model organisms microscopy experimental organism systems dna homologous recombination research and analysis methods imaging techniques chromosome biology proteins recombinant proteins chromosome pairs fluorescence recovery after photobleaching biochemistry anatomy nucleic acids cell biology genetics nematoda biology and life sciences dna recombination organisms chromosomes genital anatomy

2017

Meiotic recombination modulates the structure and dynamics of the synaptonemal complex during C. elegans meiosis

Combinatorial regulation of gene expression is ubiquitous in eukaryotes with multiple inputs converging on regulatory control elements. The dynamic properties of these elements determine the functionality of genetic networks regulating differentiation and development. Here we propose a method to quantitatively characterize the regulatory output of distant enhancers with a biophysical approach that recursively determines free energies of protein-protein and protein-DNA interactions from experimental analysis of transcriptional reporter libraries. We apply this method to model the Scl-Gata2-Fli1 triad—a network module important for cell fate specification of hematopoietic stem cells. We show that this triad module is inherently bistable with irreversible transitions in response to physiologically relevant signals such as Notch, Bmp4 and Gata1 and we use the model to predict the sensitivity of the network to mutations. We also show that the triad acts as a low-pass filter by switching between steady states only in response to signals that persist for longer than a minimum duration threshold. We have found that the auto-regulation loops connecting the slow-degrading Scl to Gata2 and Fli1 are crucial for this low-pass filtering property. Taken together our analysis not only reveals new insights into hematopoietic stem cell regulatory network functionality but also provides a novel and widely applicable strategy to incorporate experimental measurements into dynamical network models. Appropriate spatiotemporal control of gene expression is central to metazoan development. [1]. Combinatorial interactions of regulatory proteins with regulatory regions of DNA and the basal transcriptional machinery form the building blocks of complex gene regulatory networks (GRNs). The availability of whole genome sequences as well as advanced bioinformatics and high-throughput experimental techniques have vastly accelerated the identification of candidate regulatory sequences. However, experiments that can uncover and/or validate the underlying connectivity of GRNs remain both costly and time consuming. Consequently, our understanding of the functionality of GRNs even for the most studied model organisms remains superficial. Moreover, simply cataloguing ever increasing numbers of interactions between GRN components is not sufficient to deduce the underlying network architecture or function of individual modules. Unraveling the dynamical properties of GRNs will be the key to understanding their functionality. Throughout development, cells progress through a succession of differentiation steps from stem cells via immature progenitors to fully differentiated mature cells, and each of these subtypes is associated with a unique regulatory state of the GRN [1]. It is therefore essential to understand dynamical properties of the various regulatory states of GRNs, transitions between them and their interplay with intercellular signaling. It is unlikely that this goal can be achieved solely using experimental approaches. However, the development of dynamical models of GRNs offers great potential to interpret existing experimental data in order to gain new mechanistic insights. Various computational approaches have been used for regulatory network analysis in the past. Boolean models provide qualitative information about network behavior such as the existence of steady states and network robustness and are most useful for large networks or when experimental information is scarce [2], [3]. However to examine dynamical aspects, continuous ordinary differential equation (ODE) models are more appropriate. These models can be constructed with phenomenological descriptions of gene regulation in the form of Hill functions or based on more detailed biophysical mechanisms and derived using a statistical thermodynamics approach. Phenomenological models are useful for understanding the general dynamics of network topology. They are most effective for small to medium sized networks and can also be predictive of cellular behavior [4]. Models based on thermodynamics have the advantage of including an hypothesis about the biophysics of the system [5], [6], [7]. Most parameters in these models have a direct biochemical interpretation. Unfortunately the lack of knowledge about specific biochemical parameters usually makes it difficult to relate results from these models to experimental information about gene expression. Nevertheless this modeling approach has been shown to be useful in understanding certain bacterial gene regulation modules [8] and studying the effects of nucleosome dynamics in eukaryotic gene regulation [9]. The hematopoietic system has long served as a powerful model to study the specification and subsequent differentiation of stem cells [10]. Sophisticated cell purification protocols coupled with powerful functional assays have allowed a very detailed reconstruction of the differentiation pathways leading from early mesoderm via hemangioblasts and hematopoietic stem cells (HSCs) to the multiple mature hematopoietic lineages. Transcriptional regulators (TRs) have long been recognized as key hematopoietic regulators but the wider networks within which they operate remain ill defined [11]. Detailed molecular characterization of regulatory elements (enhancers/promoters) active during the early stages of HSC development has identified specific connections between major regulators [12], [13], [14], [15] and has led to the definition of combinatorial regulatory codes specific for HSC enhancers [16], [17], [18]. Moreover, these studies identified a substantial degree of cross-talk and positive feedback in the connectivity of major HSC TRs [19]. In particular, a triad of HSC TRs (Gata2, Fli1, Scl/Tal1) forms a regulatory module that appears to lie at the core of the HSC GRN [20]. This module consists of the three transcription factor proteins as well as three regulatory elements through which they are connected via cross-regulatory and autoregulatory interactions [12], [20] (Figure 1A). The details of regulatory interactions in this triad are shown in Figure 1B; only significant binding sites in the enhancers are shown for simplicity. Gata2-3 and Fli1+12 enhancers contain multiple Gata2 (GATA), Fli1 (ETS) and Scl (E-BOX) binding motifs. The Scl+19 enhancer contains ETS and GATA binding motifs. Scl, Gata2 and Fli1 are all essential for normal hematopoiesis in mice [12] suggesting that the triad is an important sub-circuit or kernel of the GRN that governs hematopoiesis. The triad architecture (Figure 1A) is very dense in regulatory connections and possesses multiple direct and indirect positive feedback loops. Such network topologies are rare in prokaryotes [21] but have been identified in other stem cell systems such as the Nanog-Oct4-Sox2 triad in the embryonic stem cell GRN [22], [23]. These observations suggest that the triad design may be associated with stem cell behavior. This idea prompted further investigation of combinatorial control by the triad TRs [20]. Generation of an enhancer library with wild type and mutant enhancers allowed the construction of different combinations of binding motifs in each enhancer. Wild type and mutant enhancers were sub-cloned into a SV minimal promoter and lacZ reporter vector and tested using stable transfection of hematopoietic progenitor cell lines [20]. This analysis produced results such as those schematically illustrated in Figure 1C. It has been suggested that the dense connectivity and positive feedback loops within stem cell GRN modules play important roles in stabilizing the stem cell phenotype [20]. However, the dynamical nature as to how this self-enforcing circuit may be initiated or indeed exited remains unclear. In this paper we construct a mathematical model of the Scl-Gata2-Fli1 triad module and characterize its dynamical properties using continuous ODE modeling approaches. We first propose a thermodynamic method of estimating free energies of different configurations of the enhancer regions from the measurements of the transcriptional reporter libraries. This method together with a proposed biochemical mechanism of distant transcriptional enhancement significantly reduces dimensionality of the network parameter space. Measurements of protein lifetimes provide experimentally informed timescales to model transient behavior of the network. We analyze the network response to physiologically relevant signals such as Notch, Bmp4 and Gata1 and show that the network behaves as an irreversible bistable switch in response to these signals. Our model also predicts the results of various mutations in the enhancer sequences and shows that the triad module can ignore transient differentiation signals shorter than threshold duration. The combination of a bistable switch with short signal filtering not only provides new mechanistic insights as to how the Scl-Gata2-Fli1 triad may function to control HSC specification and differentiation but also suggests a possibly more general role for this network architecture in the development of other major organ systems. Full quantitative characterization of the combinatorial nature of transcriptional regulation requires measurements of binding affinities between the DNA and TRs as well as interaction strengths among TRs. Moreover, the contribution of each individual TR and each possible combination to the transcriptional rate must be assessed. This information is extremely tedious to measure due to the combinatorial multiplicity of TR configurations and does not exist for the majority of experimental systems. Experimental data for synthetic libraries of transcriptional reporters that contain the gene regulatory elements is more readily available. We develop thermodynamic methods to characterize the combinatorial transcriptional regulation by distal enhancers based on this type of data and apply it to model the Scl-Gata2-Fli1 triad - a core module of the GRN of hematopoietic stem cells. Recently this system has been experimentally characterized [20]. In this study distal enhancer regions regulating the transcriptional rate of network proteins were identified and the relative contributions of each of the regulatory motifs were thereafter assessed individually and in combination by the use of a suitable transcriptional reporter (e. g. , luciferase, lacZ). The typical results from these experiments are illustrated in Figure 1; see Table S1 for the full data used. We use this data to obtain the functional form describing the transcriptional rate of the reporter-enhancer constructs and estimate the biochemical parameters characterizing this function. Below we illustrate our approach for the Scl+19 enhancer; the full model is derived in the methods section. We assume that the distant enhancers increase the transcriptional rate via modulation of chromatin remodeling rather than through direct interaction with transcriptional machinery. This assumption is motivated by the observations that activation of the Scl+19 enhancer is only revealed upon integration of the enhancer-promoter construct into chromatin and that the activity of the enhancer is independent of its position (upstream or downstream) relative to the reporter gene [20], [24]. Moreover, when integrated as single copy reporters into the genome of embryonic stem cells and assayed following 5 days of in vitro differentiation, the difference between wild type and mutant enhancer constructs lies in the number of cells that express the transgene rather than the level at which it is expressed (cf. Figure S1 and Text S1). Taken together, these observations suggest that chromatin dynamics play a significant role in the action of TRs at the enhancers. In the absence of enhancer binding, the gene can be in either open or a relatively stable closed chromatin state. In the closed chromatin state the binding regions for the TRs and the transcriptional machinery are wrapped in nucleosomes and are inaccessible; thus no gene expression is possible from this state. The closed chromatin state can spontaneously unwrap to an open state where the binding sites become accessible to allow polymerase to bind to the promoter and initiate transcription. Since most promoters bind RNA polymerase weakly, the probability of RNA polymerase binding and subsequently transcription rate is proportional to the probability of the chromatin being in the open state (; see Methods Eqs (15) – (17) ). This probability depends on the equilibrium between open and closed chromatin states. Binding of the TRs at the enhancer stabilizes the open conformation thus shifting the equilibrium towards the open state (cf. Figure S2). This way the probability of open conformation increases with increase in TR concentration or increase in binding affinity. The rate of gene expression is still given by but is now defined by a more complicated thermodynamic expression accounting for all the possible configurations of TR binding. Mutations in the enhancer site eliminate the configuration of TR binding thereby affecting but not. Below we illustrate this formalism for the Scl+19 enhancer. The Scl+19 enhancer contains binding sites for Gata2 and a Fli1 dimer and therefore can exist in closed and four different open states (enhancer empty, Gata2 bound, Fli1 dimer bound, both Gata2 and Fli1 bound). The cumulative probability of all open state configurations is then given by, where is the probability of the closed state given by (1) where subscript s denotes the the Scl+19 enhancer: is the effective closed state energy, and is the partition function given by the sum of exponentiated free energies of each state: . is an inverse temperature and hereafter all free energies are in its units. For TR-bound states, free energies are concentration dependent due to the loss of entropic degrees of freedom, e. g. for the Gata2-bound state, where denotes concentration of Gata2. (Similarly and denote concentrations of Scl and Fli1 respectively). Since the free energies are only defined up-to a constant we can choose the free energy of the open state to be zero and thus obtain the following expression for the partition function: (2) where and represent the free energies of Fli1 dimer and Gata2-Fli1 multimer binding and is the partition function for all open chromatin states. We use the subscript in all these terms to specify that they are associated with the Scl+19 enhancer and the superscript to specify the binding configuration (cf. Table S2 for notation). Direct measurements of the binding free energies in this expression may be tedious but these can be straightforwardly computed from the ratios of the transcription rates from synthetic reporter libraries with full or mutated enhancer sites. Ratios of the reporter expression levels of cell lines with wild-type (wt) and mutated (mut) enhancers can be used as constraints on the values of the binding free energies. (3) Equations similar to (3) can be constructed for all reporter-enhancer libraries and used to recursively compute the binding free energies (cf. Eqs (22) – (27) in Methods and Eqs (S. 1) – (S. 11) in Text S3). Scl, Gata2 and Fli1 form an interconnected triad of positive interactions and play an important role in hematopoietic differentiation [12], [20]. To understand the role of the unique architecture of the triad module we construct a dynamical model of the system. Assuming first-order degradation kinetics, deterministic rate equations for the change in TR concentrations take the form (4) where the functions, and describes the rates production whereas, and denote degradation rate constants for Scl, Gata2 and Fli1 respectively. Rate constants for protein degradation are estimated from known half-lives of the proteins. Since proteins are long-lived relative to mRNA, we can assume that production rates are directly proportional to the respective transcription rates (cf. Eq (28) ). In addition to distant enhancers, Notch and Bmp4 are known to serve as activators of the promoters of Gata2 and Fli1, Gata2 respectively [25], [26]. These activators increase the rate of transcription by increasing the recruitment of RNA polymerase to the respective promoter. In particular, Notch and Bmp4 increase Gata2 expression by 3. 5 fold [26] and 4 fold [27] respectively. In this case, to compute one needs thermodynamic expressions of the probabilities of multiple open conformations corresponding to binding of Notch or Bmp4. These probabilities depend upon Notch and Bmp4 concentrations (and respectively) and their binding energies and via the full partition function (subscript g stands for Gata2-3 enhancer): (5) Here is the equilibrium constant for chromatin transitions between open and closed states for Gata2 enhancer (similarly and for Scl+19 and Fli1+12 enhancers respectively). These equilibrium constants are dimensionless quantities characterizing the maximum possible fold enhancement of gene expression by the respective enhancer. The partition functions are used to compute Gata2 synthesis rate (cf. Eq (20) ). The same procedure is used to describe the rate of expression of Fli1, although in this case only Bmp4 acts at the promoter (cf. Eqs (21) ). Conversion to dimensionless form can greatly simplify the model allowing easy interpretation of simulation results. We normalize the species concentrations of Scl, Gata2 and Fli1 as, and. , and represent the mean observed concentrations of Scl, Gata2 and Fli1 in wildtype HSCs where the triad is actively expressed. In addition, and are Notch and Bmp4 concentrations normalized with respect to their promoter dissociation constants. With these normalizations, wildtype HSCs in the absence of signals would have and. We choose this state as a reference state for the estimation of free-energies (cf. Methods Section for details). The dimensionless form of equation (4) is then given by (6) Where are dimensionless synthesis rates (cf. Eq 25). Note that in the final form of our model equations the wild-type state of HSCs is always a steady state in the absence of signal. By using the parameter estimation method described in the previous section and reduction of the system to dimensionless form, we have reduced the dimensions of the parameter space and the only free parameters are the equilibrium constants for chromatin opening-closing and. In the following sections we use this ODE model to analyze steady state and dynamical properties of this triad module. We use the model developed in the preceding sections to analyze the steady state response of the triad to Notch and Bmp4. By varying and and calculating free energies that conform to the experimental predictions of mutant enhancer expression rates we can explore all regions of the relevant parameter space. Bifurcation analysis of the steady state response shows that the triad module has two stable steady states (see Figure 2). For certain values of the chromatin equilibrium constants Notch and Bmp4 can switch the triad between a low expression OFF state and a high expression ON state (Figure 2A). This switch in expression levels is irreversible and sustained even without Notch and Bmp4 signals. Therefore transient Notch/Bmp4 signals may lock the triad into the ON state. This irreversible progression switch behavior is expected from the triad module which has been reported to play a significant role in the specification of HSCs in the hemogenic endothelium. We use the above-described approaches to estimate the parameters for our model. Equations (25) – (27) relate the gene expression results from the Scl+19 enhancer to the chromatin equilibrium constant. When we use these equations to estimate the free energies and the model results match the experimental results exactly. The matching is only possible for the values of equilibrium constant above a threshold: . This lower bound is simple a consequence of the fact that in the proposed thermodynamic framework the maximal possible enhancement is given by and the experimentally measurable enhancement is 820. 51. Similarly the free energies for the Gata2-3 and Fli1+12 enhancers are estimated based on the experimental results and and respectively (cf. Methods section and Text S3 for details). The values of these constants are also limited from below by the respective maximal measured enhancer factors. In addition qualitative information about system behavior, namely its switchability as a response to physiologically relevant Notch and Bmp4 signals, places an upper bound on chromatin equilibrium constants values. For a different set of values the computed free energies are such that Notch and/or Bmp4 cannot cause the switch between low and high steady states (Figure 2B, C). As a result the system remains switchable in the very narrow range of two equilibrium constants where the full enhancer brings the transcriptional rate to a nearly saturated value. The resulting narrow ranges do not indicate lack of model robustness but rather are a consequence of strict constraints placed on free energy values by the exact matching to the experimental reporter data (cf. equations (S. 1) – (S. 11) in Text S3). In fact without these constraints the range of and for switchable bistable response extends over several orders of magnitude (cf. Figure S3 and below). If we tolerate some deviation from the experimentally measured transcriptional data we can relax these constraints and significantly enhance the range of parameter values for which the system is bistable and switchable. For example, if we allow up to 20% deviation from transcriptional reporter measurements then the values of chromatin equilibrium constants can vary by 20% and still result in switchable response (data not shown). It is quite reasonable to tolerate such levels of deviation from the experimental results because the experimental results usually have a margin of error. Therefore we find that the qualitative predictions of the model (switchable bistable response) are robust however the quantitative predictions (transcriptional data) are only as accurate as the experimental data one which the model is based. We expect the triad to be switchable in response to both Notch and Bmp4. Therefore we choose the chromatin equilibrium constants from within the narrow ranges shown above and calculate the TR-enhancer binding free energies using these chosen values. For this chosen set of parameter values the model shows an irreversible bistable response to Notch and Bmp4 (Figure 2A). Bmp4 concentrations were set to zero for evaluating the Notch dose response and vice versa. The presence of one signal reduces the threshold concentration of the other signal at which the triad switches from OFF to ON (data not shown). The calculated free energies are shown in Table S3 and used through the remaining simulations. Once the free energies of TR binding are fixed at Table S1 values, the system becomes robust to variability of chromatin equilibrium constants (Figure S3). Such changes may biologically correspond to histone modification or other physical perturbations. In response to changes over a large range the triad shows switchable and irreversible bistable responses to Notch and Bmp4 (Figure S3). Therefore the switchable nature of triad bistability is robust to several fold parameter changes. Gata1 can displace Gata2 from its binding sites in the Gata2-3 enhancer. Through competition for binding sites and subsequent chromatin remodeling Gata1 can switch the triad from high expression back to the low expression state. We represent the chromatin remodeling effect of Gata1 by including a factor 0<<1 in our expression for the rate of Gata2 gene transcription. Because the exact biochemical mechanism of the Gata1 action is not established we choose a decreasing function of Gata1 and make no other assumptions about the functional form of. We therefore, plot Gata1 dose-response curves with as the x-axis where its values decrease left to right (Figure 2D). This phenomenological description of the effect of Gata1 captures the effect it has on RNA polymerase recruitment to the promoter by initiating chromatin remodeling. Inclusion of Gata1 in our model (Figure 2D) allows the system to switch from ON to OFF states. The switching is irreversible – the system will remain OFF even after Gata1 signal is gone (). Notch and Bmp4 concentrations were fixed at zero for evaluating the Gata1 response because the concurrence of Notch/Bmp4 and Gata1 signals is physiologically unlikely. Interestingly, Gata1-deactivation is far more susceptible to noise than the activation by Notch/Bmp4. This can be concluded from the dotted line representing the unstable steady state that separates the stable ON and OFF states (compare Figures 2A and D). This line characterizes the magnitude of concentration fluctuations required for spontaneous transitions. For sub-threshold signals, this line is much closer to the stable steady state in Gata1 dose-response curves (Figure 2D) as compared to Notch or Bmp4 curves (Figure 2A). A more rigorous investigation of the magnitude of stochastic effects and their relation to separatrix of deterministic model requires a full stochastic model of the network and will be conducted elsewhere. We expect the steady state response of the Scl-Gata2-Fli1 module depends on the triad architecture and design of enhancers. The model presented above allows us to verify this claim by introducing changes in the triad design corresponding to mutations of enhancer sequence and gene knockouts and examining the effects on the steady state response. To this end, we systematically deleted TR-binding sites from each enhancer in silico and analyzed the steady state response of the system. We also analyze the steady state response of Scl, Gata2 and Fli1 deletion mutants. Mutations in the triad enhancer sequences can produce many modules with simpler architecture as shown in Figure 3. Notably, since some TR-enhancer configurations do not make a significant contribution to the enhancer activity, removal of a single enhancer binding site might effectively eliminate multiple TR-enhancer interactions. For example, the effect of Scl on the Gata2 and Fli1 enhancers is only significant when both Gata2 and Fli1 are bound to the enhancer. Therefore the probability of Scl bound enhancer configurations for these enhancers is negligible for any motif where the Gata2 or Fli1 sites on these enhancers are deleted. Keeping this in mind we analyze 10 different triad module designs that can be obtained by selective single and double mutations of enhancer binding sites. The model described above is suitably altered to predict the steady state response of these alternate designs. All relevant parameter values are taken from the full triad model. Of the 10 “mutant” designs, all 6 modules where the Scl+19 or Gata2-3 enhancers are mutated show only a single steady state with the expression of Scl, Gata2 and Fli1 comparable to the low expression state of the full triad (cf. Figure 3A). On the other hand, high levels of expression can still be observed in 4 modules with mutations in the Fli1+12 enhancer (see Figures 3B and 3C). However, in contrast to wild-type (Figure 2A), this high level of expression cannot be maintained in the absence of Notch and Bmp4. Even when the E-BOX biding site for Scl is eliminated from the Fli1+12 enhancer the system remains bistable for a range of signal. For the designs in which the GATA site in the Fli1+12 enhancer is eliminated (Figure 3C) Fli1 expression is uncoupled from Gata2 and Scl and is monostable while the responses of Scl and Gata2 are still bistable. This is expected because Fli1 autoregulation is not strong enough to produce bistability. Complementarily, we can also assess the effects from alterations of TRs rather than their binding sites. Simulations show that Scl−/−, Gata2−/− and Fli1−/− knockout mutants cannot support the high expression state of the triad. These mutants produce a phenotype similar to the enhancer mutations in Figure 3A. Comprehensive analysis of knockout mice has shown that hematopoiesis is severely impaired in all three deletion mutants [28], [29], [30], [31]. Our model suggests that the knockout of any of the triad proteins prevents the switch to ON state which is likely to affect the specification of HSCs during early embryonic development and therefore compromise the development of all mature blood cell types as seen experimentally. On the other hand, the irreversible bistability of triad response is preserved if we delete one chromosomal copy of any one of the three triad genes; however the heterozygotic mutants are expected to be more prone to differentiation (cf. Figure S4 and Text S2). This could explain why these mutants have reduced repopulation capacity [32], [33]. The dynamics of the response of the bistable triad module to a pulse of Notch is illustrated in Figure 4A. The step increase in Notch concentration almost immediately increases Gata2 concentration slightly. However Fli1 concentration remains stagnant because Scl level rises very slowly. The slow speed of Scl response is governed by its slow degradation rate (half life ∼8 hrs). Once enough Scl has accumulated, the probability of Scl being present on the Gata2 and Fli1 enhancers becomes significant. This results in a rapid increase of expression rates and the triad switches to the high expression state. The rate limiting step for switching ON the triad expression levels is therefore the slow accumulation of Scl. To further investigate the dynamics of triad switching in response to transient stimuli we have computed the minimal pulse duration that can cause irreversible switching as a function of signal amplitude (Figure 4 B, C; black lines). The results indicate that the system can be switched ON by signal pulses longer than a certain threshold level (∼42 hrs for a Notch pulse and ∼21 hrs for a Bmp4 pulse). This threshold is a few fold larger than Scl-lifetime, the longest timescale for the system. Our simulations therefore indicate that the triad module is capable of filtering transient signals that are shorter than the threshold simulation. We refer to this property as low-pass filtering – a term accepted for similar phenomena in engineering literature [34]. This filtering appears to be related to the slow turnover of Scl and the feedback loops connecting Scl with Gata2 and Fli1. To understand how slow Scl dynamics contributes to the filtering of transient Notch and Bmp4 signals we compare the dynamics of the triad module to that of a simpler network module where the Scl+19 enhancer has been deleted. We call this module the reduced module. In this reduced module Scl is assumed to be under an external regulator that controls Scl concentration. With this reduction, Scl concentration is constant and the dynamics of Gata2 and Fli1 response are not limited by the slow accumulation of Scl. For a controlled comparison of the dynamics [35] we assume that all relevant parameters have the same values as they do in the full triad model. This leaves the Scl concentration as the only free parameter. The reduced module shows irreversible bistable response to Notch, Bmp4 and Gata1 for a range of Scl values. We constrain the Scl concentration such that the threshold for OFF to ON transitions is the same for the reduced module and the full triad (Figure 4D). Notably, the separatrix between the two stable states (dotted line, Figure 4D) is much closer to the ON state for the reduced module. This suggests that the reduced module is more susceptible to fluctuations in TR levels as compared to the full triad. We now use the reduced module as described above for a controlled comparison of the dynamics of the OFF to ON and ON to OFF switching. Both bistable switches act as filters for transient signals above the threshold (Figure 4 B and C). We compared this dynamic response of the triad and reduced modules to Notch and Bmp4 pulses. The models for the two modules have the same Notch/Bmp4 thresholds and close to the threshold the minimum pulse duration for both modules is high. However at higher concentrations of Notch and Bmp4, the minimum pulse duration is much higher for the triad module than for the reduced module (16 hrs and 9. 5 hrs for Notch and Bmp4 pulses respectively). These results show how the slow dynamics of Scl allow the full triad module to act as a better low pass filter function for activation as compared to the reduced module. For a controlled comparison of the response of the two modules to Gata1 we fix the Scl concentration of the reduced module such that the threshold level of Gata1 is identical (Figure 4E). This fixed concentration of Scl is 4 fold higher for deactivation than for activation. Gata1 acts at the Gata2-3 enhancer to shut off transcription through chromatin remodeling. The slow dynamics of Scl do not affect the Gata2 concentration during this deactivation. As a result the deactivation dynamics and the minimum pulse duration for ON to OFF switching at high Gata1 concentrations (∼8 hrs) of the reduced module and the full triad are identical. The triad and reduced module are equivalent low pass filters for deactivation signals such as Gata1. Combinatorial gene regulation is ubiquitous in eukaryotes with complex DNA regulatory regions acting as integration points for multiple signals and pathways involved in gene regulation. The characterization of these regulatory regions through mathematical models is an important step towards understanding the functionality of gene regulatory networks. In order to fully characterize each regulatory element, one needs to determine dynamical functions that describe the rate of transcription as a function of TR concentrations. The most biochemically and biophysically realistic method of characterizing transcriptional regulation is rooted in statistical thermodynamics where each state of the regulatory region is assigned a free-energy so that the probability of each state can be computed from Boltzmann distribution [5]. These methods have been previously applied to bacterial systems [8] but rarely used for eukaryotic gene regulatory networks as a lack of reliable parameter measurements prevents researchers from undertaking detailed modeling approaches. Here we have developed a method for the quantitative characterization of combinatorial gene regulation by multiple TRs in eukaryotic distant enhancers. Our proposed method extends the thermodynamic approach of [36] in order to relate it to experimental transcriptional reporter assays. We develop a recursive method to estimate relevant free energies from the measurements of combinatorial libraries of transcriptional reporters. There are multiple benefits of computing free-energies of TR-DNA configurations. First, these parameters allow straightforward construction of mathematical models for quantitative analysis of system behavior with no or just a few free parameters. Second, free energies can be used for model reduction by specifically excluding thermodynamically unfavorable states and subsequent model reduction. Third, the parameters provide important qualitative insights into gene regulatory mechanisms such as cooperativity of TRs. We further reduce the number of parameters required to characterize the distant transcriptional enhancers by proposing a detailed mechanism based on the modulation of chromatin remodeling activity. Chromatin structure is known to play an important role in eukaryotic gene regulation. The organization of DNA into nucleosomes can prevent the transcriptional machinery and regulatory factors from accessing regulatory regions. The detailed mechanism of action of distant enhancer sites has not been established. It has been suggested however that its action may involve modulation of chromatin remodeling dynamics [37]. For instance, regulatory elements of the Scl-Gata2-Fli1 triad were shown to be critically dependent on integration into chromatin [12]. Here we propose a ratchet mechanism of enhancer action (cf. Figure S2). We propose that DNA can be in a dynamic equilibrium between open (promoter site accessible) and closed (promoter site inaccessible) conformations. Such a dynamic equilibrium between wrapped and unwrapped nucleosomal DNA has also been discussed elsewhere [38]. In the absence of enhancer TRs, the equilibrium is heavily shifted towards a closed state resulting in very low transcription probability. We hypothesize that binding of TRs to the enhancer site stabilizes an open conformation and thereby shifts the equilibrium towards it. This mechanism therefore allows the TRs to ratchet the spontaneous unwrapping of nucleosomal DNA and trap it in a state accessible to the transcriptional machinery. We apply this thermodynamic framework to a regulatory module hypothesized to play a pivotal role in hematopoiesis. Under this assumption the binding of Fli1, Gata2 and Scl to their enhancer sites activates gene transcription by increasing the probability of transcription rather than the rate of transcription. This hypothesis is consistent with previously reported results of studies focused on enhancer function in mammalian cells [37], [39], [40] and with our flow cytometry experiments with cells containing the Scl+19 enhancer-reporter constructs (cf. Figure S2 and Text S1). The proposed mechanism assumes that the unwrapping of DNA from nucleosomes is independent of all triad factors and thus effectively spontaneous. However chromatin modification and chromatin remodeling factors can affect these nucleosome dynamics. In particular, factors such as the Gata2 repressor Gata1 may regulate the expression by modulating free energies of DNA unwrapping through chromatin modification. By shifting the equilibrium further towards the closed state, Gata1 can suppress transcription to such an extent that TR concentrations are too low to ratchet the very short-lived open state. The recently characterized Scl-Fli1-Gata2 triad module includes a large number of transcriptional interactions resulting in multiple positive feedback loops. The complex enhancer structure makes it rather difficult to phenomenologically deduce dynamical expressions for Scl, Gata2 and Fli1 transcription. However, with our newly developed approach based on transcriptional reporter data, construction of a mathematical model of the triad becomes a straightforward task. The resulting model of the triad exhibited bistability in response to the action of Notch and Bmp4. We have chosen the free energy values for DNA unwrapping to ensure that the action of these two activators at the promoters switches the triad from low expression (OFF) state to high expression state (ON). The model predicts this switching to be irreversible – the triad will remain ON even after the signals are gone (Fig 2A). The development of HSCs in the hemogenic endothelium is known to be a Notch regulated event [41]. Notch is known to be expressed in endothelial cells and act as a regulator of Gata2 expression during the onset of hematopoiesis [26]. Bmp4 expression has also been observed in the dorsal aorta region where HSCs first develop in the embryo [18], [42]. Notch and Bmp4 are known to be mediators of HSC specification during embryonic development [41]. Our model shows how the action of Notch and Bmp4 is crucial for the OFF to ON switch of the Scl-Gata2-Fli1 triad. Since HSC specification requires Scl, our model predicts that in the absence of Notch and Bmp4, newly generated HSCs are trapped in a low expression state and hematopoietic development is compromised. The network also irreversibly switches from the ON to OFF state when reaching a threshold value of repression of Gata2 transcription by Gata1. The network will then remain in the OFF state in the absence of other signals. Interestingly, in ref. [22] the authors use a mathematical model to predict that a similar triad module in embryonic stem cells is also bistable. However their module is expected to be bistable only in the presence of activating or deactivating signals unlike Scl-Gata2-Fli1 triad that shows irreversible bistability. Our analysis indicates essential roles of all the enhancer sites included in the model in maintaining irreversible bistability in steady state dose-response curves of the triad. Elimination of any binding sites in Scl or Gata2 enhancers leads to complete elimination of bistability with only the OFF state remaining. Mutations in the Fli1 enhancer may lead to a reversible bistability phenotype in which the triad is activated only in the presence of Notch and/or Bmp4 signals above a certain threshold. We emphasize, however, that these predictions do not indicate that simpler triad networks with less autoregulation are incapable of achieving irreversible bistable switching behavior. Our goal was to predict the behavior of the triad to the mutations of the regulatory regions. If one allows compensatory changes in other model parameters one can restore the irreversible switching behavior and even set the switching threshold to be equal to that of wild-type triad. However, as indicated below the reduced modules may still display physiologically important differences in other aspects of dynamic behavior. In order to characterize the transient responses of the triad module one needs the values of kinetic parameters – lifetimes of triad proteins. Scl and Fli1 are known to be relatively stable proteins with measured half-lives of 8 hours and 2 hours respectively [43], [44]. Gata2 is comparatively unstable with a half-life less than 30 minutes [44]. This combination of short-lived and long-lived transcription regulators allows the triad to respond quickly to changes in mRNA transcription rates and at the same time, act as memory modules for history-dependent switches into and out of the HSC regulatory state. Analysis of the dynamical response of the triad to Notch/Bmp4 indicates that slow accumulation of Scl acts as a rate-limiting step for OFF-ON switching. As a consequence the triad must be exposed to Notch/Bmp4 signals for significant time periods for switching to occur. Physiologically this means that the triad motif works as a low-pass filter that responds only to transient stimuli longer than threshold duration and ignores brief, transient signals shorter than the threshold duration. All bistable switches show this type of threshold filtering of transient signals but to a different degree [45]. In our case, the response rate for the triad is limited by slow Scl dynamics and therefore multiple features of the triad network contribute to this property. For example, Scl is the slowest in degradation among the TRs and Notch/Bmp4 signals affect its accumulation only indirectly (Figure 1A). In addition, we hypothesized that the positive feedback loops involving Scl play a significant role in determining the threshold for low-pass filtering. We have confirmed this hypothesis by comparing the response of the triad to a hypothetical reduced module wherein the Scl+19 enhancer is deleted and Scl acts as an external TR for the Gata2-Fli1 feedback loop [35]. We therefore conclude that the full triad is a better low pass filter because of the rate-limiting nature of Scl accumulation and Scl-mediated positive feedbacks significantly affect the signal filtering properties of the triad. Studies in heterogeneous cell populations derived from differentiating ES cells or mouse fetal liver had suggested low level binding of Scl itself to the Scl+19 enhancer [18]. However, more recent analysis in a clonal population of blood stem/progenitor cells did not detect any binding of Scl to this element [15]. Positive autoregulatory feedback through the Scl+19 enhancer is therefore unlikely to play a significant role in stem cells, especially as the Scl+19 element does not contain a bona fide binding site for Scl which would necessitate indirect binding. Nevertheless, we have considered the addition of a positive auto-feedback loop on Scl but simulations demonstrated that it does not generate a qualitatively different scenario with the only major consequence being a further slow-down of the switching rate due to the retardation of response by positive feedback (data not shown). Gata1 acts at the Gata2-3 enhancer and is reported to actively promote chromatin modification [46]. The decrease in Gata2 concentrations is not limited by Scl dynamics because Gata1 directly affects Gata2 transcription by reducing RNA polymerase recruitment. We therefore expected that filtering characteristics of the full and reduced triad motif would be the same. We performed a controlled comparison choosing a concentration of Scl in the reduced module (with the Scl+19 deleted) that ensures the same switching threshold. The results indeed show essentially identical low-pass filtering properties of the two modules because Scl dynamics are not rate limiting in this case. Experiments have shown that the knockout of any one of the genes Scl, Gata2 or Fli1 affects the development of HSCs and leads to severely impaired hematopoiesis. Thus the expression of these TRs is critical for hematopoiesis. More recent studies have shown that these three genes regulate each other by acting at distant enhancers as activators. Results from our model provide insight into the function of this module of TRs and suggest that the triad is a central regulator that controls the specification of HSCs during early hematopoiesis and the generation of progenitors committed to differentiation from these cells. The bistable switch properties of the triad are hallmarks of a decision module. The triad switches irreversibly from the low to high expression state in response to external cues such as Notch and Bmp4 that are important for establishing definitive HSCs in the hemogenic endothelium. The bistable response predicted by the model is robust to fluctuations in parameter values. Experimental results also support this prediction [47]. The model shows that the knockout mutants are unable to reach the activated high expression state due to the all or none nature of this bistable response. Additionally the slow turnover of Scl retards the triad response to Notch and Bmp4 and thus makes it a highly effective low pass filter for noise in these signals. The response to deactivation by Gata1 is not affected by Scl dynamics. As a result the ON to OFF switch for the triad is much faster than the OFF to ON switch. Deactivation by Gata1 is also more sensitive to stochastic fluctuations in triad protein concentrations. The cells can be switched to the OFF state to produce progenitor cells committed to differentiation by fluctuations in triad TR concentrations. Thus asymmetric partitioning of these proteins during cell division can allow sub-threshold Gata1 concentrations to silence Gata2 expression in one of the daughter cells by chromatin modification. The probability of this stochastic exit from the pluripotent HSC state of the cell is governed by the Gata1 concentration in the cell. This observation is consistent with experimental analysis of a multipotent hematopoietic progenitor cell line which demonstrated that these cells exist in two distinct subpopulations when cultured under self-renewal conditions with the more differentiation prone subpopulation expressing higher levels of Gata1 [48]. Of note, the triad switches between states in an all or none fashion where overexpression of exogenous Gata2 for example could prevent deactivation of the triad by Gata1. In line with these predictions, it has been demonstrated that overexpression of Gata2 in differentiating ES cells increases the production of hematopoietic progenitors and slows down their differentiation [49]. Our model of the triad module shows that it responds differently to activation and deactivation signals. This allows the OFF to ON and ON to OFF switches to fulfill different functional requirements. The activation response is slow, irreversible and robust to fluctuations in external signals to allow the development of HSCs in a noisy intercellular signaling environment. Simulation results for the dynamics of deactivation suggest that it may be faster than the OFF to ON switch and may exploit stochastic intracellular fluctuations during the cell cycle to maintain the HSC population and guarantee a continuous supply of lineage committed progenitors at the same time. From a model based on the quantitative experimental characterization of the triad enhancers we have predicted several qualitative features of the steady state and transient response of the triad as well as its sensitivity to mutations and over-expression. We favored a deterministic model for our analysis of the triad function because of the reliability and robustness of the predictions that we have been able to extract from this approach. Even so, a stochastic model can potentially offer additional information about noise properties of the system and we intend to use results presented here to guide the construction of a full stochastic model in the future. Taken together the results presented here are consistent with prior experimental data and provide new mechanistic insights into potentially critical features of the regulatory networks that govern the specification and subsequent differentiation of hematopoietic stem cells. Moreover, our strategy of exploiting experimental data to infer biophysical properties should be widely applicable to aid regulatory network reconstruction in a wide range of cellular and developmental systems. We extend the Shea-Ackers [7] description of gene regulation to construct the deterministic models discussed above. The following assumptions are the foundation of this modeling approach, The assumption of equilibrium allows us to calculate the probability of finding TRs bound to DNA using the Boltzmann weighting factors for all configurations (occupied and unoccupied) of the DNA regulatory element [5]. The sum of the Boltzmann factors for all configurations is the partition function (7) Here is the free energy of the state (we measure free energies in units of and use). The partition function is used to calculate the probability of each of configuration. We distinguish three different types of regulatory element configurations based upon our model of nucleosome dynamics. We use these definitions to formulate the probabilities and of open chromatin with no TR binding, different enhancer bound states and closed chromatin respectively: (8) (9) Here the energy for all states is measured relative to the open chromatin state (which is set to zero). The Gottgens group cloned the Scl+19, Gata2-3 and Fli1+12 enhancers upstream of a SV promoter controlling a lacZ reporter gene and integrated this construct into the genome of wild-type HSCs that show high expression of Scl, Gata2 and Fli1 [20]. In the presence of all three TRs, the enhancer can be occupied in many different TR configurations and reporter expression is significantly higher than constructs with no enhancer. Mutant enhancers where certain TR binding sites have been deleted were also used with reporter gene constructs to measure the gene expression enhancement. The results from these experiments show that only the deletion of certain critical enhancer binding sites affects gene expression enhancement. These critical sites are shown in Figure 1 and the experimental results from [20] are included in Table S1. We use these results to simplify the model of combinatorial gene regulation in the triad. The expression of Scl is under the control of two TRs Gata2 and Fli1 with different binding sites in the Scl+19 enhancer. The enhancer can therefore be in either closed state, open state, bound by Gata2, bound by Fli1 dimer or bound by Gata2 and Fli1 dimer simultaneously. Given the various configurations of the enhancer, the derivation of the partition function is straightforward (cf. Eq (2) ). We define as the sum of the Boltzmann weights of all open state enhancer configurations for ease of representation of the probability of open chromatin states in equation (9). Note, that the binding energy includes the TR-TR interaction of Gata2 and 2 Fli1 TRs while bound to DNA. The Gata2-3 enhancer includes binding sites for Gata2, Scl and the 2 Fli1 TRs. Many TR binding sites can be deleted without affecting the reporter gene expression enhancement [20]. The binding sites for Scl, Gata2 and Fli1 shown in Figure 1 are critical for gene expression enhancement. Gene expression is decreased but still significantly enhanced if only the Gata2 or Fli1 sites are present. Deletion of all sites except Scl binding site makes the expression enhancement negligible. However deletion of only the Scl site significantly decreases the expression enhancement from the full enhancer. These results suggest that although Scl binds weakly to the incomplete enhancer by itself, the Scl-Gata2-Fli1 complex has great affinity for the Gata2-3 enhancer. Of all possible configurations of Gata2-3 enhancer occupation only the Gata2 bound, Fli1 bound, Gata2-Fli1 bound and Scl-Gata2-Fli1 bound configurations are therefore included in the partition function for Gata2-3. (10) Figure 1 also shows the critical Fli1+12 enhancer binding sites. This enhancer includes two Gata2 biding sites (primary site at 5′ end). The Scl binding site and the secondary Gata2 site (3′ end) cannot enhance gene expression by themselves. The primary Gata2 site and the Fli1 dimer sites have some effect on gene expression and together they raise gene expression ∼20 fold. Single mutation of either the Scl or secondary Gata2 binding sites has a negligible effect on the gene expression. Deletion of both sites together reduces the gene expression enhancement from ∼60 fold to ∼20 fold. Thus the Gata2 bound, Fli1 bound, Gata2-Fli1 bound and Gata2-Scl-Fli1-Gata2 bound configurations have a significant effect on the gene expression. Incorporating these experimental results simplifies the partition functions for Fli1. (11) So far we have enumerated all configurations of the enhancers. Notch (), Bmp4 () and RNA polymerase () each can bind at different promoters in the triad when chromatin is on the open state with binding affinities that are represented here as free energies and respectively. These free energies can vary for different promoters and also depend upon energy of interactions between different proteins bounds to DNA. We note that the triad enhancers bind TRs to regulate gene expression in a chromatin integration dependent manner [12]. Moreover the position of the enhancer does not affect its ability to regulate transcription. These results suggest that the enhancer bound TRs do not physically interact with promoter bound factors such as Notch, Bmp4 and RNA polymerase to affect transcription. Therefore we assume that the free energy of interaction between enhancer and promoter bound proteins is zero. We assume that the binding of Notch/Bmp4 and RNA polymerase at the promoter is cooperative. Under this assumption the binding of RNA polymerase at the promoters is enhanced by the free energies of its interaction with Notch () and Bmp4 (). In our partition functions, we now account for configurations where either the enhancer or the promoter or both or neither are occupied by the various factors. We assume that because typical promoters bind RNA polymerase weakly and use this assumption to simplify the equations below. (12) (13) (14) Interestingly, even though we assumed in our derivation that there is no physical interaction between enhancer bound and promoter bound TRs we find that the partition functions of the Gata2-3 and Fli1+12 enhancers are not separable () into distinct factors and representing the partition functions for the enhancer states and promoter states respectively. Therefore the binding of TRs at the enhancers and the promoter is not independent. This emergence of cooperativity from competition of TRs with nucleosomes has been observed experimentally [50] and incorporated into mathematical models [9]. We define, and to be the equilibrium constants of chromatin rewrapping for the Scl, Gata and Fli1 respectively. Using equation (12), the probability of RNA polymerase being bound to the Scl promoter can be written as (15) Similarly we can write the expressions for the probability of Gata2 and Fli1 promoters being occupied by polymerases. (16) (17) We note that the effect of Notch and Bmp4 on the probability of transcription from the Gata2-3 enhancer is saturable because Notch and Bmp4 concentrations (and respectively) appear in both the numerator and denominator of the expression for (cf. Eq. (16) ). Similarly the effect of Bmp4 on the probability of transcription from the Fli1+12 enhancer () is also saturable (cf. Eq. (17) ). The rate of gene expression for gene, is assumed to be proportional to the probability of promoter occupation by RNA polymerase. The proportionality constant is the rate of isomerization of RNA polymerase to the open conformation. We rearrange the rate of gene expression as (18) represents the maximal rate of expression from the promoter in the open state. is a dimensionless rate of transcription that represents the cumulative regulatory effect of all enhancer and promoter bound TRs. Using equations (15) – (17) we can now write the expressions for and. (19) (20) (21) Deletion of binding sites from the enhancer modifies, the partition coefficient for all bound configurations of that enhancer. Experimental results from the Gottgens group describe the fold-change in gene expression enhancement due to the selective mutation of certain enhancer binding sites [20]. Using their results for deletion of critical binding sites we can estimate the free energies of each TR-DNA interaction for the three enhancers. We use Scl+19 as an illustrative example. Figure 1 shows the Scl+19 enhancer and the fold expression enhancement for the reporter construct in the presence of the wildtype (wt) enhancer and three mutant enhancers: Mutant enhancer 1 (mut1) -Fli1 binding site deleted, Mutant enhancer 2 (mut2) -Gata2 binding site deleted, Mutant enhancer 3 (mut3) - all binding sites deleted. The transcription rates are normalized with the expression rate of the reporter when all enhancer binding sites have been deleted. We assume that the lacZ reporter transcription rates are proportional to the fluorescence intensities measured in these experiments because all experiments were performed in the presence of excess fluorescent substrate and wild-type and mutant constructs were assayed at the same time using the same reagents. Moreover the experimental conditions were controlled to ensure that the proportionality constants that relate various transcription rates to the fluorescent intensities are the same for different experiments. Note that the experimental results were obtained in HSCs which show high expression levels of Scl, Gata2 and Fli1 [20]. Notch and Bmp4 signals are expected to be absent in these cells [51]. We accordingly exclude all Notch and Bmp4 states from our partition functions. We can see from equations (19) – (21) that and are the probabilities of the Scl+19, Gata2-3 and Fli1+12 enhancers being in open state in the absence of Notch and Bmp4. The introduction of the mutant enhancer reporter construct is not expected to affect the growth rate or availability of RNA polymerases in a significant manner. Thus is unaffected by the deletion of binding sites. However the deletion of Fli1 binding sites eliminates the Fli1 bound state in the enhancer partition function in equation (15). Therefore is affected by deletion of binding sites. Since, using equations (22) – (24) we can relate the fold enhancement in gene expression to the free energies of TR-DNA interaction. (22) (23) (24) Equations (22) and (23) can be solved analytically for and as functions of and the concentrations and. (25) (26) The solution for is dependent on and. Using (25) and (26) we can solve for and reduce it to a function of only, and. (27) We apply this recursive procedure to uniquely determine in a similar fashion all free energies of Gata2-3 and Fli1+12 enhancers. The full equations for all free energies are presented in Text S3 (see Eqs. (S. 1) – (S. 11) ). Since mRNA is labile relative to stable cellular proteins, we assume that the mRNA concentration for the triad proteins is at steady state. We can thus directly relate the rate of transcription to the rate of production of the proteins (28) (here represents the number of protein molecules produced per mRNA lifetime). The ODEs for change in protein concentration can be written as a balance between the rate of production and the degradation/dilution rates that are linear in protein concentration (cf. Eq (4) ). The major hurdle in the analysis of this ODE model is the determination of all TR-enhancer interaction free energies from the equations described above (25) – (27) and in the supplement (S. 1–S. 11 in Text S3). The free energies can be determined from these relations if the concentrations of Scl, Gata2 and Fli1 in the wildtype cells and the constants are known. However the TR concentrations are difficult to measure in vivo. We make our equations dimensionless to avoid the measurement of actual Scl, Gata2 and Fli1 concentrations. We normalize these TR concentrations by their wildtype concentrations. In wildtype HSCs the Scl, Gata2 and Fli1 concentrations are at steady state. Let these steady state wildtype concentrations be and. Normalizing Scl, Gata2 and Fli1 concentrations with and we can rewrite equation (4) as a system of ODEs in dimensionless variables and (cf. Eq (6) ). Rates for all three enhancers as given by equations (19) – (21) can be recalculated in terms of the dimensionless variables by adjusting the free energies of each state with the appropriate concentrations. For example, . Note from equations (25) – (27) that the adjusted free energies are not functions of the steady state concentrations of Scl, Gata2 and Fli1. (29) Dimensionless rates and the wild-type, steady state dimensionless rates of transcription can be evaluated from the expressions in (29) by using adjusted free energies and. Then, and. The parameter space of free energies can now easily be explored by tuning. Since the free energies can be determined by fixing, we can also analyze the system response to Notch and Bmp4 by substituting the full expressions of and in equation (6). and represent the strength of the interaction between RNA polymerase and Notch and Bmp4 respectively. Notch and Bmp4 increase Gata2 expression in wildtype HSCs by 3. 5 [26] and 4 fold [27] respectively. At saturating concentrations of Notch (high). This implies. And similarly, . Thus are the only unknown parameters in our model. The model offers both a quantitative means of analysis of combinatorial regulation of gene expression by TRs and a succinct mathematical description of the biophysics of the regulation. The model can easily be extended to regulation involving repressors and many other situations. The reduced model where Scl is not under regulation by Gata2 or Fli1 represents a simplification of this system where the concentration of Scl is kept constant. The reduced system then comprises only the equations for Gata2 and Fli1. The time normalization is carried out relative to the Scl half life (∼) [43]. Gata2 and Fli1 have half-lives of ∼10 minutes and 2 hours respectively [44]. Accordingly, and. Our method for estimation of binding affinities reduces the number of unknown parameters in the system to three chromatin rewrapping equilibrium constants. These constants have been reported to be in the range 10–10000 [52]. We find that for irreversible bistable behavior with switchability our parameter estimation scheme restricts two of these equilibrium constants to a narrow range. We chose the following values for the equilibrium constants: , and from within the ranges. The free energy values are thereafter calculated as described above to complete the parameter set for the triad model (cf. Table S3). The same parameter values are retained for the reduced model, however the Scl concentration in this case is fixed such that the threshold concentration (the concentration at the bifurcation point) of Notch/Bmp4 is identical for both the full triad and reduced model. The system of equations for the triad described in the previous section was analyzed using a number of numerical methods and tools. The steady state characterization of both the triad and reduced modules was carried out using XPPAUT and the associated bifurcation analysis package AUTO [53]. Parameter sensitivity analysis for the chromatin equilibrium constants was also done with AUTO. The analysis of the dynamics of the ODE model was carried out using the ODE45 solver of MATLAB 2008a (R) (The MathWorks, Natick, Massachusetts). To compute the minimum pulse duration for Notch/Bmp4 signals, the integration was initiated at the low steady state and a step input of Notch/Bmp4 was introduced. The pulse duration to switch the system was minimized using the function (Optimization toolbox) in MATLAB. In all simulations only the dimensionless models were used.

Hematopoiesis—blood cell development—has long served as a model for study of cellular differentiation and its control by underlying gene regulatory networks. The Scl-Gata2-Fli1 triad is a network module essential for the development of hematopoietic stem cells but its mechanistic role is not well understood. The transcription factors Scl, Gata2 and Fli1 act in combination to upregulate transcription of each other via distal enhancer site binding. Similar network architectures are essential in other multipotent cell lines. We propose a method that uses experimental results to circumvent the difficulties of mathematically modeling the combinatorial regulation of this triad module. Using this dynamical model we show that the triad exhibits robust bistable behavior. Environmental signals can irreversibly switch the triad between stable states in a manner that reflects the unidirectional switching in the formation and subsequent differentiation of hematopoietic stem cells. We also show that the triad makes reliable decisions in noisy environments by only switching in response to transient signals that persist longer than the threshold duration. These results suggest that the Scl-Gata2-Fli1 module possibly functions as a control switch for hematopoietic stem cell development. The proposed method can be extended for quantitative characterization of other combinatorial gene regulatory modules.

Abstract Introduction Results Discussion Methods

hematology/hematopoiesis developmental biology/stem cells computational biology/transcriptional regulation

2010

Modeling Reveals Bistability and Low-Pass Filtering in the Network Module Determining Blood Stem Cell Fate

The intracellular pathogen Legionella pneumophila hijacks the endoplasmic reticulum (ER) -derived vesicles to create an organelle designated Legionella-containing vacuole (LCV) required for bacterial replication. Maturation of the LCV involved acquisition of Rab1, which is mediated by the bacterial effector protein SidM/DrrA. SidM/DrrA is a bifunctional enzyme having the activity of both Rab1-specific GDP dissociation inhibitor (GDI) displacement factor (GDF) and guanine nucleotide exchange factor (GEF). LidA, another Rab1-interacting bacterial effector protein, was reported to promote SidM/DrrA-mediated recruitment of Rab1 to the LCV as well. Here we report the crystal structures of LidA complexes with GDP- and GTP-bound Rab1 respectively. Structural comparison revealed that GDP-Rab1 bound by LidA exhibits an active and nearly identical conformation with that of GTP-Rab1, suggesting that LidA can disrupt the switch function of Rab1 and render it persistently active. As with GTP, LidA maintains GDP-Rab1 in the active conformation through interaction with its two conserved switch regions. Consistent with the structural observations, biochemical assays showed that LidA binds to GDP- and GTP-Rab1 equally well with an affinity approximately 7. 5 nM. We propose that the tight interaction with Rab1 allows LidA to facilitate SidM/DrrA-catalyzed release of Rab1 from GDIs. Taken together, our results support a unique mechanism by which a bacterial effector protein regulates Rab1 recycling. Rab GTPases play a crucial role in vesicular trafficking through shuttling between cytosol and membranes, a process that is controlled by several regulatory proteins [1]–[4]. GDP dissociation inhibitors (GDIs) preferentially interact with and deliver GDP-bound Rabs to their target membranes, where dissociation of the GDI-Rab complexes is catalyzed by GDI displacement factors (GDFs) [5]–[8]. Prenylation at the C-termini of Rab proteins is essential for their membrane association [9]–[11]. The membrane-localized Rabs are subsequently activated by specific guanine nucleotide exchange factors (GEFs) via promoting their exchange of GDP for GTP [12]–[15]. The activated Rabs then bind their cognate effectors, triggering signaling for vesicle formation [16], [17], vesicle transport [18]–[23], vesicle tethering and fusion of vesicles with target membranes [17], [24]–[27]. GTPase activating proteins (GAPs) catalyze hydrolysis of GTP in the activated Rabs and return them to the GDP-bound inactive form that is sensitive to membrane retrieval by GDIs, thus replenishing the cytoplasmic pool of Rab proteins [4], [14], [15]. The intracellular bacterial pathogen Legionella pneumophila (L. pneumophila) is the causative agent of pneumonia Legionnaires disease [8], [28]. Following invasion of host cells, L. pneumophila resides in the LCV [29] that escapes endolysosomal destruction [30]. The bacterial effector proteins, delivered by the type IV secretion system (T4SS) of L. pneumophila into the cytosol of host cells [31], actively remodel the LCV to establish an intracellular niche indispensable to bacterial pathogenesis [32]. For example, the early secretory vesicles from ER can be hijacked to the LCV, converting it into an ER-derived organelle that supports bacterial replication [33]–[36]. Rab1, required for vesicle trafficking between ER and the Golgi complex [18], [37], [38], is one of the host proteins recruited to the LCV shortly after uptake of L. pneumophila [39]–[41]. Rab1 recruitment subsequently promotes transport and fusion of ER-derived vesicles with the LCV, thus playing an essential role in the biogenesis of this organelle [42], [43]. The effector protein SidM/DrrA, by acting as a Rab1-sepcific GEF and GDF, is required for LCV recruitment of Rab1 [44], [45]. Further, the AMPylation activity of SidM/DrrA modified Rab1 by covalently adding adenosine monophosphate (AMP) to avoid the GAP recognition [46]. SidD, functions as the Rab1 deAMPylase, generating de-AMPylated Rab1 accessible for inactivation by LepB [43], [47]. Another translocated effector protein LidA is also involved in the recruitment of early secretory vesicles to the LCV [48], [49]. LidA was found to interact with Rab1 as well, regardless of nucleotide binding states, and promote SidM/DrrA activity of transporting Rab1 to the LCV [39], [45]. Unlike SidM/DrrA mutants, however, L. pneumophila mutants lacking LidA displayed a temporal delay but not a loss of Rab1 recruitment to the LCV [39]. Intriguingly, while indispensable to the recruitment of wild type Rab1, SidM/DrrA is not required to accumulate Rab1 mutant (D44N) that loses interaction with GDIs but not with LidA about the LCV [45]. These results suggest that GDIs play a negative role in Rab1 recruitment by the LCV during L. pneumophila infection. Recruitment of this Rab1 mutant, however, is dependent on LidA, suggesting that interaction with Rab1 is important for the role of LidA in delivering Rab1 to the LCV. Currently the mechanisms of how LidA cooperates with SidM/DrrA for Rab1 recruitment are not well understood. Here, we present the crystal structures of LidA (residues 224-559) in complex with a GDP-bound Rab1 mutant (S25N; residues 1-176) [50], [51] and LidA (residues 188-449) in complex with GTP-bound Rab1 (residues 1-176). Unexpectedly, the structures showed that GDP-Rab1 (S25N), a “constitutively” inactive mutant, adopted an active conformation when bound by LidA. In agreement with the structural observation, biochemical assays demonstrated that GDP-Rab1 (S25N) and GTP-Rab1 (Q70L) [52] exhibited a similar and exceptionally high binding affinity for LidA. Coupled with previous observations, data presented in current study support a unique mechanism by which LidA interferes with the host secretory vesicular trafficking. The Rab1-interacting domain of LidA used in our study is similar to the one (residues 191-600) necessary and sufficient to disrupt the secretory pathway when overexpressed in COS1 cells [49]. Consistently, In vitro data shows truncation of 188 residues from the N-terminal side or 155 residues from the C-terminal side of LidA generated no effect on the formation of such a complex (Figure S1). To further validate the interaction between LidA and Rab1, we measured their binding affinity using isothermal titration calorimetry (ITC) technique. The ITC results showed that the LidA fragment 188-580 containing the predicted coiled-coil domain interacted with the constitutively active mutant Rab1 (Q70L) with a dissociation constant of 7. 5 nM (Figure 1A). Surprisingly, the same protein also bound nearly equally well to the constitutively GDP-bound mutant of Rab1 (S25N), with a dissociation constant of 7. 6 nM (Figure 1B). A preference of LidA for the GTP-Rab1 could result from a lower abundance of GDP-Rab1 in cells. To reveal the molecular basis for the LidA-Rab1 interaction, we first solved the crystal structure of LidA (224-559) in complex with a constitutively GDP-bound mutant Rab1 (S25N; 1-176) at 1. 73 Å using molecular replacement (Table S1 in supporting information). The overall LidA-Rab1 complex adopts a compact and globular structure (Figures 2A, B). The interaction between LidA and Rab1 resulted in a 1∶1 stoichiometric complex and buried 41% (4031 Å2/9845 Å2) of the Rab1. The exceptionally large buried surface generated by LidA-Rab1 interaction is consistent with their strong binding affinity. In the structure, Rab1 is held in a large and pronounced groove made by the four “fingers” with an extensive, though not complete, charge and surface complementarities. Contacts between LidA and Rab1 are established through a combination of hydrogen bonds and hydrophobic interactions involving the switch I, switch II and the interswitch region of Rab1 (Figure 2B). LidA (residues 224-559) is composed of seven α-helices and ten β-strands, which arrange into a structure resembling a hand with the baby finger buried in the palm and the remaining four straightened (Figure 2C). Six of the seven α-helices form three extended two-stranded α-helical coiled-coils. This structural observation supports a previous prediction that LidA is a coiled-coil rich structure [48], [49]. Two of them, made by α4/α5 and α1/α7, are the “index finger” and the “middle finger”, respectively. Interaction between them is primarily mediated by packing of α1 against α5. The “thumb” stems from a two-helix coiled-coil formed by α2 and α3 helices which interact with the base of “index finger” through packing against α4. The isolated helix α6 and its following loop make extensive hydrophobic contacts with three anti-parallel β-sheets (β5/β6, β7/β8 and β9/β10), resulting in formation of the “ring finger”. The N-terminal side of α4, which is not involved in the formation of coiled-coil structures, tightly contacts the two anti-parallel β-sheets (β1/β2 and β3/β4) located between α3 and α4, thus making the “wrist”. Interestingly, the α1 and α7 stay closely together and form the middle finger. The terminal parallel helix of LidA extends to the outside of interacting region, forms an antiparallel long coiled-coil (Figure 2C). So we suppose the N- and C-terminal domains of LidA may stay spatially close cooperate with each other to perform specific function. In the complex, switch I (residues 33-48) of Rab1 is mainly sandwiched between the “middle finger” and the “ring finger” by making contacts with α1, α7 and α6. It also contacts the “index finger” α5 (Figures 3A, B). Close packing of Ile44Rab1 from switch I and Phe73Rab1 from switch II (residues 65-83) against Leu541LidA, Val542LidA, Val538LidA, Leu436LidA, Ala439LidA, Y243LidA and the aliphatic portion of Asn432LidA (from α5) appears to dominate the interactions around this interface. Additional van der Waals interactions result from contacts of Tyr532LidA with the alpha carbon atom of Gly21Rab1 and Tyr40Rab1, and Ile41Rab1 with Leu548LidA and Glu549LidA in LidA. Hydrogen bonds further strengthen this interface. In particular, in addition to forming a salt bridge with Asp443LidA, Arg72Rab1 from the switch II region makes a hydrogen bonds with the backbone carbonyl of Met536LidA. Through an intramolecular hydrogen bond, Arg72Rab1 is stabilized by the catalytically important residue Gln70Rab1, which in turn mediates a bifurcated polar interaction with the carbonyl oxygen of Thr534LidA and Glu533LidA from LidA. The C-terminal portion of the switch II, together with the region defined by β2, β3 and β4, makes extensive contacts with the base of “index finger” (Figures 3A, C). Center to this interface are the van der Waals interactions made by Phe48Rab1 and Trp65Rab1 from Rab1 with a cluster of surrounding hydrophobic residues of LidA. At periphery, a number of hydrogen bonds flank the hydrophobic interaction center. Among these, Lys61Rab1 from Rab1 forms three hydrogen bonds with Asn421LidA and the carbonyl oxygen atoms of Ile413LidA and Ser415LidA. Other hydrogen bonds include those formed by Tyr8Rab1 with the carbonyl oxygen atom of Met414LidA, Gln63Rab1 with Asn410LidA, the carbonyl oxygen atoms of Val46Rab1, ILE44Rab1 and Phe73Rab1 with Asn432LidA and Asn435LidA, respectively. In addition to hydrogen bonding with His431LidA, Tyr80Rab1 also engages hydrophobic contact with Leu406LidA. Our biochemical assay (Figures 1A, B) indicates that LidA binding is independent of the Rab1 nucleotide-binding states. To understand the underlying molecular mechanisms, we solved the crystal structure of LidA (residues 188-449) in complex with the Rab1 (WT; residues 1-191) at 2. 2 Å. The structure of the complex is highly similar to that of LidA (224-559) -Rab1 (S25N) with an r. m. s. d of 0. 678 Å for Rab1; 1. 496 Å for LidA, respectively (Figure 4A). As anticipated, GTP is well defined in the crystal structure (Figure 4A) and its interactions with Rab1 are conserved in other active Rabs (Figure S2) [53]–[55]. Surprisingly, structural comparison revealed that the conformation of LidA-bound Rab1 (S25N) is essentially identical with that of LidA-bound GTP-Rab1 (Figure 4B), indicating that Rab1 (S25N) is in the active conformation following LidA binding. While the switch II region in the free GDP-Rab1 is completely disordered (Figure 4C), it is well defined when bound by LidA. As with GTP, LidA binding induces striking structural remodeling around the switch I region of Rab1 (S25N), which swings from the edge of β-sheet to the nucleotide-binding site, with the largest displacement of 9. 2 Å at the Cα atom of Ile. Relocation of switch I results in formation of an extended β-strand β2 at its N-terminal side and hydrophobic packing against switch II (Figure 4C). Our structural analyses suggest that LidA utilizes a similar mechanism to GTP for maintaining Rab1 in the active conformation. The hydrogen bonding interactions between the γ phosphate group of GTP with Thr43 and Gly69, from switch I and switch II of Rab1 respectively, are highly conserved among Rab proteins and important for stabilizing their active state (Figure S2). Despite loss of the γ phosphate group-mediated interactions, the switch I region in Rab1 (S25N) is stabilized through its interaction with α7 of LidA (Figure 4D). Stabilization of the switch II region in Rab1 (S25N) is via its polar interaction with the loop linking α6 and α7 as well as hydrophobic contacts with α5 of LidA (Figure 4E). Deletion of the switch-stabilizing structural elements would be expected to disfavor LidA interaction with the GDP-bound Rab1 as compared to the GTP-bound Rab1. We therefore quantified LidA (residues 188-449) interaction with Rab1 (S25N) and Rab1 (Q70L) using an ITC assay. In support of the structure-based prediction, the assay showed that the LidA fragment exhibited a higher binding affinity (0. 43 µM) to the GTP-bound Rab1 than to the GDP-bound Rab1 (2. 4 µM) (Figure 4F). Interaction of the anti-parallel coiled-coil formed by the helices α4 and α5 with the switch and interswitch regions of Rab1 is reminiscent of the structure of Rab4-Rabenosyn-5 complex [54]. A superposition of the overall structures of these two complexes revealed significant structural similarity (Figure 5A). Some subtle structural differences surrounding the switch I and II are consistent with the notion that the switch regions adopt specific conformations in the GTP-binding form of Rab proteins [55]. Rabenosyn-5 forms a similar anti-parallel coiled-coil to LidA and binds the switch and interswitch regions of Rab4. Interestingly, although these two peptides have reverse orientations, they contact a significant subset of the highly conserved resides between Rab1 and Rab4, though the detailed molecular interactions differ. For example, in both complexes the highly conserved residues Phe48Rab1 (Phe45Rab4) and Trp65Rab1 (Trp62Rab4) are buried in a cluster of hydrophobic residues despite their different side chain rotamer conformations, whereas Lys61Rab1 (Lys58Rab4) and Tyr80Rab1 (Tyr77Rab4) are involved in polar interactions (Figures 5B, C). In addition to Rab1, LidA has been shown to be able to interact with other Rab member proteins such as Rab6a and Rab8b [39]. Indeed, many LidA-interacting residues in Rab1 are highly conserved in these two Rabs as well as other Rab members (Figure 6A) suggesting that LidA may have promiscuity of binding Rab proteins. To further experimentally test this, we purified 15 GST-fused Rab family members (His-fused Rab20 is exceptional) and tested their interaction with LidA using pull down assay. As shown in Figure 6B, thirteen out of the fifteen Rabs tested interacted with LidA. The binding affinities of LidA for other Rabs, including Rab2, Rab4, Rab6, Rab7, Rab9, Rab11, Rab20 and Rab22 have been quantified by ITC. All give KD values in the micromolar range. Compared with them Rab1 gives the strongest binding affinity (The KD is in nanomolar range). Thus Rab1 should be the preferred substrate of LidA compared with other Rabs (Figure 7). Thus, although previous LCV analysis identified interacting Rab GTPases such as Rab1, Rab7, Rab8, and Rab14 as novel LCV components [56], the biological relevance of the interactions await further investigation. To make clear how LidA choose substrate from so many Rabs, we attempted to disrupt the interactions of Rab1 and LidA by making single or double mutations of Rab1 and LidA along the surface of the interactions. However, none of the mutants we have tested by the pull-down assay lose their interaction with LidA. We also tested such mutants in other Rabs and did not observe an effect. So it remains undetermined how LidA selects different Rabs. Nonetheless, it is expected that the mechanism of Rab1 recognition by LidA is applicable to other Rab members given the high conservation in the LidA-interacting surface. The crystal structures presented in current study revealed that the GDP-bound Rab1 is held in the active conformation through associating with LidA. Consistent with this structural observation, quantification assay using ITC demonstrated that LidA binds strongly to both GDP- and GTP-Rab1 with a nearly equal affinity (Figures 1A, B). These results indicate that Rab1 recognition by LidA is independent of its cycling between GDP- and GTP-bound states. To LidA, the switch mechanisms of Rab1 are not operational and thus Rab1 is rendered persistently active in interaction with LidA. Such a unique function of LidA provides explanations for a body of previous observations. The prenylated GDI-free Rab1 (D44N) was shown to be primarily membrane-localized, indicating that GDI is not absolutely required for membrane targeting of Rab1 [57]. This principle is likely applicable to the delivery of Rab1 (D44N) to the LCV [45]. The strong interaction with LidA can be important for the membrane retention of Rab1 (D44N), as LidA-deficient mutants displayed a scattered distribution of this Rab1 mutant in COS-1 cells [45]. Because no GDP-Rab1 (D44N) -GDI complex is formed and LidA is able to recognize GDP-Rab1, GDF and GEF activity would be made redundant under these conditions, providing an explanation why recruitment of Rab1 (D44N) to LCV is independent of SidM/DrrA [45]. To wild type Rab1, the complex of GDP-Rab1-GDI is formed with a high affinity. The direct competition of LidA with GDI for binding GDP-Rab1 will be difficult and inefficient. Thus GDP-Rab1 has to be released from GDI by SidM/DrrA for LidA recognition [39]. These arguments likely hold true with the recruitment of the GDP-restricted Rab1 mutant (S25N) to the LCV. In uninfected cells, the dominant inhibitory effect of Rab1 (S25N) derives from the competition of this Rab1 mutant with wild type Rab1 for binding GEFs, resulting in formation of “dead-end” Rab1-GEF complexes [58]. In the case of wild type Rab1 recruitment to the LCV, this situation could be altered, because interaction of GDP-bound Rab1-SidM/DrrA complex is transient and much weaker than that of GDP-bound Rab1-LidA complex [39], [45] (Figure S3). LidA is therefore expected to be able to outcompete with SidM/DrrA for binding GDP-Rab1, which in turn facilitates the release of GDP-Rab1 from GDF (see further discussion below). As GDP-Rab1 is in the active conformation when complexed with LidA, it is conceivable that Rab1 (S25N) can still support the functions performed by wild type Rab1. In complete agreement with this hypothesis, Rab1 binding to the LCV precedes the delivery of Sec22b, GDP-locked Rab1 (S25N) can delay but not block the recruitment of Sec22b. This Rab1 mutant can support growth of L. pneumophila, at least for the first 10 hours of postinfection [42]. The strong association with LidA can facilitate the recruitment of Rab1 to LCV by SidM/DrrA. Release of GDP-Rab1 from GDIs catalyzed by SidM/DrrA (as a GDF) is a dynamic process and GDP-Rab1-SidM/DrrA is an intermediate of the reaction [59]–[61]. LidA is expected to displace SidM/DrrA from the intermediate because of its much higher binding affinity to GDP-Rab1, thus shifting the equilibrium toward dissociation of GDP-Rab1-GDI complex. In this respect, LidA is similar to GTP in that the latter is able to exchange GDP from the intermediate and form the GTP-bound Rab1, a product that is unfavorable for interaction with GDFs and thereby promotes disruption of GDP-Rab1-GDI complex. It can be imagined that lack of LidA would slow down the release of GDP-Rab1 from GDI. Consistently, deletion of LidA resulted in delayed but not a blocked recruitment of Rab1 to the LCV [39]. The inability of LidA to discriminate between GDP- and GTP-bound Rab1 may also be advantageous for L. pneumophila to retain the recruited Rab1 on the LCV. Because even GTP-bound Rab1 is subjected to hydrolysis by GAPs on LCV [43], [44], the resulting product of GDP-Rab1 is still able to be captured by LidA with a strong binding affinity. Moreover, interaction with LidA would make GDP-Rab1 less susceptible to extraction from the membrane of LCV by GDIs, because they have to outcompete LidA for Rab1 binding. Unlike other small GTPase effector proteins which mainly binding to Rab1 through interacting with the switch I and II regions that determine nucleotide-dependent interaction, LidA holds Rab1 tightly in its hand like structure with four fingers. It is unlikely that any Rab1 effector can displace LidA from Rab1-LidA complex. LidA has been known to associate with the LCVs for much longer time than SidM/DrrA. However, Rab1 does eventually leave the LCV [44], [48], so that a mechanism is needed to decrease the binding affinity between Rab1 and LidA. Since Rab1 is deeply buried in the complex, the most efficient way to release Rab1 is to force LidA to open its fingers by downstream proteins. The GTP-Rab1-LidA structure shows truncation of the middle and the ring fingers does not disrupt Rab1-LidA complex, so that the thumb or the index fingers, though necessary, may not be sufficient to release Rab1 (Figure 4A). Structure analysis indicates that the thumb and the ring finger do not pack as tightly with other part of LidA as the index and middle finger. This means that LidA' s wrist (the N-terminal half of α4 and β1–β4) which is isolated from the central interacting region, the bottom of the ring finger and thumb are accessible for other proteins to trigger a conformational change that releases Rab1 (Figure S4). Opening of the ring finger could be very important since it can expose the switch region immediately to facilitate Rab1 binding by other effectors. Recent research showed that the post-translationally modified Rab1b by the L. pneumophila effector protein SidM/DrrA retains the ability to interact with LidA and can avoid the GAP recognition [46]. Our data confirmed that the correspondent residue Tyr80 in Rab1a was also AMPylated (Figure S5A). Tyr80 is located on the interface between Rab1a and LidA, involved in both hydrogen bonding with His431LidA and hydrophobic contact with Leu406LidA (Figure 3C), so it is reasonable to speculate that the covalent-bound AMP group on Tyr80 will affect the stability of LidA-Rab1 complex. However, Rab1 interacts with the LidA palm tightly via an extensive surface, so that the interaction is difficult to interrupt. In the structure of GTP-Rab1-LidA we can see even the middle and the ring fingers are truncated, LidA can still holds Rab1 (Figure 4A). Moreover the structural feature of four fingers possesses a certain degree of flexibility. Thus the orientations of the four fingers could adjust a little bit to accommodate the additional AMP group. Consistent with previous reports [46] our ITC results showed that the binding affinity of Rab1a (K62H, 1-176) and LidA (188-580) is not significantly affected by the AMPylation (Figure S5B, C). Thus we presume that LidA can form both the Rab1-LidA and AMP-Rab1-LidA complex in vivo. The ability of LidA to form stable complex with AMPylated Rab1 is consistent with its ability to bind a number of different Rabs, which small differences in their surface residues. L. pneumophila may also benefit from this property. Formation of AMPylated-Rab1-LidA complex may protect Rab1 from the de-AMPylation activity of SidD until Rab1 is released from LidA by still unknown mechanisms. In addition to promoting SidM/DrrA for Rab1 recruitment, LidA may have other function (s). LidA interacts with the active conformation of Rab1 and exhibits no enzymatic activity [39]. Structural comparison showed that binding of LidA to Rab1 shares some features of the interaction between the effector protein Rabenosyn-5 with Rab4 [54]. Furthermore, recruitment of Rab1 is essential for delivering ER-derived vesicles to LCV [42], [43], [62]. Collectively, these results suggest that one potential function of LidA is to act as an effector of Rab1 and signal downstream components for remodeling of LCV. LidA was previously predicted and further confirmed in the present study to be a coiled-coil rich protein, one type of the best characterized Rab tethering factors [27], [54], [63]. The terminal parallel helix forms an antiparallel long coiled-coil and extends to the outside of interacting region (Figure 2C). Secondary structure analysis suggests that in full-length LidA, the length of this domain is likely to be longer than 100 Å. It is thus possible that LidA may function as a tethering factor, bridging the ER-derived vesicles and the LCV [49]. One potential advantage of employing its own protein as a tethering factor would allow L. pneumophila to be selective for recruiting ER-derived components for its growth. It is well established that a GTP-bound form of Rab GTPases interacts with their effectors for signaling. The only exception to this rule is Protrudin that interacts with the GDP-bound form of Rab11, regulating Rab11-dependent membrane recycling to promote the directional membrane trafficking [64]. While interacting with GDP-Rab1, LidA appears to be different from protrudin in that it essentially recognizes the active conformation of Rab1 and probably other Rabs [39], [43]. During the peer review of this manuscript, the structure of Rab8-LidA complex was published online. Structure comparison shows that Rab8-LidA complex adopts quite similar conformations with our structure and that the binding interface is highly conserved, consistent with our proposal that lidA recognizes Rabs in a similar way. We note that the authors report a binding affinity of LidA for GTP-bound Rab1 that is roughly one order of magnitude stronger than what we measured using ITC. At this point, it is inconsistent with our ITC data that LidA binds to both GDP- and GTP-Rab1 with a nearly equal affinity. The genes of lidA and sidM were amplified from genomic DNA of Legionella pneumophila strain Corby. The DNA fragment encoding human Rab1a (noted Rab1 throughout the text) and other 14 kinds of Rab family members were amplified from a homemade human cDNA library. All the genes encoding the mutant proteins were produced using a two-step PCR procedure. LidA, both full-length (FL) and various fragments, SidM variants include 317-545 and 1-545 were subcloned into the pET15b (Invitrogen), Rab1 variants and Rab2,4, 6,7, 8,11,14,18,23,27 were subcloned into the pGEX-6P-1 plasmid (GE Healthcare), Rab5,9, 20,22 were subcloned into both of the two vectors. Each protein was produced in E. coli BL21 (DE3). Cells grow at 37°C until the OD600 reach to 0. 8. LidA proteins were induced with 0. 5 mM isopropyl- β-D-Thiogalactopyranoside (IPTG) for 15 h at 15°C, Rab1 and other Rabs were all induced with 0. 2 mM IPTG for 15 h at 22°C. All proteins were purified using affinity, anion exchange and gel filtration chromatography. To obtain LidA-Rab1 complexes, LidA and Rab1 proteins were mixed up with a 1∶1 molar ratio at 4°C for 2 hours. The LidA (188-449) -Rab1 (1-191) complex and LidA (224-559) -Rab1 (S25N; 1-176) complex were further purified using the above mentioned procedure except the GST tag and His tag were removed by homemade PreScission protease digestion before the anion exchange step. Both of the protein complexes have a final concentration of 10 mg/mL. All purification processes were performed at 4°C unless noted otherwise. Crystallization conditions for complexes were determined from the sparse matrix screen (Hampton Research). All crystals were obtained using the hanging drop diffusion method at 20°C. LidA (224-559) -Rab1 (S25N, 1-176) complex was crystallized by mixing equal volumes of protein with reservoir solution containing 28% Jeffamin ED-2001 and 0. 1 M sodium citrate tribasic dihydrate pH 4. 8. LidA (224-559) -Rab1 (S25N, 1-176) crystals were optimized by microseeding to reach the satisfied diffraction. LidA (188-449) -Rab1 (wild-type, WT1-191) complex was crystallized by mixing equal volumes of protein with reservoir solution containing 0. 1 M Hepes pH 7. 5,25% PEG3350 and 0. 7% butanol. All diffraction data were collected at Shanghai Synchrotron Radiation facility (SSRF) beamline BL17U. To prevent radiation damage, crystals were equilibrated in a cryoprotectant buffer containing 20% ethylene glycol (v/v) plus reservoir buffer and then flash frozen in a 100 K nitrogen stream. The best crystal of LidA (225-559) -Rab1 (S25N; 1-176) complex diffracted to 1. 73 Å. The best crystal of LidA (188-449) -Rab1 (WT; 1-191) complex diffracted to 2. 2 Å. Data sets were processed using the HKL2000 software suite [65]. The crystal structures of the two kinds of LidA-Rab1 complexes were determined by molecular replacement using PHASER [66] with the coordinates of Rab1 as the searching models (PDB ID code 2FOL) [67]. The atomic models were built using COOT [68] and refined using PHENIX [69]. Data collection and structure refinement statistics are summarized in Table S1. A SA-composite omit map of LidA index finger shows the good match between the atom skeleton and the electron density map as supplementary Figure S6. All the molecular graphics Figures were generated using PyMol (http: //www. pymol. org). ITC was employed to measure the binding affinities of various fragments of LidA with Rab1 variants and 9 kinds of Rabs. All protein samples were purified in a buffer containing 20 mM Hepes (pH 8. 0) and 100 mM NaCl with tag removed by PreScission protease digestion. The final concentration of LidA (188-580) and LidA (188-449) were 0. 2 mM, Rab1 (S25N) and Rab1 (Q70L) were 2 mM, LidA (FL) was 0. 15 mM, Rabs (FL) (include 2,4, 6,7, 9,11,20,22) were 1. 5 mM. The samples were centrifuged to remove any precipitate before the experiments. All measurements were carried out at 25°C by using a VP-ITC microcalorimeter 200 (MicroCal). Titrations were carried out by titrating Rab1 (S25N) and Rab1 (Q70L) into LidA (188-580) or LidA (188-449), respectively. Rab2,4, 6,7, 9,11,20 proteins were titrated into LidA (188-580). Rab22 were titrated into LidA (FL). The titration Data were analyzed using ORIGIN data analysis software (MicroCalSoftware). 15 kinds of purified Rab family members (WT; FL) with N-terminal GST tag (His-fused Rab20 is exceptional) was immobilized onto 200 µL glutathione-Sepharose resin (GE) or Ni-chelating sepharose (GE healthcare) for His-Rab20. The resin was then washed three times to remove excess unbound protein. Untagged LidA (FL) protein solution was loaded onto the Rabs-immobilized beads and incubated at 4°C for 2 h. The loaded protein dose of LidA and Rabs was at a molar ratio of 2∶1. The resin was washed three times using the wash buffer (25 mM Tris-HCl buffer pH 8. 0,100 mM NaCl) to remove unbound LidA. Proteins binding to the resin were eluted by elution buffer (25 mM Tris-HCl buffer pH 8. 0,100 mM NaCl, 3 mM DTT, 5 mM GSH for glutathione-Sepharose resin and 25 mM Tris-HCl buffer pH 8. 0,100 mM NaCl, 250 mM imidazole for Ni-chelating sepharose). All samples were subjected to SDS-PAGE which was visualized by Coomassie Brilliant Blue staining. Crystal structure of LidA (224-559) -Rab1 (S25N; 1-176) complex and the structure factor have been deposited in the Protein Data Bank (http: //www. rcsb. org/pdb) under ID codes 3SFV. Crystal structure of LidA (188-449) -Rab1 (Q70L; 1-191) complex and the structure factor are deposited with access codes 3TKL.

Legionella pneumophila delivers 275 validated substrates into the host cytosol by its Dot/Icm type IV secretion system. Several substrates including SidM/DrrA and LidA directly interact with the host Rab GTPases and interfere with the vesicle secretion pathway. SidM/DrrA is necessary for Rab1 recruitment, function as a Rab1 specific GDI displacement factor and guanine nucleotide exchange factor. LidA has the auxiliary activity for Rab1 recruitment, whereas it is more important for the formation of the replication vacuole compared with SidM/DrrA. LidA is predicted to be the first substrate secreted by the Dot/Icm system and is critical for maintaining the integrity of the bacterial cell. Moreover, it expresses throughout the intracellular growth phase, localizes to early secretory compartments, and interacts with several members of Rab family. Here we present the crystal structures of LidA coiled-coil domain in complex with two different states of Rab1, GDP- and GTP-bound. The GDP-bound Rab1 in the complex surprisingly has the same conformation with the GTP-bound Rab1, revealing that LidA can retain Rab1 persistently in its active state. Our structures add a new insight into the regulation of the host Rab1 membrane cycle by pathogen-secreted coiled-coil effector.

Abstract Introduction Results Discussion Materials and Methods

biochemistry proteins biology microbiology biophysics bacterial pathogens

2012

Structural Insights into a Unique Legionella pneumophila Effector LidA Recognizing Both GDP and GTP Bound Rab1 in Their Active State

We investigated the spatiotemporal dynamics of HSV genome transport during the initiation of infection using viruses containing bioorthogonal traceable precursors incorporated into their genomes (HSVEdC). In vitro assays revealed a structural alteration in the capsid induced upon HSVEdC binding to solid supports that allowed coupling to external capture agents and demonstrated that the vast majority of individual virions contained bioorthogonally-tagged genomes. Using HSVEdC in vivo we reveal novel aspects of the kinetics, localisation, mechanistic entry requirements and morphological transitions of infecting genomes. Uncoating and nuclear import was observed within 30 min, with genomes in a defined compaction state (ca. 3-fold volume increase from capsids). Free cytosolic uncoated genomes were infrequent (7–10% of the total uncoated genomes), likely a consequence of subpopulations of cells receiving high particle numbers. Uncoated nuclear genomes underwent temporal transitions in condensation state and while ICP4 efficiently associated with condensed foci of initial infecting genomes, this relationship switched away from residual longer lived condensed foci to increasingly decondensed genomes as infection progressed. Inhibition of transcription had no effect on nuclear entry but in the absence of transcription, genomes persisted as tightly condensed foci. Ongoing transcription, in the absence of protein synthesis, revealed a distinct spatial clustering of genomes, which we have termed genome congregation, not seen with non-transcribing genomes. Genomes expanded to more decondensed forms in the absence of DNA replication indicating additional transitional steps. During full progression of infection, genomes decondensed further, with a diffuse low intensity signal dissipated within replication compartments, but frequently with tight foci remaining peripherally, representing unreplicated genomes or condensed parental strands of replicated DNA. Uncoating and nuclear entry was independent of proteasome function and resistant to inhibitors of nuclear export. Together with additional data our results reveal new insight into the spatiotemporal dynamics of HSV genome uncoating, transport and organisation. Virtually all DNA virus classes including herpesviruses, adenoviruses, hepatitis B virus, parvoviruses and polyomaviruses must deposit their genomes within the nucleus for transcription, genome replication and subsequent capsid assembly. Genome transport and entry to the nucleus is also a prerequisite for replication of retroviruses, lentiviruses including HIV and certain RNA viruses including e. g. , orthomyxoviruses such as influenza virus. All these viruses must navigate through the cytoplasm of infected cells, escape or counteract physical host cell barriers and antiviral processes, and engage with the nuclear envelope or nuclear pore for genome import into the nucleus [1–9]. Despite advances in certain areas [10–16] much remains to be understood concerning the detailed pathways and mechanisms involved, particularly with regard to the localisation of infecting virus genomes themselves and their regulated (or premature) presentation to the cell environment. Quantitative spatiotemporal information at the single particle level on localisation, uncoating and transport of the infecting genome is required for any complete understanding of many critical aspects of virus infection and virus pathogenesis. Among the factors which have limited the quantitative spatiotemporal analysis of genome transport and presentation are the insensitivity or ready tractability of methods to directly visualise and measure virus genomes, the inability to differentiate genomes that are encapsidated from those that have dissociated, the inability to readily differentiate incoming from replicated genomes; and the incompatibility of certain detection methods with immunohistochemistry for parallel detection of host and viral protein components. One of the most frequently used techniques, fluorescence in situ hybridisation (FISH), has provided many advances, and yet still presents several hurdles and limitations [16]. FISH inherently cannot discriminate input from replicated genomes nor, due to the harsh conditions frequently incompatible with immunofluorescence, does FISH discriminate between encapsidated genomes versus released genomes. Other routes such as the incorporation of multimerised binding sites for fluorescent DNA binding proteins, e. g. YFP-TetR [17] offer possibilities for live cell imaging but require specialised recombinant viruses or cell lines and can still be highly limited in detecting infecting genomes. Such limitations and issues of sensitivity or tractability have meant that many studies on aspects of virus infection have almost invariably relied on indirect, surrogate measures of detection of virus genomes. Reports, e. g. , on the effect of inhibitors on DNA entry or analysis of DNA sensing have inferred effects on genome localisation from protein localisation. Clearly genomes not bound by surrogate markers will not be detected for any number of reasons including unknown but specific differences between genomes, occlusion by other factors, spatial segregation dictating differential protein association, repression by chromatin, non-nuclear localisations where surrogate markers may not co-localise and many others factors. Such assays also give little information on other aspects of uncoating and the morphological state of infecting genomes. These and other considerations limit our understanding of genome entry, uncoating, nuclear translocation and physical transitions, all of which are necessary for a true understanding of the earliest processes governing virus infection and host responses. In this regard, the development of bioorthogonal metabolic precursors combined with cycloaddition to corresponding capture reagents is increasingly being exploited in various approaches to biological processes and to mechanisms in infection and immunity [16,18–21]. Analysis of DNA synthesis by labelling with alkyne-derivatised nucleosides and cycloaddition to azide-coupled fluorochromes has been evaluated in several systems [20,22,23] and used in spatial, biochemical, and systems approaches to investigate DNA replication and the cell cycle. These techniques have recently been exploited by the Greber laboratory for analysis of adenovirus (Adv) infection [16] using viruses incorporating the alkyne-derivatised nucleosides EdC (ethynyl-deoxycytidine) or EdU (ethynyl-deoxyuridine) in their genomes for the spatiotemporal investigation of genome trafficking at the single particle level. We also recently showed that EdC was efficiently incorporated into HSV replication compartments and that incubation with EdC had no significant effect on HSV plaque forming ability or spread, reflected in plaque size [24]. De novo HSV DNA synthesis has also been analysed using EdC incorporation [25]. Here we expand on these methods to produce infectious HSV containing EdC incorporating genomes, (termed HSVEdC). Using an in vitro uncoating assay on solid supports [26,27], we show that the vast majority of particles contained bioorthogonally-tagged genomes, detectable by cycloaddition to azide-linked fluorescent probes. Remarkably, if HSVEdC virions were also heat treated on the support prior to cycloaddition, virus DNA ejected from the capsid could be coupled to azide-linked fluorochromes and detectable as filamentous strands. Genomes were not detectable in virions in solution nor on cell surfaces at +4°C, where numerous capsids could readily be observed but without genome accessibility. When infection was initiated by raising the temperature to 37°C, DNA uncoating and transport in the nucleus could be observed within 30 mins and prior to synthesis and recruitment of the major immediate-early (IE) regulator ICP4. Using these assays we undertake a comprehensive quantitative spatiotemporal analysis of genome trafficking and uncoating, including analysis by three-dimensional structured illumination microscopy. Together with additional data reported here, this work provides the first direct quantitative spatiotemporal analysis of HSV genome transport and presentation to the cellular environment, revealing new processes in genome dynamics not previously appreciated and advancing our understanding of these crucial early steps in infection. Previous work from our own [24] and other laboratories [16,25] has demonstrated the incorporation of alkyne-derivatised nucleotides into HSV replication compartments and the co-localisation of nascent replicating DNA with virus replication proteins using combined click-chemistry and immunofluorescence approaches. We found that EdC (ethynyl-deoxycytidine) was more sensitive for detection of HSV replication centres than EdU (ethynyl-deoxyuridine), consistent with the high GC content of HSV DNA [24]. In this work, optimising conditions for the production of HSVEdC we found no significant effect of EdC even over relatively prolonged times (72 hrs) on uninfected cell growth or morphology (S1 Fig) and no significant effect on the efficiency of virus plaque formation nor plaque spread (Fig 1A), consistent with previous data [24]. In analysis of both single step (Fig 1B) and multi-step replication (Fig 1C), we found at most a minor effect (3–4–fold reduction) in overall yields. The particle/pfu ratio of HSVEdC produced from multi-round virus replication and normal HSV produced in the absence of precursor were compared by particle counts using normalised amounts of pfu. The results show a modest increase for HSVEdC (Fig 1D). We do not know the precise explanation for this slight increase though the value was within the range of particle/pfu we have found for normal HSV stocks (5–20). Overall even prolonged incubation with EdC at concentrations at least up to 10 μM was well tolerated, with minimal effect of virus replication and the progression of infection. EdC incorporation in uninfected cells (4 hrs) showed DNA synthesis in approximately 25–30% of cells (Fig 2A and 2B) with the spatial localisation patterns observed at higher magnification (S2 Fig) varying from discrete small clusters (S2i Fig) to more intense incorporation in prominent large focal clusters (S2ii and S2iii Fig) or homogeneous diffuse patterns (S2iv Fig). These patterns are consistent with previous spatial analysis of cellular DNA synthesis and reflect approximate stage within S-phase [20,22]. In HSV infected cells, the percentage of positive cells increased to approximately 55% by 5 hpi (hours post infection) and virtually all cells were positive for EdC incorporation by 8 hpi (Fig 2A, summarised panel b). From higher resolution spatial analysis of active DNA synthesis and localisation of the major DNA replication protein ICP8, several distinct patterns were observed (Fig 3). By 5 hpi, in those cells positive for DNA synthesis, the majority of cells showed discrete replication foci, colocalising with ICP8 (Fig 3, panels ii, iii) and reflecting the previously documented features of HSV DNA replication compartments [28–30]. We also observed populations of infected cells wherein multiple intense focal clusters of nascent DNA synthesis were observed, but in this case without any clear colocalisation with ICP8, which nevertheless still formed in distinct smaller (Fig 3, iv) or larger (Fig 3, v, vii) lobules. These latter patterns likely represent infection of cells which were in S-phase (or committed to S-phase and not prevented from doing so). The localisation of ICP8 likely represents some level of ongoing viral DNA synthesis in such S-phase cells, in the background of prominent cellular DNA synthesis. Again these results are entirely consistent with previous data [31]. We carried out a series of experiments to optimise conditions for the production of HSVEdC and then scaled up (see materials and methods) with multi-round infection initiated at low multiplicity of infection (moi; 0. 005 pfu/cell) and two pulse-labelling intervals in the presence of 5 μM EdC. Overall yields of HSVEdC virus produced were virtually unchanged from normal virus production. The lack of any pronounced effect of EdC on HSV yield and its incorporation into replication compartments does not necessarily mean that it will be efficiently incorporated or detectable in capsids. Therefore to examine the efficiency of detection of genomes in capsids we exploited an in vitro assay reported by Newcomb et al. , [26,27] which showed that HSV capsids, after absorption onto solid surfaces, underwent some form of structural rearrangement (s) such that with moderately elevated temperatures, genome expulsion from the capsids could be observed. Although the precise mechanism was unknown, these observations indicate that attachment to a solid surface perturbs the capsid, (or transmits a structural change to the portal) facilitating DNA release. Samples of HSVEdC or unlabelled HSV-1[17] were adsorbed onto borosilicate coverslips, fixed and processed for the simultaneous detection of genomes (green channel) and capsids (using anti-VP5 antibody, red channel). Typical results showing the merged image for HSVEdC and HSV-1[17] are illustrated in Fig 4, panels I and IV respectively. The individual channels for genome detection are shown in corresponding panels II and V for each virus. Using an ImageJ plugin, capsids were enumerated and the signals quantitated in each channel (see materials and methods). Particles are categorised as a positive red particle or positive green particle, requiring positive particles to be not only above the background ROI but 1 standard deviation (SD) above the background ROI. The macro produces a colour-coded overlay (panels III and VI) in which particles containing both signals are coded yellow, particles that are capsid positive but lacking a genome signal above threshold are indicated in red, and particles with a genome signal but lacking a capsid signal are coloured in green. Quantitation is shown in the right-hand panels. Approximately 700 HSVEdC capsid particles were identified in this representative field of which 95% were positive for genome detection. The small percentage of particles that do not contain a genome signal above threshold could be due to low detection signal, low EdC incorporation, defective particles, or genome release (see below). The extremely small numbers of green particles that were not detectable by VP5 immunofluorescence could also be due to defective particles or released DNA. However clearly the vast majority of HSVEdC capsids contained detectable EdC containing genomes. The control HSV-1[17] had essentially no detectable genomes above background (Fig 4, IV-VI and panel), demonstrating the extreme specificity of the reaction. From our analysis, the percentage of HSVEdC particles with detectable genomes was comparable if not slightly greater than that for similarly labelled Adv (using a combination of EdA and EdC), where approximately 90% of capsids applied to coverslips contained detectable genomes [16]. This information is important since if incorporation efficiency was low, e. g. , only 10% of particles were detected or the efficiency was unknown, then subsequent studies examining genome localisation may give an incomplete picture. Further quantitation of HSVEdC is given in S3 Fig, including the comparative distributions and variance for VP5 detected by immunofluorescence and genomes detected by cycloaddition. Gaussian distributions were fitted to each channels’ frequency data using Image J curve fitter. We used the coefficient of variation (CV, (σ/μ) x 100) as a measure of variance. The goodness of fit to a normal distribution of VP5 intensities exceeded 0. 95 with a CV of 23. 75 (S3 Fig). This variance in particle intensity is very similar to analogous types of HSV single particle analysis using either GFP-fusion proteins or antibody to capsid protein [32,33]. With this as a benchmark, we found the distribution genome signal detected by cycloaddition to have only marginally increased variance with a CV of 34. 19 and a goodness of fit to normal of 0. 94 (S3 Fig). The scatter plot for individual particle analysis is illustrated in S4 Fig for both HSVEdC and HSV, showing the vast majority of genome-positive particles with only low numbers of particles having a genome signal below threshold. HSV w/t exhibited a single focus above background, likely an artefact of detection. Taken altogether, these results demonstrate that EdC is efficiently incorporated into HSV replication compartments and subsequently into genomes in mature HSV particles, that the yields and infectivity of such particles are minimally affected and that the genomes of the majority of such particles can be detected in vitro by cycloaddition reaction. Our results by definition detect HSVEdC genomes by cycloaddition after virions are adsorbed onto coverslips. We also found in additional control experiments that the vast majority of HSVEdC genomes (99%) were not detectable if the cycloaddition reaction was performed on virions in physiological buffer, prior to adsorption to the coverslips (S5A and S5B Fig). In the original observations using purified capsids, Newcomb and colleagues observed that DNA was realised as elongated strands with progressively increasing ejection at elevated temperatures [26,27]. Although there was considerable heterogeneity between particles, release could be substantially prevented if the capsids were first cross-linked with PFA. We detected genomes within virions as punctate foci, but we did not observe elongated genome release. However our analysis was on extracellular virions and not capsids. To examine the detection and possible ejection of HSVEdC genomes further, we analysed virions that had been adsorbed onto coverslips and then subject to elevated heat treatment. The results were striking (S5C Fig). Whereas adsorption of virions at room temperature resulted in detection of genomes colocalised within capsids (e. g. panel a, also Fig 4), elevated temperature resulted in numerous elongated filamentous stands ejected from virions (panel c). This was accompanied by an increase in the numbers of capsids in which genomes were not detected as well as an increase in the numbers of punctate genome foci that were not detected by immunofluorescence. Our results are entirely consistent with the previous data obtained by electron microscopy and demonstrate that HSVEdC genomes when released from heat-disrupted virions could readily be detected on coverslips by the cycloaddition reaction. Furthermore, they indicate that when absorbed onto solid supports at lower temperatures, the genome is available for the cycloaddition reaction, presumably due to some conformational perturbation of the virion/capsid, but maintained in the confines of the capsid or virion. We next investigated HSVEdC genome transport and uncoating in cells in vivo. Virus (moi 10) was adsorbed onto cells at 4°C for 45 min and then either washed and processed directly or shifted to 37°C to allow fusion and virus entry and then processed 2 hrs later (Fig 5). At 4°C for both HSVEdC and HSV (Fig 5A and 5B respectively), numerous virus particles could be detected on cells (VP5, panels I). In contrast to adsorption onto coverslips, there was no significant genome signal for HSVEdC particles adsorbed onto cells (Fig 5A, 4°C, EdC channel, panel II). This is consistent with the lack of genome detection in physiological buffer and supports the proposal from this and previous work [26,27] for a conformational perturbation upon adsorption to artificial surfaces which is registered by the cycloaddition reaction. After shift to 37°C for 2 hrs, numerous capsids could be detected within the cytoplasm but with only infrequent detection of genomes (Fig 5A, panels IV-VI, see also below). In contrast, distinct genome foci were now readily observed in the nucleus (Fig 5A, panels V and merged VI). We frequently observed smaller genome foci in close proximity to capsids (panel VI, small angled arrows) together with larger nuclear foci (arrowheads). Low but detectable numbers of foci could be observed in a minority of cells in the cytoplasm (e. g. , panel VI, vertical arrows). The HSVEdC foci detected within the nucleus were somewhat heterogeneous in size at this time (2 hrs) and distinctly larger than the few foci detected within the cytoplasm (see below). No specific genome signal was observed for the control HSV-1[17] (Fig 5B, panels V-VI). In further control experiments the appearance of nuclear genome foci after HSVEdC infection was completely dependent upon the copper-catalysed cycloaddition reaction, was blocked by incubation with neutralising antibody prior to infection and was not prevented by prior treatment of HSVEdC virions with DNAse (S6 Fig). Altogether these data provide robust support for the proposal that the nuclear foci represent uncoated HSVEdC genomes that have reached nuclear pores, uncoated and have been transported into the nucleus, with infrequent (though detectable) presence in the cytoplasm of some cells. We next undertook a quantitative examination (Fig 6) of the relationship of moi to the numbers of genomic foci, evaluated at 30 min after shift to 37°C, in this case co-staining with anti-VP5 to detect infecting capsids. We used 30 min to quantitate the relationship with moi since foci became more difficult to quantitate accurately at later times because of the changing morphology at 2–3 hpi (see below) and because immediate early protein synthesis had almost invariably already occurred by 2 hpi. Fig 6A panel I shows a representative maximum projection image of cells 30 min after HSVEdC infection, illustrating the frequent occurrence of genomes at a consistent and close proximity to capsids, most likely representing recent uncoating events where the genome had not yet physically moved far from the capsids from which they emanated (Fig 6A, inset). At this early stage, HSVEdC nuclear genomic foci were comparatively homogeneous in size and shape and generally spherical. Mean numbers of genomes per nucleus observed at different mois (approximately 200 cells at each moi) are illustrated in Fig 6B (summarised in panel d), with frequency distributions indicated in Fig 6C. With increasing moi there was an increasing trend of genome foci in the nucleus, but this was not directly proportionate, with for example an average of 4–5 foci at moi 10,6 at moi 20 and 10 at moi 50. Moreover, while representing minority populations high numbers of foci could be observed in some nuclei (example, panel a, ii) with maximum numbers for moi 10,20, and 50 being 21,27 and 34 respectively (Fig 6B and 6D). Later in infection while there was a trend of increased numbers of nuclear foci, this was generally less than a 50% increase but more difficult to quantitate as indicated above. Capsid-free cytoplasmic foci (example, Fig 6E, vertical arrows) were infrequently observed though at higher mois these could represent 7–8% of the total foci (Fig 6E). While these were a minor population, such cells could be relevant e. g. , to overall cellular responses (see discussion). Standard wide field microscopy is limited by diffraction and additional inherent limitations in optical imaging and capture. To examine the genomes in more detail, we pursued super-resolution microscopy using 3D structured illumination microscopy (3D-SIM) [34]. We first extended the analysis of virions on coverslips combining capsid and genome detection followed by 3D-SIM, as described in materials and methods. Raw data (five phases, three angles per plane) was then computationally reconstructed and representative individual particles from a maximum intensity projection of the 3D data is shown in Fig 7A. Further particle analysis (approximately 800 particles) was performed via 2D-Gaussian fitting in each channel to calculate dimensions at full width half maxima (FWHM, summarised in table of Fig 7A). For the capsid and genome signals the mean FWHM were 144 nm and 131 nm respectively. Although the mean genome signal was slightly smaller than the mean capsid signal, we do not take this as a significant difference between capsid dimension and packaged genome dimensions, the resolution of which is beyond the limits of these techniques. On the other hand 3D-SIM combined with quantitative object analysis allowed a better resolved spatiotemporal analysis of nuclear entry in the cell. We examined genome presentation in the nucleus at early times of infection (0. 5 hpi). Cells were imaged by 3D-SIM and the data processed using the Object analyser module of Huygens image processing software in which after 3D segmentation, geometrical and spatial localisation data can be calculated for individual objects. A representative 2D field of the 3D rendered image is shown in Fig 7B (tilted to reveal the z-dimension). This field shows capsids in the red channel, genomes in green and DAPI stained nucleus in blue, with transparencies applied. An accompanying 3D video animation is shown in S7 Fig. The inherent lower optical resolution in the z-dimension than x/y-dimensions results in slightly oblong capsids. This optical limitation applies also to the genomic foci but does not affect the main conclusions on comparative volume between genomes. We compared virions on coverslips versus infected cell nuclear genomic foci at 0. 5 hpi and 2 hpi (Fig 7C). For clarity in the infected cell nuclei only the genome signal is shown. The 3D genome objects were quantified for volume (Fig 7D) and shape (Fig 7E), the latter measured as proximity to a sphere (sphericity). The results demonstrate a substantial increase (approximately 3-fold) in mean volume of the nuclear genomic foci compared to those in virions (Fig 7C panels I, II; Fig 7D). Nuclear foci at 0. 5 hpi were comparatively homogeneous and with only marginal differences in sphericity (Fig 7D and 7E). The linear length of the HSV genome is approximately 50 μm and would stretch across a typical cell several times. Our results demonstrate that while HSV genomes clearly expand, they are initially constrained and condensed to a comparatively consistent compact, roughly spherical volume. By 2 hpi, the mean volume of the foci had increased further (by another 2–4 fold), but there was also a marked increase in irregularity as the foci became more decondensed and dissipated. As infection progressed beyond 2 hrs, the genome signal became more difficult to quantitate evolving into diffuse, dissipated aggregates of lower intensity, but frequently with smaller, more condensed foci remaining (see below), usually on the perimeter of the decondensed material. To examine this alteration in genome compaction further we extended the analysis to investigate the relationship between localisation of ICP4, the major HSV IE regulator of transcription and genome morphology. Representative fields (four examples at each time point) illustrating several features are shown in Fig 8. As shown above, within 30 min after shift to 37°C, new discrete nuclear foci (green channel) were readily observed. These were detectable prior to the accumulation of ICP4 (Fig 8,0. 5 hpi). By 1 hr, together with the progressive increase in size of genome foci, as ICP4 became detectable we observed several patterns of localisation. These included cells with ICP4 recruitment to a subset of genomes with several foci still having undetectable levels (e. g. , panel V); cells with most foci accumulating ICP4 at some level (panels VI-VII) and cells with virtual quantitative accumulation of ICP4 on all genomic foci (panel VIII). At this time there was comparatively little diffuse ICP4 with the majority co-located with genome foci. By 2–3 hrs a qualitative change was observed in this association. As discussed above, the infecting genomic foci appeared as increasingly heterogeneous dispersed aggregates of lower intensity, together with residual punctate foci frequently on the periphery of the dispersed pattern (Fig 8A, 3 hpi). However while ICP4 exhibited quite precise co-localisation with the genome foci at 1 hr, by 3 hrs ICP4 was observed more within the dispersed areas of genome labelling, lacking any distinct enrichment in the remaining punctate foci. This was a distinct qualitative feature wherein foci remaining at later times on the periphery of the diffuse areas either completely lacked or were not enriched in ICP4, while foci at earlier times selectively recruited ICP4. This distinction of ICP4 localisation on earlier condensed (1 hr, panel VIII) versus later more decondensed foci (3 hrs, panel XII) is shown in expanded form (Fig 8B) with each channel separate and merged. A separate analysis by 3D-SIM is shown in Fig 8C, (2 hpi), illustrating a more condensed genomic signal (arrowed) on the edge of extended decondensed area, with ICP4 localised preferentially within the decondensed material and virtually excluded from remaining punctate focus. To examine the relationship between metabolic processes acting in the infected cell nucleus and virus genome localisation and compaction, we next investigated the effect of inhibition of transcription, viral DNA synthesis, or translation on genome localisation to the nucleus and the morphological condensation state. Cells were infected (moi 10) either untreated or in the presence of a series of inhibitors each added 1 hr prior to infection; actinomycin D (Act D, 5 μg/ml), acyclovir (ACV, 500 μM), phosphonoacetic acid (PAA, 400 μg/ml), or cycloheximide (CHX, 100 μg/ml) to inhibit transcription, virus DNA replication and protein synthesis. Control experiments confirmed activity of the drugs e. g. , ICP4 protein synthesis was completely blocked by Act D and CHX but not by PAA or ACV (S8A Fig). Representative results for genome uncoating and morphology for each inhibitor at each of four early time points are shown in Fig 9. In the absence of drug treatment, genome foci were observed in the nucleus by 0. 5 hpi together with the temporal increase in volume, irregular morphology (1,2 hpi), progressive decondensation and dissipation (3 hpi) described above. Two significant conclusions could be made from results of inhibition of transcription. Firstly there was no significant effect on either the average initial numbers or morphology of uncoated nuclear genome foci, indicating that transcription per se (for example, by a transcription-coupled ratcheting process) played no discernible role in genome uncoating and transport into the nucleus (Fig 9, + Act D). Secondly however it was clear that inhibition of transcription did have a significant effect on the progressive increase in genome volume and eventual decompaction, which were almost completely inhibited (cf, 3 hpi, untreated versus +Act D). These results were distinct from those obtained after inhibition of virus DNA synthesis. Similar numbers of uncoated infecting genome foci were initially observed, not unexpectedly. In this case however, a progressive increase in foci volume was observed, with a clear difference especially by 3 hpi for the ACV/PAA treated cells versus the Act D treated cells (cf, small arrowed foci in Act D versus larger foci in PAA/ACV panels). On the other hand, inhibition of DNA replication clearly had an effect, preventing the later dissipation and decrease in intensity seen in untreated cells (cf, 3 hpi, untreated versus ACV/PAA), indicating that these latter events were likely related to genome replication or other coupled events. The smaller genome foci size and tighter morphology in the presence of Act D compared to that observed in the presence of ACV or PAA also indicates that events linked to or downstream of transcription per se, but not requiring DNA replication, are reflected in increasing volume and irregularity and decreased compaction of the uncoated genomic foci. We repeated this analysis for Act D including relatively late times (8 hpi), comparing the fate of infecting virus genomes during normal progression of infection or when transcription was inhibited (Fig 10A). Under normal conditions, the temporal trend to increasing dissipation and decreased signal intensity continued so that by 8 hpi, the diffuse dispersed signal from incoming genomes was extremely low (Fig 10A). In distinct contrast, in the presence of Act D, not only were genomes maintained in more condensed foci, they were maintained until at least 8 hpi, with minimal changes in numbers or morphology (Fig 10A, + Act D). We also examined genome localisation in the presence of cycloheximide. Abundant accumulated data has shown that, unlike in the presence of Act D, in the presence of CHX transcription occurs but, in the absence of HSV protein synthesis, is limited to IE loci with little or no DE transcription nor DNA replication. CHX did not block uncoating and the numbers of nuclear genome foci at 0. 5 hpi were similar in CHX treated to untreated and to each of the other inhibitors (Fig 9, CHX). However, we observed a distinct feature in genome morphology at subsequent times with CHX treatment. Thus whereas inhibition of transcription resulted in the maintenance of tighter more condensed and largely singular foci (Figs 9 and 10A and 10B), genome localisation in the presence of CHX resulted in a distinct congregation and clustering of genomic foci (Fig 9, CHX, circled clusters). While individual singular foci could still be observed in CHX treated cells, this was a noticeable qualitative change in genome presentation especially at later times (see Fig 10B, comparison at 3 hpi of untreated, Act D treated and CHX treated). To examine this difference in an unbiased quantitative manner, we used the spatial statistics and focus clustering algorithm in ImageJ [35]. This algorithm (see materials and methods) examines the positions of objects within a reference structure (nuclei in this case) and assesses clustering using a normalized measure of the difference between the observed distribution of inter-point nearest neighbour distances and a completely random one. This difference is termed the Spatial Distribution Index (SDI). S9 Fig shows a schematic illustration of the analysis of the SDI. Panel a illustrates theoretical nuclei with random, clustered or dispersed patterns. The algorithm compares the cumulative inter-point distance frequencies (CDFs) for a truly completely random distribution (panel b, black lines in each graph; 95% confidence limits in grey lines) with the actual cumulative distribution obtained for foci in each example pattern (panel b, red lines in each graph). An actual random distribution (left cell) will show a distribution overlapping the theoretical random distribution while clustered (middle cell) or dispersed (right cell) distributions deviate significantly to the left or right respectively (cf, red and black CDFs in each panel). SDIs are then calculated as a probability index with a SDI close to 0 indicating a more clustered pattern, while SDIs closer to 1 indicate a pattern that is more evenly spread out. The theoretical overall distributions of SDIs for populations of cells is then calculated (S9C Fig bottom panels). A truly random pattern will show approximately even distributions of SDI values between 0 and 1 while clustered SDI distributions will show a distinct leftward shift towards lower values and evenly spaced distributions show a shift towards 1. We validated this approach using the completely random pattern of spots when viruses were applied to coverslips (S9E Fig). The CDF function of these capsid foci (S9F Fig, red line) directly overlapped with the theoretical random distribution pattern for the image (black line). Note while clusters can be observed in the distributed capsids, such clusters will occur by definition, but there is no significant difference between the overall distribution and a random pattern (S9F Fig). We applied this nearest neighbour analysis to virus genomes in infected cells for individual nuclei (approximately 50 nuclei, 3 hpi) in the presence of Act D or CHX. SDIs were calculated for the foci in each nucleus and compiled into a distribution of SDIs across the cells for each condition. Consistent with the indication from visual inspection (Figs 9 and 10B), the results indicate a clear difference in the population distribution of inter-foci distances in Act D versus CHX treated cells (Fig 10C). While the Act D pattern deviates from random to some degree across the population, this was bordering on statistical significance (p-value = 0. 056; D statistic = 0. 26). What was clear was the distinctly different trends for CHX versus Act D, indicating a highly significant change in relative localisation and clustering of genomes in the presence of CHX versus Act D (p-value < 0. 0001; D statistic = 0. 52). In a second approach to support these conclusions we examined clustering by an independent method, the distance-based K function (Ripley function). In this method (see materials and methods) using the BioImage Analysis platform ICY [36], nuclei are segmented and EdC labelled genomes located similarly to the approach above. Regions of interest (circles) with increasing radii are drawn around each detected spot and other spots located within the circles are identified. The K-function, K (r), is based on the average number of points inside a circle of radius r, calculated for the increasing radii and has an expectant value of zero for a random distribution of spots. The amplitude of the K-function can then be compared to corresponding low and high quantiles of completely random distributions (0. 01 and 0. 99 here). When K (r) is higher than the high quantile for a certain radius, the foci are significantly organised in clusters. Conversely, when K is lower than the low quantile, the foci are dispersed. The results were very clear and indicated a distinct statistical significance for clustering of genomes in CHX treated cells with search radii from 1–4 μm and a tailing off as the radii became too large to attribute significance (Fig 10D). In contrast the spatial distribution of genomes in Act D treated cells could not be attributed any distinct pattern. Altogether from the spatial analysis of the images and clustering analysis based on independent methods, the results strongly support the conclusions for a distinct difference between Act D and CHX treated cells and the proposal that transcription is recognised in the host cell and results in distinct events we have termed genome congregation (see discussion). In a previous report using an indirect surrogate measure (i. e. , β-galactosidase enzyme activity from a recombinant virus or capsid localisation), it was concluded that proteasome inhibition, which suppressed β-galactosidase activity, did so by preventing HSV genome transport to the nucleus [37]. Having established a direct assay, we addressed whether proteasome inhibition had any detectable effect for genome uncoating and nuclear import. MG132 (10 μM) was added to cells 1 hr prior to infection with HSVEdC at moi 10, and genome localisation assessed at 0. 5 hpi compared to untreated cells. The distribution of numbers of nuclear genome foci in individual cells was assessed for at least 50 cells in each condition (Fig 11A, insets show representative individual nuclei). Overall there was no significant effect of MG132 on either total numbers or distribution of HSV genomes transported to the nucleus. Similar results were obtained analysing genome localisation at 1 hr (summarised, Fig 11B). Inhibition of CRM1-dependent nuclear export by Leptomycin B treatment has previously been shown to inhibit Adv genome nuclear entry [16]. We also examined the effect of Leptomycin B (as used in the previous studies). In contrast to the effects on Adv, we observed no significant effect on HSVEdC genome uncoating and import. By comparison, when we examined nocodazole treatment, a drug which depolymerises microtubules and has been previously shown in many studies to inhibit HSV infection and capsid transport [38–40], we observed a very striking inhibition of the appearance of nuclear genomes. Control experiments for the activities of MG132 and Leptomycin B using known targets and effects confirmed their action (S8B Fig). Taken together with the positive control for the suppression of genome nuclear entry by nocodazole, we conclude that neither proteasome function nor nuclear export are required for the initial stages of HSV nuclear transport, uncoating and nuclear import. We demonstrate the efficient incorporation of EdC into virus replication compartments, colocalisation with the major virus DNA binding protein, ICP8 and features of virus DNA replication in relation to the cell cycle that are entirely consistent with previous work [28–30]. However, results demonstrating the incorporation of EdC into replication compartments and its minimal effect on virus yields do not necessarily mean that it would be incorporated into mature infectious virions. Information on the efficiency and proportion of particles that contain detectable EdC is necessary for subsequent analyses and we exploited an in vitro assay described by Newcomb et al. , who showed that HSV genomes are ejected from the capsid upon attachment to solid supports due to undefined structural perturbation (s) [26,27]. Consistent with this, we show that adsorption on glass induces a structural change in the capsid permitting access to the catalytic molecules involved in cycloaddition, including the azide-fluorochrome. In our analysis with virions the genome was retained within the particle while with purified capsids the genome was readily released [26]. We propose that while there is some structural change in the virion capsid allowing the coupling reaction to the DNA, the genome is nevertheless retained in the confines of the particle due to surrounding components. It was also previously demonstrated that when the genome was released from purified capsids, it was ejected in a polarised manner likely from the portal and that the proposed structural alteration may impact directly or indirectly on portal integrity [26]. A portal-specific alteration is possible but not necessary to explain our observations and it could be that some more global perturbation around the capsid shell could allow access to the components of the cycloaddition reaction, the largest being the azide-fluorochrome (mol wt 861). Indeed there could be distinct perturbations with one type allowing access to small compounds and another involving changes including at the portal, promoting genome release. Whether this is the explanation for our observations is beyond the scope of this work. Nevertheless the ability to identify the genome within the capsid might be exploited for other types of analysis e. g. in vitro biophysical analysis of genome transitions [57] or the identification of specific host components that may promote release. It is also interesting to compare these results with similar analyses of Adv on solid supports [16]. In the case of Adv, EdC-labelled genomes were not detected upon initial adsorption of the virus to coverslips but were observed after heat disruption. Heat treatment revealed internal protein VII (by immunofluorescence) and allowed cycloaddition labelling of the genome, which nevertheless remained tightly associated with the capsid [16]. The ejection of the HSV genome (from capsids or heated virions) versus the maintained association of the Adv genome (from heated capsids) most likely reflects differences in the pressurisation status of genomes within capsids [58] and the lack of DNA packaging proteins within HSV compared to Adv where the genome is associated with several core proteins, in particular protein VII [59]. Such differences in internal pressure and protein-genome association are likely reflected in differences in mechanism in nuclear pore engagement and genome import. Uncoated nuclear genomes could be detected within 30 min of infection at 37°C with HSVEdC. Using similar methods to examine Adv infection, genomes could not be detected in the nucleus at 30 min and quantitative analysis on nuclear entry was performed at 2. 5 hrs, a comparatively late point in our analyses, when genomes were already uncoated, decondensed and in many cases beginning to replicate. This does not necessarily indicate that HSV genome import is more rapid than adenovirus. This would require a direct parallel comparison with similarly labelled viruses in the same cells and on identical imaging systems and even then would be a qualified comparison. Nonetheless for the majority of Adv capsids, their genomes become rapidly accessible to click detection in the cytoplasm, reflecting the initial stages of uncoating [12,16,59,60]. Given differences in entry processes it is perhaps not unsurprising that unlike Adv, HSV genome entry is not sensitive to Leptomycin B inhibition of nuclear export. For HSVEdC while uncoated cytoplasmic genomes could be detected this was a minor population of the total genomes, mostly in a subpopulation of cells. Although unlikely to contribute to the nuclear genome pool, such genomes may play a distinct role in the cell population as a whole, including host responses from subsets of cells that might elicit paracrine effectors to other cells. We conclude that for HSV, most capsids do not undergo structural transitions that perturb access (at least resolvable by cycloaddition labelling with azide-fluorochromes), such transitions likely being tightly coupled to engagement with the nuclear pore. However other factors including cell type could influence capsid integrity and genome accessibility, e. g. , differences in entry by fusion at the plasma membrane versus endocytosis. The ability to positively identify genomes associated with perturbed capsids will be useful in future studies including investigation of capsid integrity as a function of cell type and the influence of prior immune stimulation by different pathways. We found no significant qualitative or quantitative difference in the presence of MG132 and conclude that proteasome function is not required for transport of the relevant capsid population or genome import per se. This is in contrast to a previous publication which indicated that proteasome function was required post-entry for efficient delivery of incoming HSV capsids to the nucleus and subsequent gene expression [37]. While there could be several explanations for the difference, in our analyses we directly measure genome import while previous conclusions were based on measurements of gene expression (β-galactosidase) at 6 hpi or on measure of capsid localisation of a fluorescent virus at 2. 5 hpi. Proteasome inhibition could have inhibited a number of processes involved in surrogate read-out later in infection or inhibited bulk capsid dynamics that were not important for early genome delivery. As summarised (Fig 12), at the earliest time detectable (stage 1, within 0. 5 hpi) uncoated nuclear genomes were in a comparatively homogeneous, roughly spherical form that had expanded to approximately 3-fold the volume within virions. Thus while there is a distinct genome decompaction after nuclear import, this is constrained in a relatively regular manner. The distribution of the numbers of genomes appearing in the nucleus and the relationship to moi bear similarity to those from physical analysis of adenovirus genome entry [16] and are relevant also to conclusions from other studies on the numbers of herpesvirus nuclear genomes that participate in transcription and replication [61–63]. We observed that at a standard moi of 10 pfu/cell (100–200 particles coating a cell), although a small percentage of nuclei could contain relatively high numbers, the mean numbers of nuclear genome foci was approximately 5 and around 90% of cells had fewer than 10. Increasing moi by 5-fold did not increase nuclear genome numbers 5-fold indicating the operation of some form of limit (s) on infection (though this could be at a number of stages) and a decreasing efficiency of nuclear import at higher mois. Similar conclusions were made for Adv nuclear entry [16]. We currently cannot discriminate the fate of capsids which do not uncoat at the nuclear pore, while for Adv genomes appear to be lost from the capsid. Based on mathematical modelling of simultaneous infections with strains of pseudorabies virus expressing individual fluorophores, it has also been estimated that an average of approximately five infecting genomes are expressed per cell at a moi of 10 and that even at moi 100 the mean is no more than 7–8 genomes [61]. Clearly, at the more extreme ends of distributions from our analyses, high numbers of genomes can enter the nucleus. However, at a standard moi of 10, the mean numbers of physical nuclear genomes are of the same approximation as the numbers of genomes that have been proposed to express or replicate [61]. One implication from this is that once imported into the nucleus, the efficiency of transcription may be relatively high. Indeed at early times, many of the physical genomes were associated with ICP4 (see below), though this does not necessarily mean such genomes are indeed transcribed. Such conclusions will require the ability to simultaneously visualise genomes and nascent transcripts or accumulating RNA. We are currently developing bioorthogonal approaches [64] with distinct chemical moieties on DNA versus RNA that allow copper-dependent and independent coupling of distinct fluorochromes to examine these questions in virus infected cells. After nuclear import, the HSV genome initially expands and continues to decondense in a series of transitions that could be delimited using chemical inhibitors. In the absence of transcription, genomes remained relatively compact and were maintained in that form in the nucleus for at least 8 hrs. De novo virus protein synthesis is therefore not required for the initial compact state of the genome, the nature of which is discussed further below. Allowing transcription but in the absence of de novo translation revealed a distinct feature, which we termed genome congregation, not observed when transcription was blocked. Several mechanisms could contribute to this process. Transcription on the viral genomes themselves, i. e. , the templates for transcription, could directly contribute to congregation, even if not all genomes were transcribing. This could be via components of the transcription/splicing apparatus somehow progressively capturing multiple genomes in a spatially restricted manner. It could also be that genome congregation is a host response to infection (with no viral proteins yet made), specifically recognising the process of transcription and sequestering genomes as a result. Further explanations are possible but understanding the process of genome congregation will be important for any full understanding of genome dynamics, competency for transcription and host responses to infection (see below). During normal infection, genomes underwent further progressive decondensation, eventually becoming difficult to discriminate and dissipating within DNA replication compartments. We frequently observed longer lived residual condensed foci usually at the periphery of replication compartments (Fig 12, stages 2–4). The intermediate stages of these genome transitions did not require DNA replication per se, with the enlarged and decondensed morphology of infecting genomes in the presence of DNA replication inhibitors being distinct from that observed in the presence of Act D or CHX (Fig 12, summary schematic view, shaded sectors). We conclude that recruitment of regulatory and/or replication factors combined with the more extensive early transcription, results in further changes and decondensation of the genome while downstream DNA replication and associated processes e. g. , recombination, branching and extensive late transcription, [65,66] results in more complete decondensation of input genomes at later times. It is possible that the longer lived condensed foci remaining on the periphery of replication compartments represent either replicated parental strands that remain as foci, or potentially a subset of parental genomes that were not acted upon by either replication or transcription. Rolling circle replication [65,67,68] acting on HSVEdC would result in one labelled parental strand remaining at the replication fork, while the other parental strand would progressively move away as replication and unlabelled progeny DNA accumulates. It is not currently technically possible to discriminate between these possibilities and other explanations are also possible but in this regard the pattern of genome association with ICP4 warrants discussion. While there was heterogeneity at an individual genome level with some genomes not accumulating ICP4, the majority of condensed genomes recruited and were enriched for ICP4 by 1–2 hrs. (As discussed above, this does not necessarily imply productive transcription from all genomes). However as infection progressed there was a clear distinction in this association. Those genomes that remained as condensed foci (found mainly on the periphery of replication compartments) were selectively depleted for ICP4 and frequently devoid of the protein altogether. One possible explanation is that these foci never accumulated ICP4, in which case they exhibit significant selectivity since ICP4 would clearly have been initially recruited to certain other foci and ICP4 was also present in adjacent decondensed replication centres in the same nuclei. Alternatively it could have been that ICP4 was recruited to many foci but different downstream pathways dictate either maintenance of a more condensed state coupled with displacement of ICP4 or progressive decondensation (and associated replication/transcription) and association with ICP4. Future work developing methods for the simultaneous visualisation of infecting genome localisation and condensation, active transcription and protein localisation will help address the nature of these relationships revealed in this work. Finally, our results on spatial analyses are relevant to the interpretation of the many previous biochemical analyses on the nature of the infecting HSV genome. MCN digestion experiments of the bulk virus genome population strongly indicate that the considerable majority of infecting genomes released from the capsid are randomly digested and not assembled into any conventional nucleosomal organisation [45–50,69]. On the other hand ChiP analyses, which usually address a minor fraction of the total DNA, suggests that histones in some form are associated with at least a population of genomes [52,53,55,70,71]. One model attempting to integrate results from different approaches proposes that infecting genomes associate with some form of nucleoprotein complex that includes histones but in a non-conventional highly distributive, rapidly associating/dissociating organisation [56,69]. We show that after capsid exit and nuclear import, genomes expand but in a constrained and distinct state and then further decondenses in a discernible fashion prior to replication, and that replication and potentially associated transcription result in further extensive dissipation within the nucleus. In addition to heterogeneity arising from overlapping temporal transitions, heterogeneity arises from subpopulations of genomes that may not associate with e. g. ICP4, or which at later times remain in a more condensed configuration. Thus certain proteins may be selectively associated with specific subpopulations of these genome, as an example the longer lived condensed foci, and thus antibodies to such proteins sample only those genome populations. Future work combining bioorthogonal chemistry for spatial analyses of genomes and ongoing transcription and replication together with immunofluorescence analysis to localise viral and host cell proteins will be necessary to resolve these questions. In conclusion, using compatible bioorthogonal nucleoside precursors for genome labelling in HSV infected cells and quantitative individual particle analysis, we demonstrate extremely efficient precursor incorporation resulting in virtually quantitative detection on an individual particle basis in the population of progeny virus. We then report a comprehensive analysis in infected cells of genome dynamics during capsid exit and nuclear import in which we; demonstrate qualitative transitions in genome condensation state linked to transcription and replication; reveal novel processes in genome congregation dependent upon transcription and show the temporal switching in regulatory protein recruitment (represented by ICP4) to distinct genome compartments. Altogether our results reveal novel aspects of the spatiotemporal dynamics of HSV genome uncoating, transport and organisation that can be integrated with previous biochemically based analyses and provide a framework for future investigation in distinct fields of host cell-virus genome. RPE-1 cells, a human telomerase immortalised retinal pigment epithelial cell line, (kindly provided by Dr Andrew McAinsh University of Warwick, UK) were grown in Dulbecco’s modified minimal essential medium (DMEM/F12, Sigma-Aldrich) supplemented with 200 mM glutamine, 10% newborn bovine serum (NCS; Gibco) and penicillin/streptomycin. The wild-type (w/t) parental strain was HSV-1[17]. Routine plaque assays were performed in RPE cells in the presence of pooled neutralising human serum (Sigma-Aldrich) at 2% or clinical grade purified human immunoglobulin (IVIg, Carimune NF, Nanofiltered, human immune globulin, CSL Behring) at 2 mg/ml, having demonstrated complete neutralisation of extracellular virus at this dose (>6 log reduction in virus titre). High multiplicity infections were performed at multiplicities of infection stated in the experiments and for routine experiments usually at a moi of 10. In control experiments for genome uncoating, the inoculum was treated with 500 U/ml DNase I (Roche) for 1 hr at 4°C, or 10 mg/ml IVIg for 0. 5 hr at room temperature. Inhibitors were used at the following final concentrations; acycloguanosine (ACV, Thermo Scientific, 500 μM); phosphonoacetic acid (PAA, 400 μg/ml); actinomycin D (Sigma-Aldrich, 5 μg/ml); MG132 (Calbiochem, 10 μM); Leptomycin B (Sigma-Aldrich, 20 nM); nocodazole (Sigma-Aldrich, 2 μM). Inhibitors were added to cells for 1 hr prior to infection. To examine the effects of EdC on cell growth, RPE-1 cells were pulsed with increasing concentrations of EdC (Sigma-Aldrich, #T511307) for 48 hr and examined by phase-contrast microscopy either live or after fixation and staining with crystal violet. Viability was determined by trypan blue exclusion using an automated cell counter. For the examination of the effects of EdC pulse-labelling on viral plaque development, RPE cells were infected at 50 pfu/well, the inoculum was then neutralised with 2% human serum after 1 hpi, and EdC was then added 2 hpi for the remainder of the assay. Plaque sizes and numbers were measured at 48 hpi using Image Pro Plus 7 software. For the effects of EdC on virus yield, RPE-1 cells were infected at moi 5 for single-step growth or moi 0. 005 for multi-step growth. Inocula were neutralised at 1 hpi with a 40 mM citric acid wash. Cells were then incubated in the presence of various concentrations of EdC (added at 2 hpi) for the duration of the experiment. Supernatant and cell-associated virus was harvested at 20 hpi (single-step) and 72 hpi (multi-step) and yield assessed by plaque titration on RPE-1 cells. RPE-1 cells were grown in roller bottles (850 cm2 surface area) to ~80% confluency. The cells were infected with HSV-1[17] at a moi of 0. 025 in 25 ml medium without serum and made to 2% NCS at 1 hpi. EdC was added to a final concentration of 5 μM at 6 hpi and again at 24 hpi. Virus was harvested at approximately 48 hpi, separating cell associated and supernatant virus by low speed centrifugation (3000 rpm, 4°C for 15 min). Supernatant virus was transferred into Oakridge tubes and pelleted in a Sorvall centrifuge RC5B using a SS34 rotor at 19,000 rpm at 4°C for 90 min. For cell-associated virus, the virus pellet was first clarified of cell debris, pelleted by high speed centrifugation as above, resuspended in PBS and applied to the top of 0. 5 ml 35% sucrose cushion in polyallomer tubes and centrifuged at 25,000 rpm in a SW55Ti rotor for 1 hr. The virus pellet was resuspended and stored in PBS. Virus titres were determined on RPE-1 cells. Particle/pfu ratios were calculated by diluting control stocks of HSV or HSVEdC to equal pfu titres, spotting standardised aliquots onto coverslips and enumerating total VP5 capsid-containing virions by automated immunofluorescence microscopy and multi-tiled image acquisition to capture and quantify the entire population. Total VP5-positive/pfu particle numbers could then be evaluated for each stock. Alternatively we examined protein profiles of standardised amounts of purified virus and quantified the amounts of the major capsid protein as previously described [72]. Similar results were obtained in comparing HSV and HSVEdC particle/pfu ratios by the two methods. The ‘Mock EdC’ inoculum used for control experiments was prepared by pulsing uninfected RPE-1 cells for 48 hr with 5 μM EdC, harvesting the supernatant and concentrating exactly as if preparing virus from infected cells. Cells grown on borosilicate coverslips, infected and labelled with EdC under the variety of experimental conditions discussed, were fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) for 10 min, quenched in 100 mM glycine in PBS for 5 min, and permeabilised with 0. 5% Triton X-100 for 5 min. Samples were then processed by cycloaddition with azide-linked fluorochromes and then blocked with 10% FBS in PBS where immunofluorescence was required. Cycloaddition and immunofluorescence were essentially as described previously [24]. Briefly for the cycloaddition reaction to detect EdC labelled DNA, PFA fixed and washed coverslips were incubated in freshly prepared reaction buffer containing 1 mM CuSO4; 10 mM sodium ascorbate; 10 mM amino-guanidine and 1 mM Tris (3-hydroxypropyltriazolylmethyl) -amine (THPTA, Sigma-Aldrich) and 10 μM Alexa 488-azide (Thermofisher) in PBS pH 7. 4. Reactions were performed for 2 hr in the dark, the reaction cocktail then removed and the samples washed with PBS, dried and mounted. For subsequent immunofluorescence, cells were blocked and stained for 45 min with primary antibodies and 45 min with secondary antibodies by standard methods and mounted in ProLong Gold Antifade Mountant (Molecular Probes). Images were acquired with Zeiss Axiovert 135 TV microscope using Zeiss x63 lens (Plan-APOCHROMAT, 1. 4 numerical aperture) and Retiga 2000R camera with Image Pro Plus 7. 0 software. The following antibodies were used: mouse anti-VP5 (Virusys, HA018; 1: 300); mouse anti-ICP8 11E2 (Abcam, #20194; 1: 100); mouse anti-ICP4 (Virusys, H1A021; 1: 400); mouse anti-α-tubulin (Sigma-Aldrich, #T6074; 1: 1000); rabbit anti-PML [73] (1: 300); mouse anti-cyclin B1 (Abcam, ab18221; 1: 500); Alexa-594 Goat anti-mouse IgG (Thermofisher; 1: 750). Samples of HSV-1[17] wild-type or HSVEdC at 1x108 pfu/ml were applied to borosilicate coverslips, adsorbed for 15 min, fixed with PFA and processed as above to detect genomes and immunofluorescence using anti-VP5 antibody to detect capsids. For experiments examining genome exit we used procedures as previously reported [26,27] where virions absorbed onto the coverslips were subject to heat treatment (70°C for 2 min) either before or after PFA fixation. Samples were then processed for detection of the genome and capsids as before. In parallel experiments, viruses were subject to the cycloaddition reaction in PBS containing the appropriate concentrations of reagents. After the reaction, samples were made to 1 mM EDTA to stop any further reaction, then adsorbed onto coverslips and processed for immunofluorescence. For quantitative analysis maximum projections were captured using the Image Pro Plus Stage-Pro function and Z-stacks were obtained with 10 slices at 0. 2 μm intervals. We used Image J and a customised plugin based on the find maxima protocol. The plugin uses find maxima and places an identical sized ROI centred on the maxima with user configurable diameter to encompass virus particles. Maxima with too close a spatial overlap or at an image edge are excluded by the protocol and can be further excluded manually before quantitation. In practice this had a limited effect given the largely monodisperse nature of analysed particles. Red (capsid) and green (DNA) intensities were measured for each ROI. Mean and standard deviation (SD) background intensities were calculated separately for the red and green channels from the area outside the identified ROIs and normalised for ROI area. Maxima–based ROIs were then compared separately against the mean background for each channel and categorised using a threshold the default of which was the mean channel background plus 1 SD. Thus, to be categorised as a red (capsid) positive particle, that particle ROI must be not only be above the mean background ROI in the red channel but at least 1 SD above that background. Frequency distributions of individual identified particle ROIs were then quantitated, calculating the bin width using the Friedman-Diaconis criteria for interquartile-ranges [74,75]. The same bin width was used for both channels in the figure for ease of comparison of the distributions. Gaussian distributions were fitted to each channel frequency data using Image J curve fitter. Cells grown on coverslips were mock-infected or infected with parental HSV-1[17] or HSVEdC by normal procedures (moi 10) at 4°C for 45 min to allow virus adsorption to the cell surface. Cultures were then either washed and fixed immediately for analysis of adsorbed virus or shifted to 37°C to allow the infection to proceed. For analysis of infected cells at very early times i. e. , 30 mins, the inoculum was then removed and cells washed and fixed. For longer times, the inoculum was removed and replaced with pre-warmed medium containing 2% NCS. Cultures were washed and fixed at various times thereafter as indicated in the text and figure legends and processed for genome detection by cycloaddition reaction and immunofluorescence as described above. Infected cells were co-stained with DAPI to allow outlining of nuclei. Images were acquired by standard wide-field microscopy (described above) or 3D-structured illumination microscopy (3D-SIM). Images were then processed using Image J denoise plugin and corrected for background. For quantitative evaluation of genome foci, images were imported into Image Pro Plus and then subject to thresholding and segmentation modules to define object masks which were quantified for various parameters. Using the DAPI outlines and the population analysis tool, the number of genomes within approximately 200 nuclei was calculated for each condition under study, differing mois, times and various drug treatments. Super-resolution imaging was performed on Elyra PS1 system (Carl Zeiss) with an Apochromat 63x 1. 4 NA oil objective lens, 488nm and 561nm excitation lasers and images were captured on a sCMOS PCO Edge camera. The camera pixel size is 6. 5 μm and with 63x objective and additional 1. 6x tube lens, this corresponds to 64 nm in the object plane. For analysis of infected cells, image stacks (2 μm) were acquired in Frame Fast mode (single multiband cube) with a z-step of 110 nm and 15 raw images (five phases, three angles) per plane. Raw data was then computationally reconstructed using the ZEN software to obtain a super-resolution 3D image stack with a pixel size of 32 nm in xy and 105 nm in z. The SIMCheck ImageJ/Fiji plugin [76] was used to perform quality control on both raw and reconstructed data and to estimate lateral (x-y) resolution (approximately 120 nm) and axial (z) resolution (approximately 300 nm). Images from the different colour channels were registered in ZEN with alignment parameters obtained from calibration measurements with either virus capsids simultaneously labelled in both red and green channels or with TetraSpeck calibration beads 0. 1 μm diameter (Thermofisher). 2D Gaussian fitting was done using the PALM analysis function in Zen with 30 pixel image window or ‘GaussFit on spot’ plugin in ImageJ. The Gaussian 1/ sqrt (e) radii were converted to full width at half maximum (FWHM) values by multiplying with 2x sqrt (2*ln (2) ). For analysis of HSVEdC capsid and genome dimensions by immunofluorescence and cycloaddition reactions, the expected dimensions can be estimated by a convolution of the SIM resolution (120 nm and 138 nm for 488 nm and 561 nm excitations respectively) with the known sizes of capsid diameter (125 nm) and genome space (100 nm) [77]. In case of the capsid by immunofluorescence, based on previous analyses [78] we estimated and additional 35 nm for the primary/secondary antibody bringing the estimated dimensions of the capsid to 160 nm. The convolutions of the SIM resolution and these sizes results in an estimated size of 200 nm and 170 nm for the capsid and genome respectively. Our measured average size is about 26% smaller than the estimate size. This was in line with measurements of calibration measurements with standard fluorescent beads of different sizes where the FWHM sizes were found to be 22% smaller than expected. Volume analysis was performed the object analyser module of the Huygens image processing suite (SVI, The Netherlands). The image is segmented into defined objects by the seed-threshold level adjustment, and connection process. Introduction of a watershed increases segmentation reliability further. Detected objects are automatically labelled and submitted to a continuous Iso Surface renderer. The segmented image is shown as a coloured iso-surface image. Object statistics are reported for each object, including geometrical data and spatial location. A simulated fluorescence process (SFP) computing algorithm allows visualization of the 3D data and production of the rendered image as an animation. Using this method we estimated volume and sphericity for genomic foci from visions on coverslips versus after entry to the nucleus. Spatial clustering analysis of EdC labelled genomes was carried out using the Spatial Statistics 2D/3D ImageJ plugin [35]. The plugin analyses the overall distribution of inter-point distances including any local clusters and calculates whether there is evidence for a non-random distribution in the population of cells. A binary mask of each nucleus is generated together with a mask of the genomes using the ‘Find Maxima’ function. The plugin calculates for every nucleus, the distances between every point and its nearest neighbour and generates a cumulative distribution function (CDF) of those distances (the G-function). To compute this function, first the average CDF of a completely random distribution is estimated over a set (500 iterations) of randomly generated point patterns, specific to each reference structure (nucleus) and the number of points (genomes) in that structure. Second, the expected variation of CDFs around their average is estimated using a second set of randomly generated binomial point patterns. The relative position of the observed CDF for the actual test set within this range of variation is used to assign a p-value to the observed pattern, termed the ‘Spatial Distribution Index’ (SDI). Point patterns that tend to clustering have an SDI closer to 0 while patterns tending to even spacing have an SDI close to 1. The CDFs of SDIs of two different populations are compared using the Kolmogorov-Smirnoff (KS) test, which is non-parametric and distribution free. A p-value for the difference between the two populations is calculated, as well as the D statistic which is the largest deviation between the two CDFs. An independent clustering analysis was performed by calculating the Ripley function (K) using the BioImage Analysis platform ICY (http: //icy. bioimageanalysis. org) as described [79]. Nuclei were segmented using the DAPI signal as above and a binary mask created. The ICY Spot Detector plugin identifies the EdC genomes contained within the nuclear mask. These were used to calculate the K function using the Spatial Analysis plugin [36]. In this approach regions of interest (circles) with increasing radii are drawn around every detected spot and other spots located within the circles are identified in the overall search area (the nucleus). The K function is then based on the number of spots that are closer than the radius, calculated for each increasing radius. The function is used to report the statistical significance of whether a distribution of points is random or clustered by comparing obtained values with critical quantiles under a completely random distribution. The amplitude of the K-function can then be compared to corresponding low and high quantiles (0. 01 and 0. 99 here). When K is higher than the high quantile for a certain radius, the foci are significantly organised in clusters. Conversely, when K is lower than the low quantile, the foci are dispersed.

Virtually all DNA virus classes as well as many RNA viruses must deposit their genomes within the nucleus for transcription, genome replication and subsequent capsid assembly. While infecting capsids have been studied by various methods and biochemical approaches have been used to investigate the bulk genome population characteristics, quantitative spatiotemporal information of the infecting genome itself at the single particle level has been lacking. This is required for any complete understanding of many critical aspects of virus infection and virus pathogenesis. Using novel techniques in bioorthogonal chemistry to produce normal non-recombinant viruses with readily traceable genomes, we provide the first direct quantitative spatiotemporal analysis of HSV genome transport and presentation to the cellular environment. Using these techniques which discriminate encapsidated from uncoated genomes and input from replicated DNA, our work provides a comprehensive analysis, using direct measures for genome detection not dependant on surrogate outputs. The results reveal completely novel aspects of early genome localisation and organisation not previously appreciated or amenable to study. Furthermore the work also provides a roadmap for similar studies in other systems and for future analysis of many aspects in different fields of the biology of infecting virus genomes early during cell infection.

Abstract Introduction Results Discussion Materials and methods

microbiology viral structure dna replication dna immunologic techniques microbial genomics research and analysis methods viral genomics genomic signal processing immunoassays comparative genomics viral packaging viral replication virions immunofluorescence biochemistry signal transduction cell biology nucleic acids virology genetics biology and life sciences genomics cell signaling computational biology

2017

Spatiotemporal dynamics of HSV genome nuclear entry and compaction state transitions using bioorthogonal chemistry and super-resolution microscopy

Using a computational model, we simulated mitochondrial deoxynucleotide metabolism and mitochondrial DNA replication. Our results indicate that the output from the mitochondrial salvage enzymes alone is inadequate to support a mitochondrial DNA replication duration of as long as 10 hours. We find that an external source of deoxyribonucleoside diphosphates or triphosphates (dNTPs), in addition to those supplied by mitochondrial salvage, is essential for the replication of mitochondrial DNA to complete in the experimentally observed duration of approximately 1 to 2 hours. For meeting a relatively fast replication target of 2 hours, almost two-thirds of the dNTP requirements had to be externally supplied as either deoxyribonucleoside di- or triphosphates, at about equal rates for all four dNTPs. Added monophosphates did not suffice. However, for a replication target of 10 hours, mitochondrial salvage was able to provide for most, but not all, of the total substrate requirements. Still, additional dGTPs and dATPs had to be supplied. Our analysis of the enzyme kinetics also revealed that the majority of enzymes of this pathway prefer substrates that are not precursors (canonical deoxyribonucleosides and deoxyribonucleotides) for mitochondrial DNA replication, such as phosphorylated ribonucleotides, instead of the corresponding deoxyribonucleotides. The kinetic constants for reactions between mitochondrial salvage enzymes and deoxyribonucleotide substrates are physiologically unreasonable for achieving efficient catalysis with the expected in situ concentrations of deoxyribonucleotides. Mitochondrial DNA (mtDNA) replication [1], and the mitochondrial nucleoside salvage pathway that generates the precursor deoxyribonucleoside triphosphates (dNTPs) for mitochondrial DNA replication, have generally been believed to function independently of nuclear DNA (nDNA) replication and cytoplasmic nucleotide metabolism. However, the observation associating mutated RRM2B (a p53 inducible ribonucleotide reductase subunit) with mtDNA depletion and at least one observation of mtDNA replication restricted to S phase in DGUOK (deoxyguanosine kinase) deficient cells now make it clear that mtDNA replication and maintenance are not always completely independent of the cytoplasmic state [2], [3]. Older evidence supported the view that mitochondrial nucleotides may be isolated from the corresponding cytoplasmic pools [4], but more recent studies support a metabolic cross-talk between the mitochondria and the cytoplasm and show that nucleotide import from the cytosol very likely contributes to mitochondrial dNTP pools in both cycling and quiescent cells [5], [6]. The mechanism of this import is unknown since the discovery that the carrier SLC25A19 (solute carrier family 25, member 19) actually is a thiamine pyrophosphate transporter and not a deoxyribonucleotide transporter [7]. Similarly, the mitochondrial monophosphate kinases of dG and T deoxyribonucleotides (key elements of the purported salvage pathway) still have not been identified. In the current picture of the mitochondrial nucleoside salvage pathway, DGUOK and TK2 (thymidine kinase 2) are the nucleoside kinases; NT5M (mitochondrial 5′, 3′-nucleotidase) is a nucleotidase; CMPK2 (cytidine monophosphate kinase 2), and isoforms of adenylate kinase (AK) are the monophosphate kinases; and NME4 is the major nucleoside diphosphate kinase (Figure 1A). Deoxyribonucleosides (dNs) are converted to dNTPs through three sequential enzyme-catalyzed phosphorylations. This is a complex process with some reactions occurring in parallel for the four deoxyribonucleosides, and some reactions using the same enzyme (for example, the first phosphorylation of dT and dC are both catalyzed by TK2) in addition to the presence (not shown in Figure 1A) of feedback mechanisms (for example, dTTP and dCTP inhibition on TK2 [8]). The physical structure of the mitochondrion provides another complication that is rarely considered in this context. The mitochondrion has an intermembrane space (between the inner and outer membranes) and a matrix compartment within the inner membrane (Figure 1B). Several contact sites exist between the inner and outer membranes. In addition to the mitochondrial enzymes listed in Figure 1A, the cytoplasmic enzymes Thymidine Phosphorylase (TYMP) and RRM2B are included in the diagram since mutations in these two enzymes are known to cause phenotypes involving defects in mtDNA maintenance [2], [9]. The mtDNA are tethered to the inside of the inner membrane, within the matrix, so it would be expected that the enzymes of the salvage pathway would also be located within the matrix. In the simplest picture of the mitochondrial salvage pathway deoxyribonucleosides are transported through the inner membrane by the ENT (equilibrative nucleoside transporter) and then phosphorylated to dNTPs within the matrix. However, evidence exists to suggest that the AK2 adenylate kinase as well as NME4 nucleoside diphosphate kinase might actually be localized to the mitochondrial intermembrane space [10], [11], not in the matrix. It is possible that other isoforms of these enzymes might localize to the mitochondrial matrix [10], [11]. If not, it is hard to understand how the salvage pathway would function without an unnecessarily complicated transport of deoxyribonucleotides back and forth across the inner membrane (arrows marked with question marks in Figure 1B). In this paper we analyze the experimentally measured enzyme kinetics of these known enzymes of this pathway. Our analysis of the mitochondrial nucleotide metabolism pathway reveals that the majority of the enzymes of this pathway are not particularly effective in the synthesis of mtDNA precursors (phosphorylated deoxyribonucleosides) either due to the affinities of the enzymes for ribonucleotides and other non-DNA precursors (dI and dUMP for example) or due to a disparity in their affinities for deoxyribonucleotides versus the expected mitochondrial concentrations of those deoxyribonucleotide substrates. Computational simulations of the function of this pathway support our analysis and indicate that a source of deoxyribonucleotides in addition to those provided by mitochondrial salvage is essential to account for the experimentally observed mtDNA replication duration of 1 to 2 hours in cycling cells. Km values were obtained from the literature [8], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] or the BRENDA database [23]. For most enzymes, we could only find a single report of kinetic parameters. For the nucleoside kinases DGUOK and TK2, we did find multiple reports of kinetic parameters. In these cases, we selected the reference providing the most comprehensive information. To compute kcat values, we first obtained reported Vmax values [8], [12], [14], [15], [16], [18], [19], [20], [21], [22] and molecular weights [13], [22], [23], [24] of the various enzymes from the literature or the BRENDA database. If the enzyme was reported to be a multimer, we added the molecular weights of the subunits to calculate the molecular weight of the holoenzyme. The quantity kcat/Km (M−1 s−1) was calculated from the reported values of Vmax/Km (with units of µmol min−1 mg−1 µM−1) using the following conversion, where Wenzyme is the enzyme molecular weight. Reported values for Km, Vmax, and the calculated kcat/Km values are provided in Table S1. Values for the concentrations of the deoxyribonucleoside, deoxyribonucleotide, ribonucleoside, and ribonucleotide substrates were used to calculate ‘ (substrate) Concentration/ (substrate) Km’ ratios. These values were used for a comparison of activities of the enzyme with different substrates and are not meant to be precise. Instead, rough order-of-magnitude concentration values were used to compare values for this ratio, which often varies by several orders of magnitude within a single enzyme for different substrates. Literature reports suggested that mitochondrial dNTP pools are higher in actively cycling cells compared to quiescent cells [25], [26], [27], [28]. We assumed a 10-fold lower concentration of deoxyribonucleotides in quiescent cells, and chose 10 µM and 1 µM as reasonable representative estimates of mitochondrial dNTP concentrations in cycling and quiescent cells respectively. The basis of these estimates are the concentrations calculated from published values in HeLa cells [29] and quiescent fibroblasts [30] respectively. We used a value of 0. 82 ml/g mitochondrial protein [31] to calculate concentrations from the measured pool sizes in HeLa cells, and we used the value of 92. 3 µm3 for mitochondrial volume per cell [32] to obtain the concentrations from the measured pool size in quiescent fibroblasts. For simplicity, we assumed ribonucleotides and deoxyribonucleotides to be equally concentrated in the three phosphorylation states (mono, di, or tri-phosphorylated). Again for simplicity we assumed all four nucleotides (dAXP, dCXP, dGXP, dTXP where X = phosphorylation state) to have equal concentrations. Nucleoside concentrations were assumed to be equilibrated between plasma, cytoplasm, and mitochondria and set at a constant 0. 5 µM using a reported value for plasma concentration [33]. Lower nucleoside concentrations have also recently been reported [34], [35]. We have kept the higher value in our analysis since this is the most conservative choice. Lower nucleoside concentration values would make the problems that we point out in this analysis even more severe. Ribonucleotide concentrations were assumed to be constant and set at 100 µM, that is, one order of magnitude higher in cycling cells and two orders of magnitude higher in quiescent cells compared to deoxyribonucleotide concentrations. This is a fairly conservative (i. e. low) choice for the ribonucleotide concentrations. For other special cases of substrates (such as dUMP, dI, or IMP) concentrations data are not readily available so we again assumed low concentration values for these substrates. The complete list of assumed concentrations is provided in Table S1. In the case that we could not find Ki values of for enzyme inhibitors, we assumed competitive inhibition so that the Ki for the inhibitor was set to be equal to the Km for that chemical as a substrate. Inhibition kinetics data [8], [12], [16], [19], [21], [22], [36] are provided in Table S1. Our group has previously published a computational model of mitochondrial deoxyribonucleotide metabolism [25]. Parameter values for the model were based, whenever available, on published experimental values [8], [12], [14], [16], [17], [19], [20], [21], [22], [23], [24], [25], [36], [37], [38], [39], [40], [41], [42], [43]. As part of the present work, we updated the model to reflect the findings since the original model was defined. We refer the readers to the previous publication for a complete explanation of the basic framework of the model [25]. Briefly, enzymatic reactions were modeled with Michaelis-Menten equations (except TK2, which is modeled by the Hill equation) and rates of change of metabolites were modeled using ordinary differential equations. The updates to the model include adding (e. g. CMPK2) and removing (e. g. SLC25A19 or DNC) pathway components and updated kinetics (e. g. inhibition terms and kinetic constants). The model was written in Mathematica 7. The model files are available as supporting information (Text S1). The model constants are also available as supporting information in plain text (Text S2) and PDF (Text S3). Deoxynucleoside transport was modeled through the ENT protein as equilibrative between the cytoplasm and mitochondria. Thus, the net rate of deoxynucleoside transport was defined using the Michaelis-Menten equation as follows: where j represents the four deoxynucleoside species (dA, dC, dG, dT) and i represents inhibitors. Vmax and Km were taken to be the same for both directions of transport. The various enzymatic reactions (i. e. , phosphorylations and dephosphorylations) were modeled using the Michaelis-Menten equation. Thus, the reaction velocity waswhere S stands for substrate and [C] stand for the concentration of any competitive inhibitors. For the reaction of dT with TK2, the above equation was modified by raising the Km and [S] terms to the power 0. 5 to represent the Hill coefficient. The model of the mtDNA polymerization process was explained in the previous publication [25]. It models polymerization using fractions of the four deoxynucleotides in the mtDNA sequence, setting the prevalence of each base in the mtDNA light and heavy strands separately to match the prevalence in the rCRS reference sequence [44]. We have modeled mtDNA replication as asynchronous [45] using the locations of the origins of replication of the light strand and the heavy strand. Differential equations for the concentrations of the various metabolites were defined by adding and subtracting the relevant reaction velocity equations. For example, for dNMPs (deoxyribonucleoside monophosphates), the following differential equation models the rate of change of a particular dNMP: where NK represents the nucleoside kinase reaction, NT represents the nucleotidase reaction, NMPK represents the forward and reverse monophosphate kinase reactions. The kinetic constants and inhibition parameter values are available in Table S1. We used this updated model to test the hypothesis that a source of deoxyribonucleotides in addition to intra-mitochondrial salvage is essential for completing mtDNA replication in cycling cells in the experimentally observed time of 1–2 hours [45]. To be conservative, we set the ‘target’ replication time to be 2 hours (requiring an average replication rate = 33136 (nucleotides) /120 (minutes) = ∼276 nucleotides/minute). We ran simulations with a simulation time of 120 minutes (2 hours), with all dynamics including mtDNA replication starting immediately at the beginning of the simulation. We also tested a target replication time of 10 hours (requiring an average replication rate = 33136 (nucleotides) /600 (minutes) = ∼55 nucleotides/minute) – reasoning that in quiescent cells the time constraints for completing mtDNA replication may be more relaxed. Transport of deoxynucleotides from the cytoplasm to the mitochondrial matrix was modeled in a simple manner, by setting a constant production term of either deoxynucleosides, dNMPs, dNDPs (deoxyribonucleoside diphosphates), or dNTPs. Transport was modeled as occurring at only one phosphorylation level at a time, in order to assess the effectiveness of transport at each level. The essence of our simulation experiments was to test whether mtDNA replication was completed in the target time under varying levels of added molecules, including no addition, of various (A, C, G, T) deoxynucleosides and deoxynucleotides. We note that in principle the additional source of deoxynucleotides in this model does not necessarily have to be import from the cytoplasm, but could also be from other unknown intra-mitochondrial sources. However, considering the evidence that nucleotide transport does occur between the cytoplasm and mitochondria [5], [6], we assume that the additional source we have modeled corresponds to import from the cytoplasm. We tested multiple ‘transport profiles’. A transport profile is composed of simply the rate of the transported deoxynucleosides and deoxynucleotides. For each transport profile, we ran 100 simulations each beginning with a different, randomly selected (with uniform probability) set of initial mitochondrial concentrations of each deoxynucleoside and deoxynucleotide. As an initial test of the level of exogenous precursor transport needed, we set equal rates of import for all four (A, C, G, T) nucleosides (or nucleotides) at a particular phosphorylation level and then let the rate of import vary from 0 to 1200 molecules per minute, in increments of 100. Thus, for example, for testing whether transport of deoxynucleosides alone suffices, we ran 13 sets of 100 simulations. In each of those 13 sets, deoxynucleosides alone were imported at equal rates for each of the four nucleosides, in increments of 100 starting from 0 and up to 1200. Such simulation sets of 13 different import levels were conducted similarly for each phosphorylation level of the four deoxynucleoside species. The initial conditions of the simulations were set randomly with a uniform distribution over a set range. The allowed range (minimum and maximum) of initial deoxynucleoside concentrations was 0. 05 µM to 5 µM and the range of initial deoxynucleotide concentrations was 0. 1 µM to 10 µM. We set the concentrations of ribonucleosides, ribonucleotides, and non-canonical deoxynucleosides and deoxynucleotides to be proportional to the randomly selected dN and dNXP concentrations (see Table S1 for details), and held these concentrations (which only acted as inhibitors) constant throughout the time course of the simulation. The simulations were repeated 100 times with varying initial conditions. We extended the transport analysis further by obtaining the minimum number of molecules of each transported dNTP required for mtDNA replication to be completed in 2 hours (representing cycling cells) or 10 hours (representing quiescent cells). For the simulations to determine the minimum transport profiles, we tested whether the replication rate exceeded 55 (‘quiescent cells’, fixed initial concentrations: dNs = 0. 5 micromolar and dNXPs = 1 micromolar) or 276 (‘cycling cells’, fixed initial concentrations: dNs = 0. 5 micromolar and dNXPs = 10 micromolar) nucleotides per minute. We started at equal import of all four dNTPs at a rate such that replication would be completed in slightly less than the target time (2 hours or 10 hours). Next, we decreased the import of one dNTP at a time to check whether the target replication rate was observed. We continued this relaxation process until we obtained the minimum transport for each individual deoxyribonucleotide species necessary to support the target replication rate. In Michaelis-Menten kinetics, kcat and Km are the basic parameters of an enzyme-substrate reaction pair. The parameter kcat is the number of substrate molecules catalyzed per enzyme molecule per unit time and Km is the substrate concentration at which the reaction proceeds at half-maximal velocity. High kcat and low Km values imply a fast and efficient reaction, and thus, a high kcat/Km ratio indicate that this substrate is catalytically preferred by the enzyme. We searched the literature [8], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] and databases [23] and gathered the available data on the reaction kinetics of enzymes of mitochondrial nucleotide salvage. Figure 2A shows a plot of kcat/Km values. Each group of bars is for one enzyme, and within each group the bars are arranged from lowest to highest so that the best substrates lie to the right on each plot. For clarity, the substrates that are DNA precursors (presumed to the ‘proper’ substrates of these enzymes) are in green, and non-DNA precursor substrates are in red. The kcat/Km values cover a very wide range and so are plotted on a logarithmic scale. Figure 2A shows that each of these enzymes has significant reactions with non-DNA precursors. More importantly, except for TK2, none of the mitochondrial enzymes have DNA precursors as their preferred substrates, as seen from the fact that the substrates which lie to the right in each group of bars are non-DNA precursors. Prior work [46] has estimated the theoretical maximum of kcat/Km for an enzyme-substrate pair. This maximum is constrained by the diffusion limit, and was estimated to be ∼108 per M per second [46]. Compared to the diffusion limit, the kcat/Km values for reactions of the mitochondrial salvage pathway with DNA precursors are orders of magnitude lower (range = 888 to 5. 63×105 per M per second). In summary, in both absolute and relative terms these enzymes of the mitochondrial salvage pathway (with the possible exception of TK2) do not appear to be optimized for discriminating mtDNA precursor substrates from chemically related non-precursor substrates. To put the kcat/Km results in Figure 2A in perspective, Figure 2B is a plot of kcat/Km for the various substrates of the mitochondrial DNA polymerase gamma (POLG). In contrast to the enzymes of mitochondrial nucleotide metabolism, Figure 2B shows that, as expected, DNA precursors are preferably discriminated by POLG. This is true both absolutely and relatively. The kcat/Km values for the dNTP substrates approach the diffusion limit of ∼108 per M per s, and the values for dNTP substrates are many orders of magnitude larger than the kcat/Km values of the ribonucleotide substrates. GTP and UTP kinetics data are not shown because the POLG kinetics with these potential substrates have not been measured. While the ratio kcat/Km captures the efficiency of a reaction between an enzyme and a substrate, it does not take into account the expected physiological concentration of the substrate, which may vary by several orders of magnitude between ribonucleotide and deoxyribonucleotide substrates. The ratio of ‘ (substrate) Concentration/ (substrate) Km’ provides information that is complementary to that revealed in the previous section by the ratio kcat/Km. When the substrate concentration is much smaller compared to the Km, the enzyme is sensitive to substrate concentration and can thus operate at a range of velocities. However, the velocities in this range would be smaller than the maximum possible velocity. Depending on the relation between maximum possible velocity and the required rate of enzymatic output, substrate concentrations smaller than Km can be a detriment. This is the case for the mitochondrial salvage enzymes because mtDNA replication has to satisfy certain time constraints. We searched the literature [8], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] and databases [23] for Km values of the mitochondrial salvage enzymes for various substrates and their expected in situ concentrations. Figures 3A and 4A show a plot of Concentration/Km values for all of the enzyme-substrate pairs for which we could find data. As before, each group of bars is for one enzyme, and within each such group the bars are arranged from lowest to highest value of the ratio. Preferred substrates would be expected to have higher concentrations relative to the reaction Km and thus would fall to the right in each enzyme. Substrates that are DNA precursors are plotted in green, and non-DNA precursor substrates are in red. Figure 3A shows Concentration/Km values at higher mitochondrial concentrations (‘cycling cells’) of the deoxyribonucleotide substrates (10 µM). The Concentration/Km values for DNA precursor substrates range from 0. 001 to 0. 19. Thus, none of the reactions involving DNA precursor metabolism in the mitochondria would be running at maximal reaction velocity. In fact, since all reactions involving DNA precursor substrates have Concentration/Km values less than 1, none of these reactions would be expected to be running at even half-maximal velocity. It is apparent that these enzymes of mitochondrial nucleotide metabolism have significant affinities for non-DNA precursors. In many cases, the enzymes have higher affinities for non-DNA precursors than for the DNA precursors. In the case of the nucleoside kinases TK2 and DGUOK, although they have higher affinities for DNA-precursors, there is less than 10-fold difference from their preference of non-DNA precursors. For some reactions, the expected substrate concentrations are orders of magnitude lower than the reaction Km values (range: 0. 001 to 0. 19). The same trends exist in the values of Concentration/Km assuming lower mitochondrial concentrations of substrates (Figure 4A). Moreover, comparing Figure 4A (low deoxyribonucleotide concentrations) to Figure 3A (high deoxyribonucleotide concentrations), the disparity between DNA precursors and other substrates is more striking with an order of magnitude decrease in the Concentration/Km ratio of the DNA precursor substrates (range: 0. 0007 to 0. 19). This was expected as we assumed mitochondrial ribonucleotide concentrations to be constant and independent of high or low mitochondrial deoxyribonucleotide concentrations. To place these enzyme kinetics values in context Figures 3B and 4B show positive and negative examples justifying the principle of using the ratio Concentration/Km as a measure of substrate preference. The Concentration/Km ratio for dNTP substrates for POLG is about an order of magnitude larger than the ratios for reactions with ribonucleoside triphosphates (rNTPs) (Figure 3B). It is noteworthy that POLG is the only enzyme of the mitochondrial ‘salvage’ (DNA replication) pathway whose DNA precursor substrates have expected concentrations that are larger than the enzyme Km values. For our negative example, we considered SLC25A19 (formerly named the deoxynucleotide carrier (DNC), now identified as the thiamine pyrophosphate carrier) [7] to be a suitable choice. In contrast to POLG, the Concentration/Km ratios of DNA precursor substrates of SLC25A19 are low both in the absolute and the relative sense. Dolce et al [13] published data on the Km and Ki values of substrates (we used Ki as a proxy for Km if Km was not reported) that were tested for transport by the SLC25A19 protein, and it is seen in Figures 3B and 4B that DNA precursor substrates (green bars) are not the preferred substrates of this enzyme. Eventually, it was discovered that the function of SLC25A19 had been misinterpreted [7], [47]. When we compare the concentration/Km plots of the mitochondrial nucleotide metabolism (Figures 3A and 4A) to those of POLG and SLC25A19 (Figures 3B and 4B) we observe that the Concentration/Km values of DNA precursors with the enzymes of mitochondrial nucleotide metabolism are at the same level as the Concentration/Km values of the DNA precursors with SLC25A19, even though these DNA precursors are not the physiological substrates of SLC25A19. We note that SLC25A19 was not used a negative example in Figure 2B because the enzyme kinetics values (kcat) were not available for the relevant deoxyribonucleotide or ribonucleotide substrates. As a side observation, we are intrigued by the fact that at lower dNTP concentrations the Concentration/Km values for rNTPs are essentially equal to those for dNTPs for polymerization by POLG (Figure 4B). This observation reveals that discrimination by POLG in this case is perhaps almost completely dependent on the corresponding reaction Vmax. As the Vmax (or kcat) of rNTPs with POLG are much lower than those for dNTPs, it is possible that in quiescent cells POLG faces more interference by ribonucleotides, thus obstructing the polymerization of deoxyribonucleotides into the DNA molecule being synthesized and at the same time promoting the incorporation of ribonucleoside triphosphates in the DNA strand. This is consistent with the reported incorporation of ribonucleotides in replicating mtDNA [48]. As an initial analysis of the function of the salvage pathway, we used the Michaelis-Menten equation to calculate reaction rates under assumed substrate concentrations. We ignored the effect of inhibitions. This implies that the reaction rates we calculated (number of substrate molecules catalyzed per enzyme molecule per minute) were the upper-bound of the rates at the estimated concentrations, because inhibitions would act to lower these rates. We call such reaction rates ‘effective velocities’. As we assumed deoxyribonucleoside concentrations to be constant at 0. 5 µM and deoxyribonucleotide concentrations to be either 10 µM (high, ‘cycling cells’) or 1 µM (low, ‘quiescent cells’), we obtained two sets of effective velocities for the enzymes of mitochondrial nucleotide metabolism – one approximating the behavior in cycling cells, and one approximating the behavior in quiescent cells. Figure 5A is a plot of the effective velocities of nucleoside kinases versus NT5M. Remember from Figure 1A that NT5M is the nucleotidase that reverses the action of the nucleoside kinases, so the amount of material fed into the salvage pathway depends in part on the balance between these two groups of enzymes. The substrate dCMP is absent for Figure 5A because no reaction was observed between NT5M and dCMP [19]. Note that the nucleoside concentrations were assumed to be constant, so the high and low concentration rates are only given for NT5M. Because of inhibitions and competing reactions, the deoxyribonucleoside output from NT5M would be much lower than represented here, but it is still instructive to compare objectively the disparity between the forward and reverse reactions at the first phosphorylation level. It is clear that the theoretical maximum velocities (at the assumed concentrations) of NT5M reverse reactions are many-fold higher than the maximum velocities from nucleoside kinases. While the situation is poor for the dG and dT substrates, it is extremely poor for the dA substrate where the reverse reaction has well over an order of magnitude advantage over the forward reaction. Furthermore, NT5M may not be the only nucleotidase in the mitochondria, thus exacerbating this issue [19]. In addition to these qualitative comparisons of substrate preferences of mitochondrial nucleotide metabolism enzymes, we analyzed the reaction kinetics further to approximately quantify the flow of substrates through this enzymatic pathway. For simplicity, we ignored the inhibition terms in the Michaelis-Menten equations (inhibitions would further reduce reaction velocities). We could then investigate the effect of kcat, Km, and substrate concentrations on the upper-bound of velocity of the reactions at assumed substrate concentrations and compare the estimated velocities to the expected requirements for completing one round of mtDNA replication in a specified amount of time. It has been reported that one round of mtDNA replication in cell culture takes ∼1–2 hours to complete [45]. To be conservative, we assumed that mtDNA replication takes 2 hours to complete. To replicate 16,568 bases pairs on two mtDNA strands in 2 hours, ∼276 nucleotides are required per minute on average (with the log scale on Figure 5B it is unnecessary to precisely divide this quantity into the specific numbers of dATP, dCTP, dGTP, and dTTP molecules needed for the human mtDNA sequence). Figure 5B shows the effective velocities of some of the enzymes of mitochondrial nucleotide metabolism (DGUOK, TK2, CMPK2 and AK2). These are all the enzymes for which we found data that would enable us to calculate effective velocities. To facilitate comparison across these enzymes, some data for DGUOK and TK2 are repeated in Figure 5B from Figure 5A. As before, nucleoside kinase velocities in Figure 5B are the same for high or low concentration conditions because nucleoside concentrations are assumed to be constant. There exists a many-fold difference in the output of the four dNMPs, with dA nucleosides being fed into the salvage pathway by DGUOK at a rate many orders of magnitude lower than that required to support mtDNA synthesis. Assuming a 2 hour replication duration and an approximately 276/4 nucleotides per minute substrate requirement, the number of molecules of the DGUOK enzyme per mitochondrion required to catalyze the requisite output of dAMP is close to 3000. The poor kinetics of DGUOK with dA is not the only problem with the dA pathway. Although there could be multiple AK isoforms in the mitochondria, some of them are reported to be lacking kinase activity and none of them appear to catalyze dAMP phosphorylation with comparable efficiency to that of AMP phosphorylation [49]. This is verified for AK2 as seen in Figures 2A, 3A, and 4A. A calculation of dCDP production by CMPK2 at low assumed dCMP concentrations shows that more than 1000 CMPK2 enzymes per mitochondrion would be required to produce the necessary dCMP output per minute (assuming an approximate requirement of 276/4 nucleotides per minute). This result is important considering that CMPK2 expression was undetectable in many tissues [22], thus implying that CMPK2 function may not be essential for the production of mtDNA precursors as has been noted previously [22]. The data on the kinetic parameters of the human mitochondrial nucleoside diphosphate kinase (NME4 in Figure 1) is scarce (Km for dTDP of approximately 1 mM, which is 100 to 1000 times the physiological concentration of dTDP) [17], which is why it is not included in Figure 5B. An NDPK isolated from the pigeon mitochondrial matrix preferred ribonucleoside diphosphates over deoxyribonucleoside diphosphates by several fold [42]. Surprisingly, it appears that both AK2 [11] and NME4 [10] are localized in the mitochondrial intermembrane space, thus suggesting that if their reaction products participate in the mtDNA precursor synthesis, they would then have to be imported into the mitochondrial matrix. Although dAMP is not the preferred substrate for AK2 (Figures 2A, 3A, and 4A), AK2 still has a very fast reaction with dAMP (Figure 5B). Good efficiency with dAMP and the localization of AK2 in the intermembrane space instead of in the mitochondrial matrix seem to contradict each other regarding the role of AK2 in mtDNA precursor synthesis. To test our conclusions and to build upon them, we used an updated computational model to perform simulations of deoxyribonucleotide dynamics and mtDNA replication within the mitochondrion. Our comprehensive computational model allowed us to investigate the dynamics and origins of mitochondrial dNTPs. Our modeling is based on experimentally measured kinetics and model results enable us to quantitatively track the concentrations as well as the balance of the various deoxynucleosides and deoxynucleotides over time within an individual mitochondrion. Furthermore, the mitochondrial salvage pathway is complex and a systems analysis of this pathway as a whole is an important companion to the study of the individual enzyme kinetics. Figure 6 shows our simulation results. The X-axis represents the number of molecules of each nucleotide supplied to the mitochondrion in the form of a ‘source’ term in the differential equations in addition to the output from salvage within the mitochondrion. Each value on the X-axis is the sum of molecules supplied of all four species. For example, the X-axis value of 400 means that 100 molecules per minute of each of the four (A, C, G, T) species were supplied. The Y-axis represents the average (over 120 minutes of simulation time) mtDNA replication rate that we observed, calculated as number of nucleotides replicated divided by the time taken to replicate them. Initial values for the substrate concentration in the mitochondrion were randomly varied over a set range as described in the Methods section. Each Y value corresponds to the mean of 100 replication rates from 100 simulations with differing initial substrate conditions. The standard deviations were far smaller than the mean values (and are therefore not shown in Figure 6) indicating that the simulation was not sensitive to the initial substrate conditions. We compared the observed replication rates to those required to complete mtDNA replication in 2 hours (‘cycling cells’) or 10 hours (‘quiescent cells’). For the mtDNA length of 33,136 nucleotides (replicating both strands), these would be 33136 (nucleotides) /120 (minutes) and 33136 (nucleotides) /600 (minutes) respectively or approximately 276 nucleotides per minute and 55 nucleotides per minute respectively. Since the mean observed replication rates with no additional nucleotides supplied (0 on the X-axis) fall below the 2 hour line, it is clear that the output from mitochondrial salvage cannot account for an mtDNA replication duration of 2 hours. In fact, even when a 10 hour replication target was set, mitochondrial salvage alone is an inadequate source of dNTPs, though only a slight amount of additional substrate supplied by transport is needed in this case. Next, we note that both deoxynucleoside as well as deoxynucleoside monophosphate import are insufficient to support a 2 hour replication target. Transport of either dNDPs or dNTPs is sufficient to achieve the target replication rate. The profiles of dNDP and dNTP transport are indistinguishable from one another on Figure 6 because of the extremely fast kinetics of NME4. Transport of approximately 48 molecules per minute for each of the four nucleotide species was required to complete mtDNA replication in 2 hours. The longer replication time of 10 hours required a transport of 15 dNTP molecules per minute for each of the four nucleotide species. These rates were determined from the simulation by transporting all four nucleotide species at equal rates and with fixed initial concentrations (as described in Methods). We next addressed the question of the minimum transport of each dNTP species necessary to support the target replication. As described in the Methods section, the assumption of equal transport of the four dNTP species was relaxed to find the minimal amount of transport separately for each dNTP species required to meet the mtDNA replication rate goal. To achieve a replication rate of at least 276 nucleotides per minute (‘cycling cells’), 47,31,48, and 48 molecules per minute of dTTP, dCTP, dATP, and dGTP were required. Thus, for this condition of relatively fast replication, transport of all four nucleotide species at similar rates is necessary. The total dNTP transport rate sums to 174 nucleotides per minute, a large fraction of the 276 dNTPs per minute consumed by the mtDNA replication. For the slower mtDNA replication with a target of 10 hours, Figure 6 shows that a relatively small amount of transport of dNTP molecules per minute suffices. To achieve the replication rate target of at least 55 nucleotides per minute (representing slow mtDNA replication in ‘quiescent cells’), individual dNTP transport rates of 0,3, 8, and 15 molecules per minute of dTTP, dCTP, dATP, and dGTP were required. Due to the complexity of the system (a nonlinear one because of feedbacks and inhibitions), slightly different but often practically similar transport profiles were observed to result in similar replication rates. For example, for cycling cells the transport profile of 41,34,48, and 41 molecules per minute also achieved the replication rate target. For quiescent cells, the profiles of 2,2, 8, and 15 and 0,2, 8, and 15 molecules per minute (practically identical to the transport profile given above) also achieved the replication rate target. In summary, rapid replication of mtDNA requires a substantial additional source of all four dNTPs (or dNDPs) to supplement the limited kinetics of the mitochondrial salvage pathway. Under the conditions of quiescent cells, the primary requirement is for the transport of dATP and dGTP molecules, and the vast majority of the dNTPs consumed by the mtDNA replication can be provided by the salvage pathway. Based on this analysis of the enzyme kinetics three properties of the mitochondrial nucleoside salvage pathway are thus apparent: From the kinetics perspective mitochondrial nucleotide metabolism as defined by this set of enzymes (Figure 1) cannot be expected to be the primary source of dNTP substrates for the rapid replication of mtDNA molecules. Since ribonucleotides exist at higher concentrations than deoxyribonucleotides, enzymes that take both ribonucleotide and deoxyribonucleotide substrates will, in situ, not favor the catalysis of deoxyribonucleotides. This is certainly true for enzymes that possess higher affinities for ribonucleotides, but also for those enzymes that have only slightly better kinetics for deoxyribonucleotides. In these cases ribonucleotide substrates will simply out-compete the deoxyribonucleotides substrates owing to the relative abundance of ribonucleotides. One plausible interpretation of this analysis is that import of cytoplasmic deoxyribonucleotides is the primary source that supplies the direct precursors for the replication of mitochondrial genome while the mitochondrial salvage pathway acts as a back-up metabolism with a minimal role to play in cycling cells. The occurrence of deoxyribonucleotide transport between the mitochondria and cytoplasm and the substantial contribution of cytoplasm deoxynucleotides towards intra-mitochondrial dNTP pools have been demonstrated [5], [6]. Our results make it possible to comment on why this must be so, due to the kinetic properties of the enzymes of mitochondrial salvage. Our results also enable us to conclude that import of deoxyribonucleotides is in fact essential to support an mtDNA replication time of ∼2 hours. Furthermore, simulations based on these enzyme kinetics indicate that this import occurs either at the dNDP or dNTP level. In cells where cytoplasmic deoxynucleotide concentrations are low, mitochondrial salvage would assume a greater role and, in combination with some other supply such as RRM2B mediated reduction of ribonucleotides in the cytoplasm followed by deoxyribonucleotide transport into the mitochondrion, would produce the dNTPs for both the replication of mitochondrial DNA and perhaps repair of nuclear DNA. Possibly, the dNDPs produced by RRM2B activity might first undergo the terminal phosphorylation by NME4 in the intermembrane space (Figure 1B) and may then be imported into the mitochondrion matrix at the dNTP level to combine with the dNTP pool from intra-mitochondrial salvage. Indeed, this is consistent with defects in the mitochondrial salvage pathway having their most severe phenotype in post-mitotic tissues. That mitochondrial salvage has only a back-up role in supporting mtDNA replication is one explanation why DGUOK and TK2 deficiency phenotypes are tissue-restricted and not systemic. The kinetic characteristics of the cytoplasmic counterparts of mitochondrial salvage enzymes expose informative parallels and distinctions between the cytoplasmic and mitochondrial pathways of nucleotide metabolism. The good activity of the mitochondrial enzymes (except the nucleoside kinases TK2 and DGUOK) with ribonucleotide substrates implies that these enzymes might play as important a role in ribonucleotide production to support RNA synthesis as they do in supporting DNA synthesis. Based on our analysis, we would argue that future studies of the kinetics of the mitochondrial salvage enzymes would benefit from a broader characterization of the kinetics, particularly the activity of the enzymes with ribonucleotide substrates relative to the activity with deoxyribonucleotide substrates. The majority of the cytoplasmic counterparts also show a preference for non-DNA precursors (such as dUMP) and ribonucleotide substrates [12], [43], [50]. However, the role of nucleoside salvage as a source of dNTPs for nuclear DNA replication is generally assumed to be minimal. In the S-phase of the cell cycle, ribonucleotide reductase irreversibly converts ribonucleoside diphosphates to deoxyribonucleoside diphosphates, and subsequently, deoxyribonucleotides originating from de novo sources proceed to become the predominant precursors to nDNA replication. Thus, ribonucleotide affinities of these cytoplasmic enzymes not only provide the ribonucleoside diphosphates for ribonucleotide reduction but also ensure an adequate supply of RNA substrates. The terminal kinase (NDPK) of the cytoplasmic salvage pathway accepts both ribonucleoside and deoxyribonucleoside diphosphates, and the products can then be appropriately diverted for either RNA or DNA synthesis. Salvage enzymes of thymidine metabolism fit nicely into such a model - examples being the excellent kinetics of TK1 and TK2 with dT, and those of cytoplasmic deoxythymidylate kinase (essential for both salvage and de novo pathways of dTTP synthesis) with dTMP – since thymidine is not an RNA substrate and because of the crucial allosteric control exerted by thymidine nucleotides on ribonucleotide reductase [51] as well as feedback control on mitochondrial TK2 [21]. Such similarities in the enzyme kinetics of the parallel mitochondrial and cytoplasmic metabolisms lead to the question of a ribonucleotide reductase connection to mitochondrial nucleotide metabolism. Such a connection is hinted at by the data supporting a connection between the mitochondrial dNTP pool and the ribonucleotide reductase RRM2B [2], [6]. There has been at least one report of ribonucleotide reductase activity within the mitochondrion [52], though this has never been confirmed as far as we are aware of. Our simulations show that mitochondrial salvage is inadequate to account for the observed replication time of ∼1–2 hours in cycling cells. It is likely that the deficit is supplied by import from the cytoplasm. We propose that deoxyribonucleotide import into the mitochondria not only does occur, but is in fact essential to replicate and maintain mtDNA in cycling cells. Furthermore, in our simulations, import at the monophosphate level was not able to support mtDNA replication under the constraint of a replication duration of 2 hours or less. Our observation that either dNDP or dNTP transport are able to nearly identically support mtDNA replication is due to the extremely fast kinetics of NME4, the nucleoside diphosphate kinase. The fact that the NME4 kinetics for the conversion of dNDP to dNTP are fast lends weight to the hypothesis that transport occurs mainly at the dNDP level, and not at the dNTP level which would bypass the NME4 activity. Our results are not necessarily in disagreement with previous reports that observed that supplementation with external dA and dG or dAMP and dGMP rescued mtDNA depletion [3], [28]. In those cases, it was undetermined whether these externally supplied substrates changed their phosphorylation level prior to or after entering the mitochondria, or even the cell. In the study conducted by Saada [28], in patient fibroblasts harboring DGUOK defects while dGTP pools were reduced compared to controls, dATP pools were only moderately affected. In this study when these patient fibroblasts were given external supplementation of both deoxyguanosine and deoxyadenosine, mitochondrial dGTP notably increased, while the increase in mitochondrial dATP was less pronounced. Our observation on the inefficient kinetics of DGUOK with dA is consistent with these findings. We are not aware of any studies on the effects of pyrimidine supplementation in TK2 deficiency. We have chosen a somewhat arbitrary target replication time of 10 hours for the mtDNA in quiescent cells. It has been reported that even in quiescent cells (rat hepatocytes), mitochondrial DNA is subject to rapid turnover [53]. Moreover, it is plausible to suspect that long replication durations might compromise the integrity of either or both the template and the synthesis strand by increasing the probability of damage to the exposed DNA or unfaithful replication (deletions, frameshifting, etc). Therefore, it is possible that the mtDNA replication time may be practically constrained to a shorter duration than 10 hours. In that case deoxynucleotide import could be essential even in quiescent cells. There is a lack of data on mtDNA replication times in quiescent cells, a critically important gap in our knowledge since quiescent cells are the most severely affected cells in most forms of Mitochondrial DNA Depletion Syndromes (MDS). The fact that clinical conditions arising from altered intra-mitochondrial dNTP pools mostly manifest in postmitotic tissues is consistent with our results. The possibility of there being more than one deoxynucleotide transporter, say one for purine deoxynucleotides and one for pyrimidine deoxynucleotides, might explain why mutations in TK2 and DGUOK which are both nucleoside kinases produce different phenotypes. It is plausible that there exists more than one mitochondrial deoxynucleotide transporter whose expression levels, possibly in conjunction with other factors, contribute to tissue specificity of mtDNA depletion syndromes. There have been reports [54], [55] asserting a role of PNC1 (pyrimidine nucleotide carrier encoded by solute carrier family 25, member 33 or SLC25A33) in nucleotide import into mitochondria as well as mitochondrial maintenance. PNC1 was able to transport a variety of metabolites, including purine and pyrimidine ribonucleotides and deoxyribonucleotides, with a preference for UTP. Intra-mitochondrial UTP accumulation decreased in response to siRNA-transfection against PNC1. Mitochondrial ADP, ATP, and GTP levels were not significantly altered but the effect on dNTPs was not investigated. Suppression of PNC1 was associated with reduced mtDNA while overexpressed PNC1 was associated with increased mtDNA relative to controls. Since UTP is a cofactor of the mitochondrial helicase (PEO1 or twinkle), mtDNA levels might have been altered through increased or decreased UTP [54]. It is also possible that these consequences resulted from a lack of RNA primers or lack of mtDNA precursors that might be substrates of PNC1. However, PNC1 mRNA was undetectable in skeletal muscle [55], a tissue that is a target of TK2 defects. Interestingly, ribonucleotide reductase overexpression caused mtDNA depletion in skeletal muscle of mice [56]. Also, per mg protein, PNC1 appeared to transport roughly 1. 5 times more UTP compared to dTTP, the next most transported substrate. At this time, the role of PNC1 in transporting deoxyribonucleotides for mtDNA synthesis is inconclusive. Import of radioactively labeled dTMP into mitochondria has been observed [57]. However, it was also observed that a fraction of the labeled dTMP was degraded as well as phosphorylated in the growth medium, leading to the possibility that the transport of phosphorylated states other than the monophosphate may have occurred. A transport activity with preference for dCTP has also been observed [58]. It has been proposed that low basal TK2 expression in muscle renders the tissue vulnerable to TK2 defects, while overlapping substrate specificity of cytosolic dCK prevents mtDNA depletion from mutant DGUOK in tissues where dCK expression is high [59]. While mtDNA defects that have a basis in mutated salvage enzymes might conceivably be rescued by other factors such as overlapping substrate specificity of cytoplasmic enzymes, this hypothesis cannot account for phenotypes relating to POLG defects. Importantly, the fact that phenotypes from mutations in POLG are also tissue-specific and not systemic indicate that other factors, such as rates of mtDNA turnover or energetic demand of tissues might also be a factor in the basis of tissue selectivity. In a recent review, Liya Wang discussed deoxynucleoside salvage enzymes and their association with tissue specific phenotypes of mtDNA depletion [60]. It was hypothesized that since mtDNA turnover rates are different in different tissues and also because dNTP pools show organ-specific differences, it would be expected that the regulation of dNTP pools would also be different for different tissues. Because both muscle and liver have high amounts of mtDNA and also of mtDNA turnover, and since the dTTP pool is lowest in muscle and the dGTP pool is smallest in liver, it was proposed that these tissues would be especially vulnerable to mutations in TK2 and DGUOK respectively. Other contributing factors could include limiting RRM2B, thymidylate synthase, or nucleotide transporter activity. In our opinion, it is probable that there is more than one underlying principle that explains tissue specificity – vulnerability of tissues to mutations might be from a combination of various factors such as transcriptional compensation, turnover rates, energetic demand, etc and that different forms of mtDNA depletion syndromes may trace their etiology to different factors. Based on their experiments with perfused rat heart, Morris et al concluded that in isolated perfused heart, there is no de novo synthesis of dNTPs [61], stressing the importance of TK2 in rat heart. This could indicate that our observations on the inadequacy of mitochondrial salvage enzymes may not hold across all tissue types. It is also possible that the deoxyribonucleotide pools in rat heart arose in part through salvage mediated by residual TK1 activity. In a recent report of a TK2−/− H126N knockin mouse [62], the authors observed TK1 to be the main thymidine kinase component in heart, compared to TK2 in the brain. In this mouse, phenotypic manifestation of TK2 deficiency was related to TK1 down-regulation and transcriptional compensation. Although by postnatal day 13 both brain and heart had suffered substantial mtDNA depletion, in contrast to brain, heart was spared as respiratory chain proteins were still at normal levels in this organ when assayed at postnatal day 13. This could indicate a difference in the importance of TK2 in the heart tissue of rats compared to mice. A recent report claimed that a cytosolic localization of TK2 is present in many rat tissues [63]. For the knockin mouse [62], a compensatory mechanism involving increased mtDNA transcription through suppression of MTERF3 (mitochondrial transcription termination factor 3) expression was implicated in alleviating some of the effects of mtDNA depletion. It was unclear why dTTP but not dCTP levels were affected and whether cytosolic ribonucleotide reduction had any influence in this observation. In humans, even in quiescent patient fibroblasts with only 5–40% of residual TK2 activity, mitochondrial and cytosolic dTTP pools were unaltered [30]. This finding would be consistent with the possibility that the activities of mitochondrial salvage enzymes may not be strictly necessary even for quiescent cells. Alternatively, it is also possible for there to be practical important differences between species with regard to this metabolism. In their study of the rat heart, Morris et al [61] noted that although known as a substrate of TK2, dU was not converted to dUMP possibly due to ENT1 nucleoside transporter not being localized to mitochondria in rodents, unlike humans, suggesting that dU may not be transported into the mitochondria in rodents. It has been noted that genes involved in MDS (mtDNA depletion syndromes) etiology are essential for life in mouse models [60]. However, the severe phenotype of knock-out mice is not identical to the phenotype in humans [64], although multi-organ phenotypes have come to light in humans also [65]. This divergence could perhaps be due to species differences or because of the complete absence of enzyme activity in knock-out models [64]. One limitation of modeling biochemical pathways is that kinetic parameters as reported in the literature and obtained from recombinant enzymes may not reflect the in situ reality, for instance, if the enzyme conformation is unknowingly affected in the in vitro analysis, or if the assay conditions do not represent the cellular environment adequately. Similarly, we have relied on the literature and our judgment for selecting appropriate concentrations and enzyme copies within the mitochondrion. In our analysis we have assumed a nucleoside concentration of 0. 5 µM [33]. There have been reports of nucleoside concentrations of approximately 50-fold lower [34]. Such lower concentrations would have two effects on this analysis. First, the problems that we point out concerning the function of the nucleoside kinases TK2 and DGUOK would be even worse with significantly lower nucleoside concentrations. Second, there is a more subtle problem that the enzyme kinetics for TK2 were measured at much higher substrate concentrations (1 µM to more than 100 µM) [21]. If the true substrate concentrations were on the order of 10 nM, then the kinetics would have to be extrapolated to much lower concentrations, which could introduce additional uncertainty in the kinetic constants. Finally, our estimate of time taken to replicate mtDNA (2 hours) comes from a study of mouse cells [45]. It is worth mentioning that POLG kinetics suggest that polymerization itself is capable of proceeding at a rate much faster than 2 hours [14]. A more comprehensive investigation into mtDNA replication durations in a variety of human cells and particularly in the cell types affected by mtDNA depletion syndromes would thus be very beneficial. For simplicity, we assumed that only one mtDNA molecule is replicating at any given time in a particular mitochondrion. If two or more mtDNA molecules were replicating simultaneously, then the deficit in the required dNTPs would be even larger than our analysis indicates. It should also be noted that mitochondria are very dynamic and undergo continuous fusion and fission. However, the effects of fusion and fission on the mitochondrial dNTP content would most likely average out. While fusion of two mitochondria would result in a larger dNTP pool (measured as number of molecules per organelle), fission would result in a smaller dNTP pool. Since the known elements of the mitochondrial salvage pathway do not have sufficient enzyme kinetics to support mtDNA replication in the observed duration of ∼1–2 hours, then, an alternative source of mtDNA precursors must be essential. Despite the intensive focus of research on this pathway associated with mitochondrial depletion syndromes, it seems likely that our knowledge of mitochondrial nucleotide metabolism is still incomplete and that this pathway might need to be considerably expanded in the future to include new enzymes, mechanisms, nucleotide transporters and modes of regulation.

The powerhouses of human cells, mitochondria, contain DNA that is distinct from the primary genome, the DNA in the nucleus of cells. The mitochondrial genome needs to be replicated often to ensure continued generation of ATP (adenosine triphosphate) which is the energy currency of the cell. Problems with maintenance of mitochondrial DNA, arising from genetic mutations as well as from antiviral drugs, can lead to debilitating diseases that are often fatal in early life and childhood, or reduced compliance to therapy from patients suffering drug toxicity. It is therefore important to understand the processes that contribute to the upkeep of mitochondrial DNA. The activities of a set of enzymes, which together generate the chemical building blocks of mitochondrial DNA, are important in this regard. We used computational methods to analyze the properties of these enzymes. Results from our approach of treating these enzymes as a system rather than studying them one at a time suggest that in most conditions, the activities of the enzymes are not sufficient for completing replication of mitochondrial DNA in the observed duration of around 2 hours. We propose that a source of building blocks in addition to this set of enzymes appears to be essential.

Abstract Introduction Methods Results Discussion

enzymes metabolic networks nucleotides bioenergetics enzyme kinetics mitochondrial diseases theoretical biology biochemistry simulations metabolic pathways biology biochemistry biochemical simulations nucleic acids genetics energy-producing organelles metabolism computational biology genetics and genomics human genetics

2011

Enzyme Kinetics of the Mitochondrial Deoxyribonucleoside Salvage Pathway Are Not Sufficient to Support Rapid mtDNA Replication

The zebrafish Danio rerio is a powerful model system to study the genetics of development and disease. However, maintenance of zebrafish husbandry records is both time intensive and laborious, and a standardized way to manage and track the large amount of unique lines in a given laboratory or centralized facility has not been embraced by the field. Here, we present FishNET, an intuitive, open-source, relational database for managing data and information related to zebrafish husbandry and maintenance. By creating a “virtual facility, ” FishNET enables users to remotely inspect the rooms, racks, tanks, and lines within a given facility. Importantly, FishNET scales from one laboratory to an entire facility with several laboratories to multiple facilities, generating a cohesive laboratory and community-based platform. Automated data entry eliminates confusion regarding line nomenclature and streamlines maintenance of individual lines, while flexible query forms allow researchers to retrieve database records based on user-defined criteria. FishNET also links associated embryonic and adult biological samples with data, such as genotyping results or confocal images, to enable robust and efficient colony management and storage of laboratory information. A shared calendar function with email notifications and automated reminders for line turnover, automated tank counts, and census reports promote communication with both end users and administrators. The expected benefits of FishNET are improved vivaria efficiency, increased quality control for experimental numbers, and flexible data reporting and retrieval. FishNET’s easy, intuitive record management and open-source, end-user–modifiable architecture provides an efficient solution to real-time zebrafish colony management for users throughout a facility and institution and, in some cases, across entire research hubs. The fecundity, rapid development, external fertilization, amenability to both forward [1–7] and reverse [8–11] genetic approaches, conservation of core vertebrate protein-coding genes [12], small size, and inexpensive husbandry costs make zebrafish a powerful vertebrate model for studying human physiology and disease [13,14]. Moreover, their optical transparency [15–17] and readily available suite of genetic tools (Tol1 and 2 [transposable element] tranposase-mediated transgenesis, Gal4/UAS [yeast Gal4 transcriptional activator/upstream activation sequence] gene expression system, Cre/loxP [bacterial Cre recombinase gene/locus of X (cross) -over in P1] genetics, etc.) [18–27] and fluorescent reporters (e. g. , gCaMP [green fluorescent protein, calmodulin, and M13 peptide], lifeact fluorescent proteins, etc.) [18,28] have made zebrafish the premier model system for studying vertebrate biology in real time. These features, combined with their impressive regenerative capacity, also make zebrafish ideal for studying vertebrate tissue and organ regeneration [29,30]. Given these advantages, it is no wonder that fields as diverse as developmental neurobiology [31] and cancer [32] have leveraged the embryonic and adult zebrafish, respectively, to make valuable biological insights. As more and more labs embrace this model system and shared-use zebrafish “core” facilities continue to be built across the world, there has been an explosion in the amount of transgenic and mutant lines generated by the community, creating a significant need for accurate, automated, real-time colony management tracking software. This need for a centralized database is not only desirable at the level of an individual laboratory group but in some cases is necessary throughout a facility and institution or even across an entire research hub as researchers desire to build a more cohesive zebrafish community. Despite the importance of effective data management in animal research, many investigators employ handwritten notebooks or spreadsheet applications for managing small and medium-sized animal colonies. While these ad hoc data entry approaches offer the benefit of being simple to adopt, they do not scale with increased user numbers (even within a single laboratory), and they are not practical across an entire facility. Relying upon individuals to enter the correct line nomenclature (e. g. , allele) and accurate information (e. g. , sex, date of birth, etc.) unnecessarily exposes these “databases” to avoidable operator error. Additionally, paper records have the significant drawback that they cannot be accessed simultaneously by multiple users, they cannot be viewed remotely, and they can be misplaced or destroyed. On the other hand, electronic spreadsheet databases suffer from their own limitations, as multiple users cannot access them simultaneously in real-time. In addition, data stored in this format are challenging to mine; they typically lack a uniform/controlled lexicon, and best practice rules for data entry are absent, thus making them highly susceptible to operator error. Moreover, simple spreadsheets (e. g. , Excel) cannot navigate complex data structures, such as breeding schemes or pedigrees, as required in a model colony management software. Paraphrasing Silver [33] and as epitomized in the Jax Colony Management Software for mice [34], an ideal database would track (1) individual animals and their ancestors, (2) matings between animals, (3) progeny born from such matings (e. g. , litters) and the individuals within litters that are used in experiments, and (4) experimental materials (tissues and DNA samples) obtained from individual animals. Such a database should also format records so that determination of relationships between any components of the colony is possible (e. g. , which tissue came from which animal). Ideally, this platform would feature an intuitive graphical user interface, have a robust online community to support trouble shooting or modifications to the underlying architecture or functionality, and also be open source. A handful of software solutions specifically designed for managing zebrafish animal husbandry exist. A summary of some of the currently available zebrafish colony management applications, both commercial and open source, is provided in Table 1, while their functionalities are summarized in Table 2. Available solutions for data management also vary significantly in terms of cost, implementation requirements, flexibility, and level of user support. Additionally, few of these options are open source, and even fewer also record phenotyping and experimental data. These drawbacks, along with the cost of commercial options or the potentially intimidating learning curve associated with adopting open-source options—such as creation and maintenance of structured query language (SQL) databases, along with knowledge of programming languages like Python or personal home page (PHP) to manage them—may explain the reluctance of the zebrafish community to embrace these solutions. Here, we describe FishNET. FishNET is a comprehensive relational database application developed specifically to meet biomedical research community demands for a well-engineered, flexible database system supporting zebrafish animal husbandry and data management. FishNET meets the aforementioned criteria for an ideal record-keeping system with several added benefits, including a remotely accessible virtual facility view; a pedigree function; barcode-scannable labels for tanks, crosses, embryos, and fry; a calendar function that emails users for events such as line turnover or graduation of fry to the nursery; sick fish reports; records for genotyping protocols; and real-time records of all fish tanks, their use, and water quality in a facility for billing purposes and Institutional Animal Care and Use Committee (IACUC) reporting purposes. FishNET runs directly on Macintosh or PC computers using the FileMaker Pro Advanced (FMPA) application. Hosted databases can also be viewed remotely via web browsers on PC, Macintosh, or Linux computers by incorporating FileMaker WebDirect. Furthermore, it can be accessed on iPad and iPhones through the FileMaker Go application in the App Store (https: //apps. apple. com/us/app/filemaker-go-18/id1438460792? ls=1). The functional features of FishNET include support for controlled vocabularies (such as uniform allele nomenclature to avoid confusion regarding the origin of a transgenic or mutant line), multiuser capabilities, quick and accurate data reporting, pedigree tracking, animal husbandry workflow, sample tracking, and experimental data capture. The expected benefits of FishNET are improved vivaria efficiency, increased quality control for experimental numbers, and flexible data reporting and retrieval. FishNET is a freely available tool, and FileMaker software is a well-supported application with a large user base. Free trial versions of FMPA 17 are available at https: //content. filemaker. com/filemaker-trial-en-rf. FileMaker Go (for iOS mobile devices) is available at the App Store (https: //www. filemaker. com/products/filemaker-go/). A full database, as well as an empty shell, of FishNET is freely available at http: //www. wythelab. com/wythe-lab-databases (or via S1 File). Zebrafish protocols were approved by the IACUC at Baylor College of Medicine and the University of Texas MD Anderson Cancer Center. FishNET employs a normalized relational database that can be run through FMPA. The relational data model that underlies FishNET minimizes data redundancy, enforces data integrity, leverages controlled vocabularies to ensure proper allele designations, and enables accurate data retrieval against large data sets. The user interface enables addition of new functionality and enhancement of existing functions without major modifications to the basic infrastructure, enabling easy adaptation to user-specific needs. FMPA ensures safe multiuser concurrent data entry and editing in real time. Additionally, FishNET features a built-in barcode generator and barcode reader, and when these features are combined with a validated printer (as we demonstrate herein), together, they provide instant recognition and tracking of all fish stock information and related actions on a mobile device. An overview of the two primary configurations of FishNET is shown in Fig 1A. FishNET can be set up to run as a standalone database on a personal computer running a Windows or MacOS operating system. Alternatively, FishNET can be hosted on a central computer running a Macintosh or Windows operating system using FMPA Server software to support small- to medium-sized research groups in which FMPA client software is installed on all user desktop and laptop computers (e. g. , 1 license of FMPA per accessing computer). This same FMPA Server configuration also supports use of FishNET through the FileMaker Go App on mobile handheld devices with barcode readers running iOS 11. 2 or later (e. g. , iPad, iPhone). Handheld computers that have been tested include iPad Air 2 as well as iPhone 6s and iPhone X. In the server configuration, FishNET can be accessed—with appropriate user authentication capabilities—on any internet-capable device running Safari, Chrome, Internet Explorer, or Microsoft Edge internet browsers using FileMaker WebDirect. The cost of setting up and hosting FishNET will vary depending on the configuration. The number of users and their need to modify the underlying database architecture (e. g. , do they need to function as an administrator or simply upload, access, and download data) will dictate whether a standalone license is sufficient or whether multiple licenses that each connect to an FMPA Server (hosted via an on-premise server) or FileMaker Cloud (hosted remotely via FileMaker, not described here) are required. At this time, an annual FMPA license costs $540 for 5 licenses per year (academic pricing) with FMPA Server included, while a permanent individual instance of FMPA costs $324 and FileMaker Cloud costs $400 (user licenses not included). The cost to host FMP Server will vary by institution and will also depend on whether the institution hosts it on their own server (which requires Windows Server 2016 or newer) or if it is instead hosted locally by the laboratory (for instance, on a computer running MacOS Sierra 10. 12 [or newer] software). A guide to determining which license is appropriate for your group is provided in Fig 1B, as are images of the stationary and mobile stations in Fig 1C and 1D, respectively. A more detailed comparison between the costs of FMP Server and Cloud is provided in S1 Table. The base hardware requirements of FMPA and FishNET are relatively modest. FishNET runs smoothly on a late 2014 Mac mini with a 2. 6 GHz, dual-core Intel i5 (2278U) processor with 8 GB of 1,600 MHz LPDDR3 onboard memory, 3 MB virtual memory, and a 1-TB hard drive and 802. 11ac Wi-Fi wireless networking and Bluetooth 4. 0 wireless technology requiring 100–240V AC. While the database itself can be set up multiple ways, below, we describe two standard configurations. One option is an immobile station on a bench top within the facility room that requires a wireless barcode reader integrated with the computer (to read barcodes and enter data) as well as a label printer (for the racks, tanks, crosses, and fry) (Fig 1C). For multiple users, a more sensible configuration may be to locally host FishNET (this requires an FMPA Server license, which is included in the purchase of a team FMPA license—which covers 5 or more individual licenses). This set-up allows for simultaneous, multiuser access and data entry using either FMPA on the same network or through an internet web browser on the same network via FileMaker WebDirect. In this configuration, access on the same network could also be achieved on a handheld device using the FileMaker Go application (which is restricted to iPads and iPhones running iOS 11. 2 or above). Individual licenses for FMPA would only be required for users that want to easily print labels with barcodes and run the automated genotyping annotation script (see below) or those that need to modify the underlying database architecture. The hardware and costs for this configuration are outlined in Table 3. A second configuration is for a mobile station that can travel throughout an entire facility (Fig 1D). The costs and components of this set-up are shown in Table 4. A basic overview comparison between the computing software and hardware costs for all FishNET configurations is provided in Table 5. Regardless of the hardware set-up, the main drawback of using a commercial relational database platform is that while it is open source, non-coding–based, and user modifiable, a facility or laboratory must purchase a permanent or annual license to use FMPA. Fortunately, several flexible options exist, as outlined below. Additionally, the FileMaker Go App is compatible with Apple mobile devices (e. g. , iPhone and iPad). If FishNET will be hosted, then a computer with a Dual-Core CPU processor, 8 GB of RAM, and at least 80 GB of hard drive storage running Windows Server 2016 Standard Edition (with Desktop Experience), Windows Server 2012 R2 with Update, MacOS High Sierra 10. 13, or MacOS Sierra 10. 12 will be required. The FMP Server license is included when you obtain at least 5 licenses of FMPA, making this the logical choice for most user groups. To protect the user’s data, FMPA Server offers AES-256 encryption. It also supports third-party secure sockets layer (SSL) certificates to establish secure links between the server and web browsers. Furthermore, using an institution’s wireless local area network (Wi-Fi) to transfer files from the FMPA Server to individual users adds another layer of security (provided that the institution requires user authentication to connect to the network/has a secure firewall). A centralized database, such as FishNET, offers the advantage of simplifying data redundancy and mirroring (e. g. , creating backups). When using a single instance of FMPA on a single computer, a user or lab can back up the entire contents of the hard drive, along with the FMPA database file (s), using a standard backup program (e. g. , Time Machine, SuperDuper! , etc.), storing these copies on an external hard drive or remotely in the cloud. In the case of FMPA Server, the program by default creates a backup daily, storing the last seven backups (and successively rewriting over them from oldest to newest), with the option to keep specific backups at will. In this case, we suggest copying FishNET to an external drive. We have developed a system for positional identification of tanks and lines based on user-defined “Facilities, ” “Rooms, ” “Racks, ” “Tanks, ” and “Lines. ” The underlying infrastructure of this system is a relational database. A typical database is composed of tables of individual entries or “Records” (as they are referred to in FMPA) that contain data (in the case of FishNET, these data are information such as tank number, genotype, and date of birth). In a relational database, these records relate to one another through shared or common data (e. g. , tank identifier number, genotype, etc.), and as such, these records (and all of their associated data) can be sorted, queried, and viewed independently or in a list format. Tables of records within FishNET can be broadly grouped into seven categories (or “layouts, ” in FMPA terms): Tanks, Crosses, Lines, Harvests, Nursery, Labs, and Statistics. Information regarding the allele (Lines), parental strain (s) /mating history (Tanks and Crosses), progeny (Nursery), owner of the tank/IACUC number (s) associated with a tank (Labs and Tanks), usage (Harvests), and water quality/births/deaths (Statistics) are all stored in these various records, enabling robust and detailed records of zebrafish husbandry within a research group or across an entire facility. An eighth category, Virtual Facility, which contains unique identifiers for the facility, room, rack, row, and column of a given tank, creates a virtual facility containing room (s), rack (s), tank (s), and nursery of unique zebrafish lines, thus enabling users or administrators to view all occupied tanks within the facility (as well as the pertinent information for each individual tank: sex, age, genotype) and easily and efficiently track and locate tanks within a facility. A final category is the “Calendar” layout, which relates to all other tables (e. g. , Crosses, Tanks, Nursery, etc.). This category has email functionality to create reminders for graduating fry to the main system, turning over lines, or any other user-defined event. Finally, each tank, cross, and fry/larvae, as well as each rack, is uniquely barcoded to enable, fast, reliable data entry and functionality (such as moving tanks within or across racks, setting up matings, recording dead fish, etc.). Once you have determined the correct FMPA configuration for your needs (e. g. , locally hosted server, cloud, etc. [Fig 1A and 1B]) and purchased and installed the software, visit http: //www. wythelab. com/wythe-lab-databases and download the latest version of FishNET. If using a single standalone FMPA license, the user only needs to open the FishNETv1. fmp12 file and proceed with the set up. If using the FMPA Server, the downloaded file needs to be placed in the FMP Server database folder. In macOS, this folder is placed by default in /Library/FileMaker Server/Data/Databases, or in Windows, in C: /Program Files/FileMaker/FileMaker Server/Data/Databases. Below, we provide a logical, step-by-step guide for creating your virtual facility, entering in lines, and populating racks with these lines, an overview of which is provided in Fig 2A. An overview of how resources relate to one another within FishNET can be found in Fig 2B. In short, (1) a tank with adult zebrafish is given a tank unique identifier (TUID). If two adults are crossed to (2) produce offspring, a cross unique identifier (CUID) is assigned to the mating. Embryos resulting from this cross (3) are given a nursery unique identifier (NUID), and they can either be processed for experimental purposes (4) and be given a harvest unique identifier (HUID) or (1) graduated to the system and given a new TUID. After depositing the downloaded file in the appropriate folder (as outline above), there needs to be a one-time activation in the FileMaker Server Admin console to bring FishNET online. This can be done by selecting the database and selecting “Open” (Fig 3A). In order to add the database to individual FMPA instances, users need to select “Add App From Hosts, ” then add the host using the IP address of the server. This can be found in the FileMaker Server Admin console (Fig 3B). Upon opening the software, a window asking for username and password will appear. Enter “admin” as the username and “admin1234” as the password to access the database. Upon selecting sign in, the landing page should be visible (Fig 3C). Here, you will see a directory below the main FishNET icon. You can also select from the pulldown “Layout” menu in the upper left-hand corner for navigation. From the “Landing Page, ” or “Home, ” you will first select File/Manage…/Security. By default, an “admin user” is set up. If the database will be shared between different laboratories, we recommend setting up a password to access the administrator account because this account can access all records in the database and can also change the underlying architecture of each layout, whereas the individual labs will only see and modify records that belong to their respective laboratories and are prevented from modifying database-wide records (see Video 1: https: //youtu. be/BwCD6-r6MM4). Creating labs and individual users will be addressed later. After generating an administrator profile, the next step is to configure the facility (or facilities). This functionality is restricted to the administrator of the database and not individual users. From the landing page, select the “Configuration” button. In this layout, in the first column, the name of the facility needs to be entered (Fig 4A). In the second column, the different types of tanks being used are specified by the size and number of spaces they occupy on the rack (s). Following the same nomenclature, indicate the different tank capacities (for instance, if you are using a 5-liter tank, enter “5L”) ordered from smallest (nursing tanks) to largest. The “Spaces Occupied” column is used as an approximation of the size of the tank in the zebrafish rack. Fig 4B shows an example set-up of a Tecniplast rack and the corresponding spaces occupied for tanks in each row. This information is used to create a customized virtual rack view. After this initial configuration has been created, the number of rooms within a facility must be specified. This can be done by selecting “Virtual Facilities” from the landing page; upon clicking the button, a list of facilities should be shown in a pop-up scroll-down menu (Fig 4C). Selecting a facility will direct the user to the “Facility View” layout. Here, the option to “Add Rooms” can be found (Fig 4D). Selecting “Virtual Facility” from the main menu bar can also direct the user to the rest of the facilities where rooms need to be set up. It is important to note that in order to use the Virtual Facility function, all Facilities and Rooms need to be properly labeled prior to inputting any records. This cannot be changed later without having to change all records linked to a Room or Facility. Finally, once the Facility and number of Rooms have been specified (and named), the number of racks per room must be entered by selecting the “View Room” button (which takes users to view an individual room) (Fig 4E). When viewing an individual Room, a user can select “Add Rack” to create digital racks that will reflect the physical room. For each rack that is added, the administrator will choose from predefined rack configurations corresponding to the current major commercial aquatic habitat manufacturers (e. g. , Techniplast, Aquaneering, and Pentair) (Fig 5A). Pictures of the different aquatic systems can be viewed in the “Rack Layouts” section. Selecting one of these options creates a default layout of tank spaces matching these normal vendor configurations (Fig 5B). Upon selecting a rack layout, a virtual view of the rack will be shown; in the Virtual Facility View, individual positions can be labeled with barcodes by selecting the “Print Rack Labels” function (Fig 5C, indicated with red arrow). If the facility has a unique or custom rack layout, then select the “Custom” option. To set up a custom rack configuration, select the number of tanks in each row (A, B, and so on) by entering in the top number, then specify the number of spaces they occupy by entering in the lower number (Fig 5D). Repeat this for each row on the rack. The “Facility, ” “Room, ” and “Rack” categories provide spatial organization to the database to create a “Virtual Facility” that can be explored by the administrator, animal husbandry staff, and scientific users alike (more functions within this framework will be discussed after lines and tanks have been introduced). The categories of “Lines” and “Tanks” will be superimposed upon this user-defined framework later (see Video 2: https: //youtu. be/6aNV2MroMc0). Next, individual laboratories and users within these laboratories will be entered into the database. Before beginning this section, have the IACUC/animal protocol number (s) for the lab as well as the email address, phone number, and complete name of each user within a laboratory group readily available. The contact information for labs and users will later be assigned to individual tanks on the racks, as well as all crosses and any fry to be raised, allowing for automated reminders about graduating fry to the nursery (or the main system), dead fish notifications, or providing details for contacting other users regarding available lines within the facility. To create a laboratory group, first select the “Labs” button from the landing page (or dropdown layout menu). Once you are in the “Laboratories” layout, click the “+ New Record” button at the top of the FileMaker task bar (Fig 6A). Add all of the pertinent information to this empty field to create a laboratory. Once the lab has been entered, you can then populate it with users by selecting the “Lab Members” button. Within the “Lab Members” layout, click the “New Record” button to add users (Fig 6B). This information will later be used to establish ownership of individual tanks, crosses, and embryos within a facility. Finally, each lab should enter in a valid IACUC protocol number and protocol duration. This information will be associated with all tanks, crosses, and actions of users within a laboratory (and present on all tank, cross, and nursery labels) (see Video 3: https: //youtu. be/vfBigAefij4). By default, all new laboratories have access to view and modify the complete database. Below are a series of permissions that an administrator can restrict if they wish to limit individual laboratory members’ access to certain features or functions within FishNET (otherwise, ignore the following steps). Specifically, the following commands will allow individual users to “See and Modify” all information (e. g. , genotype, number, sex, location) for tanks owned by all users in a common laboratory, but tanks from other laboratories within the facility (including administrators’ stock tanks) are hidden and the records locked for modification. Administrators, such as animal facility technicians, can view all information and record mortalities and health problems and can work with any tank (e. g. , to graduate fry or to retire a sick tank). For this type of access, first create a new user or group account (Laboratory) with restricted access to the database by selecting File/Manage/Security. Then, select “New Account” and make sure the option to authenticate via FileMaker File is confirmed (Fig 6C) (note that the “Account Name” must match the name of the laboratory within FishNet). Type a password for the new account, then select “Privilege Set” and choose “New Privilege Set. ” A pop-up window will open, and within that window, change the privilege set name to the name of the laboratory. After the name is changed, make sure all the data access and design options are set to “All Modifiable, ” then click Records and select “Custom Privileges” (Fig 6D). A new window will be opened where you can see all of the tables that store information in the database. Each table will have a different setting. Here, you can select “Fish Lines, ” then change the “Delete” option to “No. ” Select “Tanks” and change the “View” option to “limited, ” then enter the following text into the pop-up window: Laboratories: : Lab Name =" Wythe" , replacing “Wythe” with the desired laboratory. Repeat this same operation for the “Delete” option. In order to complete the set-up for each new laboratory, the administrator must set up the appropriate permissions for all data tables. For a fully functioning lab, refer to Fig 6D and modify each field as shown. Follow the previous instructions for every instance of “limited” privileges, referring to Table 6 for necessary settings (see Video 4: https: //youtu. be/NrI7TOiSCvs). Lines, in this case, refers to distinct alleles that follow nomenclature established by the zebrafish information network (ZFIN) (https: //wiki. zfin. org/display/general/ZFIN+Zebrafish+Nomenclature+Conventions). Novel lines, such as those generated by Clustered Regularly Interspaced Short Palindromic Repeats/CRISPR-associated protein-9 nuclease (CRISPER/Cas9) mutagenesis, can be also be entered in FishNET. To enter an allele, from the landing page, select the “Lines” tab (or select the “Fish Lines” layout from the Layout pulldown menu) (Fig 7A). If your database has no lines entered, you will be taken to a blank record. Under “Record View, ” you will only be able to view one line or record (e. g. , a unique allele) at a time. Use the forward and backward icons at the top upper left of the database window to scroll through the records manually (Fig 7B, indicated with blue arrow). To create an allele in “Record View, ” populate the following fields in the blank record: locus, allele name, nickname, line unique identifier (LUID), lab of origin, line ZFIN ID, originating strain, PUBMED ID, laboratory the line was received from, cloning notes, and general notes (Fig 7B). Importantly, whatever designation is used for the “Allele Name” field will be redeployed throughout the database in dropdown menu choices, ensuring uniform nomenclature within a research groups, across other labs, and throughout entire facility (see Video 5: https: //youtu. be/ONgTj538AmU). Each line that is entered is assigned an LUID number (beginning with L). Each subsequent record is automatically assigned the next available LUID number (L + 1) (Fig 7B). FishNET also allows for users to upload and store attachments associated with each line, such as associated publications, representative images of example phenotypes, or expression patterns that can function as a reference for sorting offspring (Fig 7B, indicated by asterisks). Additionally, one can link genotyping protocols to individual lines such that the protocol can be accessed by selecting the “Genotyping Protocols” layout or button within the “Record View” field (these will be discussed in more detail in the following section). While users or owners of the line are not defined when a line is created, later, if tanks are populated with a line and placed on a rack, they will be visible in the “Lines: Record View” window under the “Tanks Available” heading. Only live tanks will appear within this window. Clicking on the magnifying glass provides a detailed view of the tank, including its genotype, date of birth, sex, number, the owner of the line, its status (e. g. , juvenile, adult), location in the facility, and the paternal and maternal animals that generated the tank (Fig 7B). To view all lines that have been entered into a database, within the “Lines” layout, select “List View” (rather than the individual “Record View”) (Fig 7C). This generates a simple, scrollable list of every allele, including the locus, allele, and ZFIN ID. To define a new locus in the “List View” layout, simply input text into the space provided (or select “New Record”), and the new locus will be indexed and will appear as an option in any subsequent input. Enter the correct genetic manipulation, locus, allele, and ZFIN information, and the rest of the database will populate the allele name using this designation. For adding additional details, select the “Record View” layout and finish entering in all pertinent information. To view all lines that are actually available within a facility or user group, select “Compact View. ” Users can navigate this view using the “Sort By” button to segregate the animals by any of the following criteria: genotype, facility, lab, or user. This view will show users the TUID, facility, and lab that own any tank (s) of a given genotype. Additionally, users can click on the magnifying glass for a detailed view of that particular tank (location, parents, age, etc.) (Fig 7D). To create a genotyping protocol, from the landing page, select “Genotyping Protocols” (Fig 8A, indicated with red arrow), then select the “New Record” button. Here, you will name the protocol and enter the ZFIN identification number and any notes, as well as PCR primers and thermocycler conditions, expected product size, and a representative agarose gel image, by dragging and dropping any. PNG or. JPEG file into the PCR example container (Fig 8B). For optimal viewing, gel images should be resized to 4 inches width and a resolution of 300 pixels/inch. To scroll through available protocols, use the FileMaker forward and backward keys available at the top of the screen or use the search window. The underlying logic of FishNET that enables the tracking and searchable features relies upon a TUID that is assigned to every new tank. From the “Tank List” layout (Tanks > List View or Landing Page > Tanks), one can see a list of all tanks, each with a TUID, stored within the database (Fig 9A). All pertinent information (sex, genotype, age, owner, and location) are present in this view. To create a tank, click the “Detailed View” tab (or, from the main dropdown menu, select “Tanks” and go to “Tanks Detailed View”). In that layout, select “New Record. ” Then, select the loci and the appropriate allele from the dropdown list (that should have been populated by following the instructions outlined in Entering Lines). After this, select the “User” of the tank (this will autopopulate the user’s lab, email address, institution, phone number, and IACUC information). If the TUID of the maternal or paternal parent tank is known, enter it now to establish a pedigree for this new tank. The TUIDs can be entered manually, or you can select the field (s) with your cursor and then scan a barcode for the animal of interest. Be sure to enter the date of birth, tank size, and number of fish for the TUID record. The status of the line will be color-coded according to the date of birth (fish younger than 2 weeks are white, fish older than 2 weeks but less than 3 months are yellow, fish older than 3 months but younger than 1 year are green, fish older than 1 year are red, and retired or euthanized tanks are black). Additionally, the database automatically creates a turnover date for 1 year after the date of birth. Users can choose to receive an automated email reminder for line turnover by selecting the “Set Reminder” calendar button above the “Turnover Date” field (but this must be done after establishing the User/Owner of the tank to set up an email address for the reminder). Finally, a user must assign a location for the tank by selecting the “Facility” dropdown menu and entering in the Rack, Row, and Column for the new tank. After entering that information, select the “Add to Rack” button to place the tank in the virtual facility. Alternatively, one can click the “Add to Rack” button in the header field and then scan in the barcode location on the rack followed by the barcode on the tank to establish the location of the tank (see Video 6: https: //youtu. be/fp1ZA1NSeEI). Once the user adds the tank to a rack, selecting the “Locate Tank” button will show where the tank in question is located within the virtual rack window (Fig 9B). Each mating generates a CUID number to track every mating and to generate pedigrees. To create a cross, select the “Crosses” button from the header or landing page menu (or select “Cross” from the pulldown menu). Then, in the header menu, select “Set Up Cross. ” This option can also be found in the Tank Detailed View (Fig 10A, top). A pop-up window will appear asking for the paternal TUID and the maternal TUID (Fig 10A, bottom). These can be entered either manually by text or by reading the barcode of each parent tank. Next, a pop-up window will ask what type of mating is being setting up: a bulk cross, a trio mating (defined as one male crossed to two females), or a pair. The next window will ask how many of this type of mating are being set up (e. g. , 8 pairs). After filling in this field, the application will show you a mating label while asking you if you want to repeat the cross. If you are only setting up one type of cross (e. g. , just pairs) and selected the appropriate number, then chose “No. ” If you wish to set up another type of mating from this cross, such as a few trio tanks, then select “Yes. ” The cycle repeats until you chose to no longer repeat this cross. After choosing “No, ” the screen will return you to the “Crosses” layout. To simplify the view, you can choose to “View Active Crosses” rather than “View All Crosses. ” In the header field, you will now have the option to print mating labels for each of the individual mating tanks that you set up. These labels display the genotype, sex, TUID, and location of the parent tanks, as well as the owner of the cross. If additional mating labels are needed, select the “Print Additional Labels” field in the header. We also use these to label 10-cm plates of embryos from crosses (and thus print extra labels). Active crosses that are less than 1 day old will be labeled green in the “Crosses” layout, while active crosses more than 1 day old will be labeled red to indicate that the fish should be returned to the system (or fed). Retired crosses will be colored gray (and users should retire every cross once the mating pairs are taken down and returned to the system). This convenient color-coding system enables easy visual tracking of active crosses within a facility or user group (Fig 10B) (see Video 7: https: //youtu. be/1EkTnV2AfdI). FishNET also has the capacity to track all larvae that are destined to be graduated to the main system, enabling real-time tracking of all tanks on a nursery, as well as the success of fish survival rates, through the “Statistics” button within the “Virtual Facility” layout. To raise fry from a cross, select the “Raise Fry from Mating” button in the header menu (Fig 10B). A pop-up window will ask you for the date of birth for the fry, as well as the number of fry for a single tank to raise (this will help determine survival on the system, as discussed later). Within the “Nursery” layout, each tank of fry is assigned an NUID. After filling in this information, a new pop-up window will ask if you wish to raise more fry from this cross (e. g. , additional plates/tanks). At day 5, select the “Graduate to System” button in the top right of the header. Upon selecting this button (located in the header field of the “Nursery” layout), a pop-up window will ask the user to enter NUID (s) that will be used to form a new Tank (in the event that more than one dish of fry of the same genotype are used to create a new tank on the system) (Fig 10C, right). This automatically generates a new, unique TUID record that contains all of the previous NUID information (allele, owner, strain, etc.) (Fig 11A). Once a new TUID is created, a user may select the “Add to Rack” function, as described previously, to “place” the tank in the virtual main system (see Video 8: https: //youtu. be/9wATt56mVsg). The calendar layout/function (Fig 11B) enables users to access a facility-wide calendar with all related events (crosses set up; system reminders such as change baffle size, alternate food, graduate to main system; etc.). Selecting “Add Event” will take the user to a field where the title of the event, date, and facility are entered. Finally, the calendar function has an option for email reminders that automatically populates information using the laboratory user field (s). This function can also be used to copy administrators or animal husbandry staff on reminders. Another novel, to our knowledge, feature of FishNET is the ability to track the pedigree of embryos or adults that are used for experiments (as well as all experimental conditions) in the “Harvests” layout. This layout generates a novel HUID that pulls from either the parental cross CUID (for embryos) or the TUID (for an adult) to create a unique record with the date of the harvest, experimental treatment, and number of fish collected (Fig 12A). Additionally, using the “Create Resources” tab, one can generate detailed records for every embryo or adult within a harvest (Fig 12B). While nonessential for facility management, many individual labs may find this functionality helpful for tracking embryos or adults that have been harvested and stored in the freezer and for keeping track of sections, mRNA, or protein samples. Tagging samples with a unique Resource ID and labeling a sample “H0001-1” is easier and clearer than writing out an entire genotype, sex, date, and treatment on one slide or an Eppendorf tube. Furthermore, there will be a permanent record of the parents, animal, genotype, and treatment associated with this HUID. Additionally, images (such as genotyping gels) and other relevant data can also be pasted within these “resource” pages (Fig 12C). To automate annotation of genotyping gels, we created an Apple script (FishNET_Genotype. scpt file) that can take the record list from the harvest section of FMPA into Adobe Photoshop Creative Cloud 2017 to annotate images of PCR gels (this function is only available in Mac OS X). This requires a completed genotyping protocol (S1A Fig), which can be found under Lines/Genotyping Protocols. A special security permission is also needed to allow Apple scripts in FMPA: select File/Manage/Security… Select Extended Privileges and give full access to “fmextscriptaccess. ” To start, download the script (S2 File). Open the script in Script Editor and follow the simple instructions to adapt it to your computer (S1B Fig). We recommended adding a shortcut for the script to your menu bar. From the script button on the menu bar, select “Open Scripts Folder/Open Users Script Folder” and deposit the downloaded script file there (now the script can be run from the Script menu in the menu bar). Before running the script, a genotyping gel should be open and resized in Photoshop to 4 inches width by 3 inches wide, with a resolution of 300 pixels/inch (S1C Fig). Run the script and enter the HUID number you want to annotate. Make sure all names match the correct bands on the image, then select “OK” in the pop-up window asking, “Are ready to add the gel to the database? ” Select “Yes” to create a new genotyping record. Drop the new annotated gel image into the empty field and select the appropriate name of the genotyping protocol as seen in S1D Fig. The annotated image is now present in the detailed view of all records linked to the selected harvest ID (S1E Fig). Because many laboratory groups use Excel spreadsheets to manage their existing colony, we provide a simple way to import this information into FishNET. Once the laboratory, members, facility, and rooms are set up in FishNET, one can import preexisting line data. In Excel, ensure that the following columns are populated: Locus (Tg, Et, etc), Allele Name, Lab of Origin, and General Notes (Fig 13A and S3 File). After all fields are entered, save the document in Tab-delimited text (txt) format. To import these data into FileMaker, go to the “Lines” tab and select File/Import Records/File. Next, select the. txt file where the data were saved. Once selected, the user has to match the source field from Excel to the target fields in FileMaker Pro. Activate field mapping and importing by selecting all fields that should be imported (the middle symbol between “Source Fields” and “Target Fields” should change to an arrow) (Fig 13B). “Add new records” should be selected, as well as “Don’t import first record (contains field names) ” (Fig 13B). Similarly, it is possible to import data from a preexisting FileMaker database. In the original database, go to the layout you wish to export (Tank list, Lines, etc.) and select File/Export Records. Then, “save” the new file as an FMPA-type (fmp12) file. Next, select the fields to be exported and select “Export. ” Finally, to import these data into FishNet, follow the above instructions, selecting the new. fmp12 file as the source of import. We placed an emphasis on the reporting capabilities of the database, which allow for storing fish census and usage reports for the whole facility in an exportable, Excel-compatible format file. This graphical report provides an overview of all fish morbidities and mortalities, and individual fecundity records are maintained for each line across the entire facility. FishNET also tracks all animals, can generate user-defined fish use reports (as required for annual IACUC protocol reporting), and, as previously mentioned, generates graphical reports for water quality data. Within the “Virtual Facility” layout (Fig 14A), selecting the “View Stats” function (Fig 14B) will take users to a graphical report of the number of tanks and their status (i. e. , adult mating age, juvenile, older than 1 year) and the number of total fish and tanks (Fig 14C, top). Because of limitations in graphing capabilities within FMPA, we have opted to generate reports using Google Charts Application Programming Interface (API), a third-party software (as outlined below in the Importing Water Quality section). Selecting the “Vitality Summary” button in this layout will display the number of births across a rack, room, or facility, with weekly, monthly, or yearly reports (Fig 14C, bottom). These graphical reports for facility stats and fish stats are both populated by user-entered data from dead fish reports, mating records, and tanks that are retired. In this “View Stats” layout, because different systems have unique headers and labels for data, we have simplified uploading external data. The management and partitioning of users and labs in FishNET, combined with barcode labeling, will allow the PIs and animal facility managers (or staff) to track animal usage and husbandry and enable accurate billing and IACUC reporting in real time. FishNET can also alert users if fish require attention or care. A caretaker can create a follow-up report in a tank detailed view (Fig 15A). A list of all follow-up reports can be found in the tank list view by selecting fish reports (Fig 15B, indicated with red arrow). For laboratories running FishNET through FileMaker Server, it is possible to generate automated email reminders to users and/or facility administrators. If this feature is desired, the administrator must open the script workspace (S2A Fig, left red arrow). Here, find the script called “Send Follow-Up” (right red arrow), and within the script, select the configure button (yellow arrow). In the send mail option (S2B Fig), select “Specify” next to the simple mail transfer protocol (SMTP) Server option. A new window will pop up in which the user administrator must enter their lab or institutional SMTP Server information. For calendar reminders, this process needs to be repeated on the “Send Reminders” script (S2B Fig, right). Once that is finalized, the user needs to go to the Filemaker Server Admin Console (S2C Fig). Under Configuration/Script Schedules (red arrow), select Create Schedule (indicated with yellow arrow) and select Filemaker Script as Schedule Type and click on “Set Database” (S2D Fig, red arrow). In the new window, select FishNET, and then enter the administrator username and password (S2D Fig, right). From here, go to select script (S2E Fig) and pick “Send Follow-Up” from the list. Lastly, choose how frequently you would like to send facility users the emails. This process can be repeated for the calendar alerts by adding a new schedule and selecting the “send reminders” script. A graphical overview can be found in the “Mortality Summary” in the “Stats” category in the Facility, Room, or Rack layout (Fig 15C, left). Once a report is generated in the mortality summary, three pie charts give an overview of the cause and symptoms present in the fish, followed by charts with the total number of dead fish. This report can be generated weekly, monthly, or yearly. A weekly report that can be printed can be found in the “Tank List” view by selecting “Fish Reports/Dead Fish Reports” (S3A Fig). The user can select the facility and the dates to be included in the report (S3B Fig). Furthermore, a complete list of dead fish can be found in the “Dead Fish List” in the “Virtual Facility Room” list. We have enabled FishNet to generate water quality reports (e. g. , conductance, pH, etc.) using the interactive Web service, Google Charts. Because of the limited options for data display in FileMaker, we have instead used Google Charts to generate graphical reports because it is more flexible in terms of graphing options, colors, and data input options. To generate these visual reports, export existing system water quality data as a. CSV file from a rack or facility of interest. Then, open FishNET and navigate to “Virtual Facility” and the specific “Room” containing the rack that the data correspond to. Next, select “Virtual View” for the individual rack of interest, then, in the next window, select “View Stats” (Fig 16A). Once in the individual rack stats view and in the present month of interest for the data, select “Water Quality Report” and “create new record” to import data (Fig 16B). Two different fields with either “Days” or “Values” will appear for each parameter field (conductivity, pH, temperature, nitrites, nitrates, and chlorine). From the. CSV file, simply copy the date of data acquisition (dd/mm/yy, or just the day number) in the first field and enter the values in the second field (making sure that the number of days and values are correctly related to one another). If there are multiple values recorded on one day, the system will display a bar graph with the average values and corresponding error bars. During the course of optimizing FishNET, we tested numerous printer configurations to find a sturdy, waterproof readhesive label that is also barcode compatible in a size usable with both large and small aquarium tanks as well as 10-cm petri dishes for fry. After much trial and error, we recommend using a BBP33 Brady printer and permanent, barcoded polyester labels for each tank location label on a rack. Unfortunately, the BBP33 printer only supports Windows computers. While some third-party vendors have developed BBP33 drivers for Mac, those we tested were unreliable. Instead, as a workaround, we suggest installing Windows 10 using Parallels Desktop for Mac (https: //www. parallels. com/). As a cheaper alternative, one may use a Dymo 450 LabelWriter Turbo printer (Berkeley, CA, USA), but note that the labels are permanent and their removal may require scratching the tanks (details available, as well as printer layouts, upon request). Printing layouts for various labels are provided within FishNET under the main layout dropdown menu under the heading “Printing Layouts. ” From here, users can visualize all of the printer layouts for petri dishes (for embryos), fry on the nursery, small tanks, large tanks, racks, and crosses. Each of these layouts are modifiable, and the label size can be changed to one that best meets a user’s needs, including which information should or should not be present on the labels. To create barcodes in FishNET, the Code 39 Barcode Font from IDAutomation (IDAutomationHC39M Code 39 Barcode. ttf) is required in any computer printing barcodes. It can be downloaded for free here: https: //www. dafont. com/idautomationhc39m. font. Once the font is properly installed, the user needs to go to each printing layout (S4A Fig) and select “Edit Layout. ” In edit mode, select the barcode rectangle and change its font from Arial to IDAutomationHC39M (S4B Fig). A barcode should appear. Then “Exit” the layout and “Save” changes. Repeat this process with each printing template. After configuring a facility, within the “Virtual Facility” layout, a user can select their “Facility, ” then select an individual “Rack, ” and within this layout, a button to “Print Coordinates Labels” is present in the upper right area of the header. Selecting this function will create a pop-up window asking, “Which Row do you want to Print? ” After entering the desired row of the rack you are currently using (e. g. , “A”), then selecting “OK, ” the labels for one row will be printed. For labeling the tanks on the main system and nursery, as well as the petri dishes containing embryos, we prefer to use a removable label, which has the added benefit of reducing adhesive buildup on tanks while also eliminating errors in transferring information (such as date of birth, genotype, sex, etc.) that occur during manual transcribing of labels. For an example of a completely labeled rack and tanks, see S4C and S4D Fig. To make input of tank locations easy, a barcode-compatible function is ready using the “Add Multiple Tanks” function in the Virtual Rack template, in which a sequential series of pop-up messages will prompt the user to scan the tank barcode followed by the rack location barcode (printed above). A common error in FMPA (at least within our team) is that after importing data or deleting a record, a user would like to reset the TUID, CUID, LUID, or NUID number to a previous value (to eliminate gaps in numbering and ensure consistency). To do so, select “File” in the FMPA menu, go to “Manage, ” then select “Database, ” and a pop-up window will appear. Within this pop-up window, go to the “Tables” section and double click the table you want to edit (e. g. , “Fish Crosses”). Then, double click the unique identifier Field Name (e. g. , “CUID”), and a pop-up window with options for the field will appear. Then, select the “Auto-Enter” section. Here, the “Serial Number” option will have a checkmark, and below this option, the next serial value is defined (e. g. , “C0032” if you are in “CUID”). To change this next value, simply edit the number after the first character (e. g. , change “C0032” to “C0031” if you needed to delete one erroneously created cross record/CUID). After the changes are made, click the “OK” button to close the “Field Options” pop-up window, then select “OK” again to close the “Manage Database” window, and then finally save the changes. Despite the increasing demand to ensure rigor and reproducibility at each step of the research endeavor, a robust, affordable, and intuitive archival database for zebrafish animal husbandry records has not been developed and widely adopted by the zebrafish community. We have created a facile, network-accessible relational, open-source database that meets the needs of researchers, animal husbandry staff, and institutional animal oversight committee members alike that creates and preserves comprehensive, detailed records for an individual lab or entire zebrafish facility. Additionally, such centralized, comprehensive record-keeping and the data visualization tools contained within FishNET will limit unnecessary duplication of lines within and across laboratories, ensure timely line turnover, and should flag fish husbandry or facility-wide issues (e. g. , water quality, decreased fecundity, etc.) in real time, allowing for institutions to reduce the overall number of animals required for experiments and save critical research dollars. Using the open-source, non-coding-based FMPA platform, the underlying architecture of FishNET can be modified by any end user to meet their unique needs or directly expanded upon to increase database functionality. FishNET also scales according to user demand, as does the FMPA platform, enabling functionality for research groups as small as one laboratory or as large as an entire institution. Future updates to FishNET will be available at http: //www. wythelab. com/wythe-lab-databases. In regard to migrating data to updated versions of FishNET, FileMaker Pro offers a cost-effective data migration script package (FileMaker Data Migration Tool), available with the FileMaker Developer Subscription (https: //store. filemaker. com/product/FDS) for $99 a year. Additionally, several third-party solutions are available online as well, although we have not yet validated any such software solution. Continual upgrades, along with integration of improved FMPA software and computer hardware technologies, will ensure that this inventory system continually evolves to meet the needs of the zebrafish community.

FishNET facilitates remote tracking of individual zebrafish lines and links associated biological resources to enable robust and efficient colony management and storage of laboratory information related to zebrafish.

Abstract Introduction Materials and methods Results Discussion

fish ecology and environmental sciences methods and resources relational databases vertebrates computer hardware computers animals animal models osteichthyes developmental biology model organisms experimental organism systems molecular biology techniques embryos information technology genotyping research and analysis methods embryology computer and information sciences animal studies molecular biology zebrafish databases eukaryota water quality computer architecture biology and life sciences organisms

2019

FishNET: An automated relational database for zebrafish colony management

Globally, vaccine-preventable diseases remain a significant cause of early childhood mortality despite concerted efforts to improve vaccine coverage. One reason for impaired protection may be the influence of prenatal exposure to parasitic antigens on the developing immune system. Prior research had shown a decrease in infant vaccine response after in utero parasite exposure among a maternal cohort without aggressive preventive treatment. This study investigated the effect of maternal parasitic infections on infant vaccination in a more recent setting of active anti-parasitic therapy. From 2013–2015,576 Kenyan women were tested in pregnancy for malaria, soil-transmitted helminths, filaria, and S. haematobium, with both acute and prophylactic antiparasitic therapies given. After birth, 567 infants received 10-valent S. pneumoniae conjugate vaccine and pentavalent vaccine for hepatitis B, pertussis, tetanus, H. influenzae type B (Hib) and C. diphtheriae toxoid (Dp-t) at 6,10, and 14 weeks. Infant serum samples from birth, 10 and 14 weeks, and every six months until age three years, were analyzed using a multiplex bead assay to quantify IgG for Hib, Dp-t, and the ten pneumococcal serotypes. Antenatal parasitic prevalence was high; 461 women (80%) had at least one and 252 (43. 6%) had two or more infections during their pregnancy, with the most common being malaria (44. 6%), S. haematobium (43. 9%), and hookworm (29. 2%). Mixed models comparing influence of infection on antibody concentration revealed no effect of prenatal infection status for most vaccine outcomes. Prevalences of protective antibody concentrations after vaccination were similar among the prenatal exposure groups. These findings are in contrast with results from our prior cohort study performed when preventive anti-parasite treatment was less frequently given. The results suggest that the treatment of maternal infections in pregnancy may be able to moderate the previously observed effect of antenatal maternal infections on infant vaccine responses. The global burden of vaccine-preventable diseases remains high, especially among children under 5 years old, with 1 to 2 million deaths recorded annually [1]. The encapsulated bacteria Streptococcus pneumoniae and Haemophilus influenzae type B (Hib) cause the majority of child deaths from pneumonia, of which >99% occur in less-developed countries [2]. Widespread anti-pneumococcus and anti-Hib vaccination programs have been implemented by the WHO’s Expanded Program on Immunization, with reductions in rates of vaccine-targeted pneumonias and of carriage of vaccine-covered serotypes [3]. However, these successes often fail to highlight the fact that children in developing nations often have less robust responses to vaccines. This has been observed in several campaigns, for example programs administering Bacille Calmette-Guérin (BCG) for tuberculosis prevention [4], and those administering typhoid [5] and measles vaccinations [6]. A recent polio outbreak in Africa highlighted this problem, wherein a Nigerian polio epidemic spread to Ghana, Botswana, and Kenya, despite > 90% vaccination rates in those nations [7]. Many factors are likely involved in apparent reductions in vaccine efficacy, including nutritional deficits, cold chain problems, and incomplete coverage and/or uptake within mass immunization campaigns. However, a growing body of evidence supports the hypothesis that chronic parasitic infections may also influence response to vaccination. Intestinal helminth infections have decreased immunization efficacy in animal models [8], and in human studies, BCG and tetanus vaccine antibody responses have been found to be diminished in the setting of Schistosoma infections [9,10]. In addition, childhood responses to tetanus, Hib, and typhoid vaccination appear to be attenuated by malaria infection [11–13]. While the immunomodulatory effects of parasites have been extensively studied, the host-parasite relationship and its fetal effects during pregnancy are poorly understood. Our group has observed that chronic maternal parasitic infections can influence the developing immunity of the child in utero. Transplacental shift of parasite antigens exposes the fetus to materials that evoke immunomodulatory effects such that parasite-specific B- and T-cells are already present at birth [14]. In our previous 2006–2009 study, such parasite exposure was associated with a decreased response to Hib and diphtheria immunization in early infancy [15,16]. These effects appeared to be due to an “imprinting” phenomenon that positively or negatively skewed neonatal immune response to parasite-specific antigens [17], and in turn, may have had a bystander effect that impaired infant responses to unrelated antigens such as those found in early childhood vaccines [18]. In many countries, endemic parasitic infections remain a significant public health challenge. Prenatal screening and treatment for these infections is becoming standard for antenatal care based on WHO guidelines, but maternal parasitic infections continue to occur and/or relapse at significant rates. While vaccination against encapsulated bacteria is also becoming the norm in the developing world, there is potential that this effort may be hindered by prevalent parasite exposures. In our previous study of a 2006–2009 mother infant cohort, we found maternal infections during pregnancy to be associated with reduced infant responses to diphtheria toxoid and H. influenzae type B polyribitol phosphate (PRP) [15,16]. In the present 2013–2015 study, we revisited the effects of a mother’s prenatal parasitic infections on her infant’s response to early childhood vaccination, specifically, against the previously affected diphtheria and Hib, and in the wake of the introduction of 10-valent pneumococcal conjugate vaccine in the Kenyan schedule [18], against Streptococcus pneumoniae serotype vaccine antigens 1,4, 5,6B, 7F, 9V, 14,18C, 19F and 23F. Ethical approval of this study protocol was obtained from the Kenyatta National Hospital Ethical Review Committee (protocol #P85/03/2013) and from the Institutional Review Boards at Case Western Reserve University (IRB #01-13-13) and Stanford University School of Medicine (protocol #IRB-31468). Participating mothers provided written informed consent for their own participation and that of their infants. In this prospective cohort study, pregnant women were enrolled at the Msambweni County Referral Hospital antenatal clinic in Msambweni, Kenya, a predominantly rural area on the southern coast with high co-prevalence of parasitic diseases [19,20]. Enrollment occurred between July 2013 and July 2015. Mothers were followed until delivery, and their newborn infants were subsequently followed until up to three years of age. Inclusion and exclusion criteria and censoring events for the cohort study are listed below: Inclusion criteria: [Note: Pregnant women were allowed to enroll irrespective of their gestational age, although they could not enroll at delivery because we could not assure adequately informed consent. However, potential participants were strongly encouraged to come to the clinic for prenatal care early in the second trimester (ideally <16 weeks gestation) both to ensure adequate prenatal care for the mother and unborn infant and to provide sufficient follow-up time to collect antenatal study samples from the mother. Maternal cohort participation in antenatal clinic care ranged from 1 week to 35 weeks; the median was 15 weeks (interquartile range 11–19 weeks). ] Exclusion criteria: Censoring events for the mother and newborn (and, by default, the infant’s mother): Maternal participants were screened at each prenatal visit and delivery for malaria (by blood smear and DNA PCR), for S. haematobium (by urine filtration egg counts and plasma anti-soluble worm adult protein [SWAP] IgG4) [21], for intestinal helminths (by quantitative stool microscopy using the Kato-Katz method [22]), and for lymphatic filariasis (by detection of circulating serum filarial Og4C3 antigen) [23]. Mothers received intermittent preventive treatment for malaria in pregnancy (IPTp) with sulfadoxine-pyrimethamine at each monthly clinic visit, and were actively treated for any symptomatic intercurrent malaria with artemether-lumefantrine. Upon clinic enrollment, all mothers were dispensed an initial ‘preventive’ dose of mebendazole (500 mg). Any participants who tested positive for intestinal helminths received additional single 500 mg dosing with mebendazole within 72 hours. In accordance with Kenyan Ministry of Health guidelines in place during the period of the study, pregnant mothers were not given treatment for schistosomiasis or filariasis during the antenatal period. Blood, urine, and stool samples were collected from mothers and infants at all study visits, as well as infant cord blood. On delivery, infant anthropometrics and temperature were recorded. APGAR score was assessed at 1,5, and 10 minutes, and gestational age was estimated by dates and by the revised Dubowitz clinical measurement [24], as prenatal ultrasound examination was not available. Infants received standard immunizations per the Kenya Ministry of Health guidelines, including the pentavalent diphtheria-tetanus-whole cell pertussis-hepatitis B-Hib vaccine, and ten-valent pneumococcal-conjugate vaccine for serotypes 1,4, 5,6B, 7F, 9V, 14,18C, 19F, and 23F (Synflorix, GSK) at 6,10, and 14 weeks of age. Serum antibody levels to S. pneumoniae serotype antigens, Hib PRP, and diphtheria CRM197 toxoid were measured using a fluorescent multiplexed bead-based immunoassay (Luminex, Austin, TX) as previously described [25]. The twelve vaccine antigens were coupled to carboxylated microspheres and then incubated with patient serum. Beads with subject antibodies bound to the antigen were then quantified on a BioPlex MAGPIX multiplex reader (BioRad, Hercules, CA). A stock serum (composed of 007SP (anti-pneumococcal polysaccharide (PnPs) human serum [26] and 09/222 (anti-Hib human serum, NIBSC, UK) was used to create a standard curve for each assay run. IgG levels against the vaccine antigens were then determined by interpolation of the fluorescent signal observed for the subject sample within the standard curve of the stock serum. Results were analyzed using R statistical software (R Core Team, 2016). Based on mothers’ prenatal infection history, three classes of infants were compared: i) those of “uninfected” mothers, which was defined as women who had no evidence of any parasitic disease at any visit; ii) those of “infected” mothers who had one or more infections at any time; and iii) those of mothers who had “two or more infections, ” which meant having evidence of two or more concurrent infections at any one time before birth. Baseline characteristics were compared via unpaired t-tests, Pearson’s Chi-square testing or Fisher’s Exact test, as appropriate. Individual parasitic infections considered in the analysis included malaria, S. haematobium, hookworm, and/or filaria, as well as a category of “any soil-transmitted helminth” (i. e. , positive either for hookworm, Trichuris trichiura, Ascaris lumbricoides, or Strongyloides stercoralis). A linear mixed effects model was used to compare the effect of each infection, whether at any prenatal visit, at delivery, or at either time, on the trajectory of antibody concentration against each of the studied vaccine antigens. Antibody concentrations were log-transformed to achieve approximately normal distributions then a model was constructed for each antigen response with fixed effects for infection, gender, and visit number, as well as a random effect for unique subject identification number. Infection variables were considered having a significant main effect on the model prediction if they had a P value ≤ 0. 05. A second model was then constructed to examine the interaction between infection status and child age at the time of visit in predicting antibody concentration, using a fixed effect for interaction of infection and time point, along with the previously included variables. In this model, an interaction at a particular time point was considered statistically significant at P ≤ 0. 05. Correction for multiple comparisons was not included in this analysis. We determined the percentage of infants having protective antibody levels at birth, 6 months, and 24 months by enumerating the tested subjects who had antibody levels above threshold concentrations previously established as providing protection against invasive bacterial infections at each time point. The cutoff values were: ≥ 0. 35 μg/ml for pneumococcal serotypes, ≥ 0. 1 IU/ml for diphtheria, and ≥ 1. 0 μg/ml for Hib [27]. From July 2013 through July 2015, a total of 764 mothers were enrolled during prenatal visits, with 660 followed to delivery (Fig 1). Mean estimated gestational age (by last menstrual cycle) at the first prenatal visit was 23. 1 weeks (95% CI 22. 7–23. 5). During the course of their prenatal care, 686/764 (90%) of mothers received at least one dose of sulfadoxine-pyrimethamine as IPTp, and 709 (93%) received mebendazole for STH infections. Sixty-six mothers (8. 6%) received artemether-lumefantrine treatment for acute malaria during their pregnancy. After delivery, 576 mothers remained enrolled in the study; the 84 drop-outs were due to: i) delivery off-site (n = 40,6. 1%), ii) failure to collect samples at delivery (n = 18,2. 7%), iii) stillbirth or early infant death (n = 21,3. 2%), iv) antenatal maternal death (n = 1,0. 2%), v) preterm birth (n = 3,0. 5%), or vi) subject withdrawal from study (n = 1,0. 2%) (Fig 1). Among these 576 mothers, 566 (98%) had received antenatal IPTp antimalarial prophylaxis with sulfadoxine-pyrimethamine. As IPTp was dispensed at each ANC visit, the total number of doses taken during pregnancy varied with mother’s clinic participation. The median number of doses given was three (range = 0 to 7 doses). Sixty-six mothers (11. 4%) had been given artemether-lumefantrine at some point during pregnancy for symptoms of acute malaria. Eighty percent of mothers with detectable malaria parasitemia (N = 231) cleared their parasitemia with therapy. Additionally, 555 of the mothers (96%) received one or more doses of prenatal mebendazole, and 133 (23%) had received two or more supplemental mebendazole doses due to interval positive stool egg detection. Overall, 116/198 (59%) of women with detectable stool STH cleared their infections with therapy during their pregnancy. There were 592 infants born to the 576 cohort mothers, due to 16 twin births. Twenty-five infants (4. 2%) were lost to follow-up before the first immunization; continuing cohort infant participants were followed for a mean of 17. 6 months (SD 8. 2, range 1–37). In the data presented below, cord blood antibody levels of the newborns subsequently lost to follow up were included in the determination of the prevalence of protective levels of antibodies at birth. Baseline characteristics for the mothers fully tested for infection and their infants are shown in Table 1. Uninfected mothers were, on average, 1. 9 years older than infected mothers, and 2. 0 years older than mothers having two or more infections, but these differences were not statistically significant. Uninfected women also had a higher average body mass index [BMI] compared to infected mothers or polyparasitized mothers, but these differences also did not reach statistical significance. Monthly household expenditures, bed net use, HIV prevalence, primigravid status, maternal parity, and delivery hemoglobin were not significantly different among the groups. However, none of the parasite-uninfected mothers was HIV positive (compared to 5. 6% of parasite-infected mothers), and fewer of the parasite-uninfected mothers were in their first pregnancy (14%) as compared to infected mothers (24%). There were no significant differences between these groups for all infant characteristics measured at birth, including sex distribution, birth weight, birth length, APGAR scores, estimated gestational age (Dubowitz score), or rate of preterm births. Parasitic infections were common in the maternal cohort (see Table 2). A total of 433/576 (75%) of mothers of participating infants had had at least one parasitic infection during pregnancy, and 249 (43%) of the mothers were infected at delivery. Only 36 women (7. 2%) who completed all testing had no evidence of infection by any of the eight parasites tested for in our study. The most common maternal infection was malaria, with 231 (40%) of mothers found to be infected at antenatal visits, 49 (8. 5%) at delivery, and 257 (45%) at any time point of surveillance. The mothers’ next most common infection was S. haematobium with 253 (44%) found to be infected. There were 168 hookworm infections (29%), and 132 women were filaria-infected (23%). Soil-transmitted helminths, as a class, were also quite frequent, with 34% of women having evidence of at least one intestinal helminth infection. During prenatal testing, 233 mothers (40%) tested positive for more than one parasite and 66 mothers (11%) had more than one infection at the time of delivery. Unfortunately, a significant fraction of mothers did not provide stool (n = 200/576,35%), urine (n = 176,31%), or have S. haematobium or W. bancrofti blood testing (n = 136,24%) completed at delivery. Therefore, helminth parasite prevalence at delivery for these infections was calculated only for the group with non-missing data. In order to avoid any misclassification, mothers were considered as “uninfected” only if they had all parasite testing completed and were negative for all pathogens. As a result, 79 women who were negative only on partial testing were formally excluded from the “uninfected” group included in our analysis. Cord blood plasma was available from 573/592 (97%) of the infants followed after birth. In follow up, we were able to examine and test 567 infants at least one time point after birth. Of these, serum was available for 407/567 (72%) of infant subjects at 10-weeks of age, and 440 (78%) at 14-weeks. The 6,12,18, and 24-month follow up visits provided data from 462 (81%), 432 (76%), 322 (57%), and 202 (36%) individual infants, respectively. There were only few samples available to analyze from the 30 and 36 month visits, with only 69 (12%) and 9 (1. 6%) of the 567 infants represented at these time points. In comparing children whose mothers were infected (either during prenatal care or at delivery) to children of uninfected mothers, there was a significant difference in anti-pneumococcal polysaccharide (PnPs) 23F IgG response levels (p = 0. 047), as shown in Fig 2. For this antigen, the ‘infected’ group had 0. 13 μg/ml less antibody on average compared to the ‘uninfected’ group, with the largest differences noted after the age of 18 months. The effect of maternal parasitic infections was not consistent. We found opposite and significant differences between children whose mothers had prenatal or delivery infections and children of uninfected mothers in terms of their anti-PnPs 19F antibody levels, as shown in Fig 3. For this antigen, our linear mixed effects model indicated that the ‘infected’ group had higher anti-PnPs 19F concentration beginning at six months of age (p = 0. 007) and continuing until three years of age (p = 0. 006). Using the same modeling approach, we found a small but statistically significant (p = 0. 001) effect in anti-diphtheria CRM when mothers had infections at delivery, again with higher antibody levels in the children of infected mothers from six to twelve months of age. (Fig 4) We also examined the separate effects of individual infections and found that many of the more prevalent parasitic infections during pregnancy were associated with minor, mostly enhancing effects on infant anti-vaccine antibody levels after 18–24 months of age. The parasite exposures were malaria, schistosomiasis, filaria, hookworm, and an ‘any soil-transmitted helminth’ category. The magnitude of their effects was typically less than 0. 1 μg/ml IgG for any antigen (see S1–S5 Figs for details). In assessing the impact of two or more antenatal maternal infections (‘polyparasitism’), we noted non-significant increases in antibody concentrations in later infancy for antigens PnPs 5,7F, and 9V among children of infected mothers (S6 Fig). Of note, we did not observe significant differences in post-vaccination IgG levels among children of mothers with heavy (≥ 50 eggs/10 mL urine, N = 15) vs. light (1–49 eggs/10 mL urine, N = 36) vs. no S. haematobium infection (N = 383). In addition, there were no significant differences in infant responses between those with malaria-parasitemic mothers who did (N = 185) or did not (N = 47) clear their parasitemia during antenatal care. Finally, there were no differences in response between infants whose mothers did (N = 116) or did not (N = 82) clear their documented STH infections during pregnancy. We next determined the proportion of infants who achieved successful vaccine protection, i. e. , who had concentrations of antibody considered protective against invasive disease. (Fig 5, S1 Table). At birth, the uninfected group had a wide range of antibody concentrations against the 12 antigens tested, with rates of protective titers ranging from 3% for PnPs 1 to 83% for PnPs 14. At six months, PnPs 6B, 7F, 9V, 14,18C, 19F, 23F, and Dp-t-CRM antibodies were at protective levels for >80% of the children whose mothers were uninfected prenatally (range of 81–96%). By 24 months, these rates of protection decreased, with percent protective levels ranging from 0% to 67%, with protective rates <50% for six of the twelve antigens. However, the number of children whose mothers had been uninfected, and who had serum available for testing at this age, was low (S1 Table, N = 9). Patterns of protection for the maternally-infected and–polyparasitized groups were similar (Fig 5). The only significant differences found were lower rates of protection against PnPs 7F and higher rates of protection against PnPs 19F at age 24 months among children of infected mothers (S1 Table). This prospective cohort study examined the effects of maternal prenatal parasitic infections on later infant vaccination responses to diphteria toxoid and to antigens used in pneumococcal and Hib conjugate vaccines. It found limited influence of prenatal maternal infection status on infant post-vaccination antibody levels. In contrast to the findings in our previous 2006–2009 cohort study [15,16], most of the current 2013–2015 cohort’s infants with in utero exposure to parasitic infections had an increased or equivalent antibody response to nine out of ten pneumococcal antigens, to diphtheria, and to H. influenzae type B, as compared to children of uninfected mothers. Post-vaccination levels of IgG were significantly lower only for S. pneumoniae serotype 23F, the most common serotype found among local school age children [28]. Most of the observed differences in IgG levels between exposed and unexposed infants were transient during the first two years of life; when comparing putative protection rates based on threshold antibody levels, prenatally exposed infants were similar to their unexposed peers. These results are in contrast to previous findings of studies of the negative impact of concurrent or prenatal exposure to parasitic infections on vaccination efficacy, both from our group as well as others [9–11,14,17,18,29]. This was, however, the first study of the effects of parasitic infections in pregnancy on infant response to the S. pneumoniae conjugate vaccine. Of note, two other studies have shown increased vaccine antigen responses in infants exposed to maternal parasitic infections. Specifically, following BCG vaccination, IFN-γ response to PPD challenge was higher among infants with prenatal T. cruzi exposure [30], and plasma vaccine-specific IgA levels after oral polio and rotavirus vaccination was greater in infants of mothers with prenatal helminth infections [31]. An important consideration for the differences in results may be due to the fact that, in the current 2013–2015 cohort, nearly all of the mothers received more aggressive anti-parasitic therapy for malaria and intestinal helminths during pregnancy, whereas this was not the case in the previous 2006–2009 cohort. Because prophylaxis is now standard of care in Kenya, it was not possible to test the differential effect of preventive treatment on the immune response in the infants in a more robust trial design, but intercurrent treatment enhancements may have reversed the detrimental effect of prenatal parasitic infections that we previously observed [15,16]. It should be noted that certain infections such as S. haematobium and Trichuris would not have been effectively cleared by the anti-parasitic antenatal treatment with mebendazole given to the mothers in this cohort. However, the current, more aggressive IPTp dosing and follow-up treatments for detectable Plasmodium and STH infections were important secular changes between our earlier 2006–2009 cohort (which had limited IPTp and albendazole preventive therapy) and the 2012–2015 cohort studied here. In the 2006–2009 cohort, IL-10 immunomodulatory cytokine responses to filaria and to S. haematobium were significantly associated with reductions in Hib and diphtheria vaccine responses and/or duration of effect [16]. In current 2012–2015 cohort, prenatal prevalence of filariasis among mothers was 23%, lower than the 44% found in the 2006–2009 cohort [15], this effect likely related to a regional filariasis control campaign implemented during intervening years. The lower prevalence of filariasis and the more effective preventive/suppressive therapies used for parasite control may have reduced the impact of parasitic infections on Hib and Dp-t vaccination responses (as observed in 2006–2009 cohort [15,16]), taking them down to insignificant levels in the current cohort. Our study classified infection status based on positive testing without respect to clinical status. We used highly sensitive parasite testing for malaria, filaria, and urogenital schistosomiasis, which detects parasitic exposure but does not diagnose clinical disease. Most women in the study were asymptomatic at the time of testing, and it is possible that symptomatic disease would have had a more significant imprinting effect than we observed in our cohort. It is possible that the single stool exams used in our study may have missed very light infections with STH, which may have led to misclassification of infection status. However, there were no heavy STH infections detected in our cohort. In addition, we did not find significant differences between children of mothers who had heavy vs. light intensity S. haematobium infections, nor did we find differences between children of mothers who did or did not clear their malaria or STH infections with outpatient treatment. In clear contrast to our earlier 2006–2009 study [15], a randomized, placebo-controlled trial in Uganda investigating the effects of prenatal anti-helminth therapy with albendazole and/or praziquantel on infant responses to the pentavalent diphtheria-tetanus-pertussis-HBV-Hib vaccine found no effect of anti-helminthic treatment (or antenatal malaria or helminth infections) on infant antibody responses to the five antigens at one year of age [32]. A separate study in western Kenya [29] has examined the impact of third trimester Plasmodium and/or S. mansoni infections on later infant immune responses to parasites and to routine vaccinations. They observed no effects on tetanus or diphtheria protection at the 2 year-old follow up, but found a significantly lower level of protection against measles at this age. We agree with these researchers that the impact of antenatal parasitic infections is quite complex, and undoubtedly influenced by the type and severity of infections, the mother’s underlying nutritional and re-exposure status, and the type of vaccine administered (i. e. , whether peptide plus adjuvant, polysaccharide conjugate, or live-attenuated). It is even possible that that prenatal exposure to parasites followed by antiparasite treatment may result in enhanced infant vaccine responses (compared to no exposure) to some vaccine antigens, as we found in the current study. This study was limited in its analysis of antibody responses among the cohort infants due to the low number of mothers documented to be completely free of the parasitic infections of interest. Non-adherence to interval testing and losses to follow-up limited the available sample size for serologic testing. In our statistical model comparing the interaction of infection status on antibody levels at specific time points, the most significant effects were often found in the 30 and 36 month visits, when there were smaller numbers of observations available for analysis. This significance cannot be extrapolated to the groups as a whole, so that overall exposure class findings are less robust, i. e. , we are not able to reliably determine if there is a multi-year effect on vaccine antibody levels in parasite-exposed children. We had a much larger sample pool for earlier time points during the infant follow-ups, especially in the 6–24 month period, which is the critical period when maternally-derived passive protection declines. Although Fig 5B suggests that, at 24 months, among children of infected women (i. e. , those prenatally-exposed to parasites) there may be lower rates of protection against pneumococcal serotypes 7F, 9V, 18C, and 23 F and against Hib, the low number of unexposed children available for comparison does not allow for any clear statistical inference in this regard. Of note, our statistical analysis, which included multiple outcomes and multiple time points, did not adjust for multiple comparisons, such that some of the apparent differences reported may be due to random variation in outcomes. Future studies involving stratified enrollment based on more sensitive multiplex diagnostics and allowing for longer interval follow-ups will provide better power to answer this question. In summary, this study yields additional longitudinal information on the relative impact that fetal parasite exposures may have on vaccine efficacy and early immunity to encapsulated bacterial pathogens. As the global health community moves forward to provide safe and effective vaccinations for S. pneumoniae, diphtheria, and H. influenzae [3], it remains more important that we continue to refine our understanding of the ways in which these vaccinations are best administered. Because parasitic infections alter the expected immune response to vaccination in a number of settings [9–13,15,16,29], and because the regions that bear the brunt of mortality from encapsulated bacteria also carry the highest prevalence of parasitic infections, a more in depth knowledge of parasite-vaccine interactions is clearly needed. In the meantime, efforts to assure that adolescent and adult women are included in STH, filariasis, and schistosomiasis control programs should be strengthened, as this group is often wrongly excluded from preventive treatment because of fears about possible adverse drug effects in pregnancy. However, the safety and efficacy of treatment during pregnancy has now been established in placebo-controlled randomized trials [33,34], and expanded treatments for all women should contribute to global improvement in maternal health, as now prioritized by World Health Organization guidelines [35].

This mother-baby cohort study continued our investigations into the potential impact of a mother’s parasitic infection (s) during pregnancy on a baby’s ability to respond to early life vaccinations. In a rural Kenyan setting where malaria and helminth infections are common, we tested infants’ anti-vaccine antibody responses over time for up to three years of age after early vaccination against Streptococcus pneumoniae (the pneumococcus), diphtheria, and Haemophilus influenzae B (Hib). In contrast to the results for our previous 2006–2009 cohort, for whom antenatal parasite exposure reduced responses to diphtheria and Hib, our more recent 2013–2015 cohort did not show consistent evidence of an effect of antenatal maternal infection on subsequent infant vaccine responses. We conclude that the impact of antenatal infections on infant immune response can be mitigated, and that present-day screening and preventive therapies during pregnancy may have achieved this effect.

Abstract Introduction Methods Results Discussion

children medicine and health sciences immune physiology maternal health obstetrics and gynecology immunology tropical diseases parasitic diseases vaccines preventive medicine age groups women's health infants infectious disease control antibodies vaccination and immunization families birth public and occupational health immune system proteins infectious diseases proteins people and places biochemistry helminth infections physiology population groupings biology and life sciences malaria labor and delivery

2019

Parasitic infections during pregnancy need not affect infant antibody responses to early vaccination against Streptococcus pneumoniae, diphtheria, or Haemophilus influenzae type B

Conjugative transfer of the integrative and conjugative element ICEclc in the bacterium Pseudomonas knackmussii is the consequence of a bistable decision taken in some 3% of cells in a population during stationary phase. Here we study the possible control exerted by the stationary phase sigma factor RpoS on the bistability decision. The gene for RpoS in P. knackmussii B13 was characterized, and a loss-of-function mutant was produced and complemented. We found that, in absence of RpoS, ICEclc transfer rates and activation of two key ICEclc promoters (Pint and PinR) decrease significantly in cells during stationary phase. Microarray and gene reporter analysis indicated that the most direct effect of RpoS is on PinR, whereas one of the gene products from the PinR-controlled operon (InrR) transmits activation to Pint and other ICEclc core genes. Addition of a second rpoS copy under control of its native promoter resulted in an increase of the proportion of cells expressing the Pint and PinR promoters to 18%. Strains in which rpoS was replaced by an rpoS-mcherry fusion showed high mCherry fluorescence of individual cells that had activated Pint and PinR, whereas a double-copy rpoS-mcherry–containing strain displayed twice as much mCherry fluorescence. This suggested that high RpoS levels are a prerequisite for an individual cell to activate PinR and thus ICEclc transfer. Double promoter–reporter fusions confirmed that expression of PinR is dominated by extrinsic noise, such as being the result of cellular variability in RpoS. In contrast, expression from Pint is dominated by intrinsic noise, indicating it is specific to the ICEclc transmission cascade. Our results demonstrate how stochastic noise levels of global transcription factors can be transduced to a precise signaling cascade in a subpopulation of cells leading to ICE activation. Integrative and conjugative elements (ICE) are a newly recognized class of mobile DNA elements in prokaryotes [1]–[4]. ICE come in different families, represented by the host cell range and gene similarities, but all have a similar mechanistic ‘life-style’ [2]. Under most circumstances the ICE resides in one or more positions in the host chromosome like a prophage [5]. At frequencies of typically less than 10−2 per cell and under particular growth conditions or environmental signals ICE excise by recombination between short direct repeats at either end (within the attachment sites attL and attR) [6]–[8]. The double-stranded excised ICE can undergo DNA processing as for plasmid conjugation [9], and transfers a single-stranded ICE-DNA to a new host cell. In the new host cell the ICE-DNA is replicated and integrates by site-specific recombination between the ICE-located attP-site and the chromosomal attachment site attB [1], [2]. Interestingly, many ICE integrate in genes for tRNA [10] and ICE integrase sequences suggest phage ancestry [11]. ICE have attracted broad interest because, similar to plasmids, they can carry a large number of auxiliary genes in addition to the genes necessary for their basic functioning, which can provide selective advantages to the host cell. For example, several ICE carry genes for antibiotic resistance [12]–[14], for iron scavenging [15], [16], for diguanylate cyclases that can enhance host survival [17], for plant symbiosis [18] or for metabolism of chloro- and aminoaromatic compounds [19]–[21]. Although some ICE have been detected by their self-transferability, a large number of ICE-related elements with unknown mobility has been discovered through genome comparisons [22]–[24]. Some of those may be mobilized with help of other elements [25], but others may represent elements in retrograde evolution that once were capable of initiating conjugation, but which are now rendered immobile [23]. In more general terms one therefore often speaks of ‘genomic islands’ [1] or ‘regions of genomic plasticity’ [24], which include both ICE and ICE-like elements. Genome comparisons among closely related strains have suggested that a significant fraction (perhaps as much as 20%) of strain-to-strain variation may be due to the presence of different types of genomic islands [24], [26], [27]. Such comparisons have further implied that genomic islands are largely responsible for the adaptive capacities of prokaryotic species [28]. Although several ICE have been genetically and functionally characterized, and their general importance for bacterial evolution and adaptation is now widely appreciated, still very little is known about their cell biology [2]. One of the most intriguing aspects of the functioning of an ICE is its low frequency of conjugation (e. g. , 1% or less of a population of cells), which suggests that in only very few individual cells in a clonal population a decision is made to activate the ICE. The types of mechanisms and regulatory control that can achieve such low frequency differentiation are still widely unexplored. Some ICE bear regulatory systems controlling excision that involve phage-type repressors [29]–[31], which therefore may behave similar as the phage lambda bistable lysogenic/lytic switch [32]. Other ICE-classes, however, bear no gene functions with significant homologies to known phage lytic switches. Previously, we showed that excision and transfer of the element ICEclc in Pseudomonas knackmussii B13 must be the consequence of a bistable switch that culminates in the activation of the intB13 integrase promoter (hereafter named Pint) in 3% of cells during stationary phase [33]. ICEclc is a 103-kb sized element with strong homologies to a large number of genomic islands in Beta- and Gammaproteobacteria, and is named after its propensity to provide the host cell with the capacity to metabolize chlorinated catechols, encoded by the clc genes [21]. Two identical ICEclc copies reside in the chromosome of strain B13, which are interspaced by 340 kb (Miyazaki, unpublished). Activation of the intB13 integrase leads to excision and formation of a closed circular ICEclc intermediate [33]. Transfer of the circular intermediate is dependent on a DNA relaxase, which makes a single-stranded break, but, exceptionally, can initiate transfer at two origins of transfer (oriT) on ICEclc [9]. Single cell studies using fluorescent reporter fusions showed that Pint activation was preceded by and dependent on expression of a protein named InrR (for INtegRase Regulator) in the same individual cell (Figure 1). InrR is encoded in a small four-gene operon on ICEclc under control of another bistably expressed promoter (PinR) [33]. This suggested that ICE excision and activation in general may be the consequence of a bistable switch, and that the frequency of ON-setting is a determining factor for the frequency of ICE conjugation. Bistability as a phenomenon is most well-known from competence development and sporulation in Bacillus subtilis, which lead to phenotypically differentiated cells [32], [34], [35]. Although bistability is thought to originate from stochastic expression noise, this in itself is not sufficient to ‘lock’ cells in different phenotypic behaviour, but rather needs to be amplified and stabilized by regulatory mechanisms that include double positive feedback loops or double negative loops [32]. On the other hand, it is conceivable that the noisiness sets the threshold for the proportion of cells that display the bistable trait. The goal of the underlying work was to explore whether noisiness may lay at the basis of determining the proportion of cells in which ICEclc becomes active. We focused our attention on both Pint and PinR promoters, which are expressed during stationary phase and only in a subpopulation of cells [8], [33], [36]. Initiation of ICEclc transfer in stationary phase cells further suggested involvement of a specific sigma factor such as RpoS (σs). RpoS is the stress-starvation sigma factor that in P. aeruginosa controls the expression of some 772 genes at the onset of stationary phase [37], 40% of which have also been identified as quorum-sensing controlled. Deletion of rpoS in P. aeruginosa does not result in a dramatically changed phenotype, although such mutants survive 50-fold less well to heat and salt shocks than wild-type, and produce less extracellular proteins such as elastase, exotoxin A, and alginate [38]. In order to establish the role of a stationary phase sigma factor in activation of ICEclc, we identified an rpoS-gene in P. knackmussii B13 and studied the effects of interruption and subsequent complementation using single-cell reporter gene fusions to Pint and PinR. Interestingly, a B13 wild-type equipped with a second rpoS gene copy displayed a much higher subpopulation of cells expressing both Pint and PinR promoters. To study whether actually individual cell levels of RpoS could be somehow deterministic for the activation of ICEclc we replaced native rpoS by a gene for an active RpoS-mCherry fusion protein. Finally, we measured contributions of intrinsic and extrinsic noise on Pint and PinR promoters from covariance in the expression of double gene reporters placed in single copy on different parts of the B13 chromosome [39]. Our results indicate that individual cells with the highest RpoS levels in the population are more prone to activate Pint and PinR, which suggests that the stochastic variation in RpoS levels across a population of cells is transduced into ICEclc activation and transfer in a small subpopulation. In order to identify the rpoS gene of P. knackmussii strain B13 we used PCR amplification with primers designed against conserved regions in a multiple alignment of rpoS sequences of P. aeruginosa, P. putida KT2440 and P. fluorescens (Figure S1). The nucleotide sequence of the amplified fragment from strain B13 showed high homology to a set of rpoS genes from other pseudomonads, with a percentage nucleotide identity between rpoSB13 and rpoS from different P. aeruginosa strains of 83% over 989 bp. The predicted amino acid sequence of RpoSB13 positioned most closely to that of P. aeruginosa PAO1 (Figure S2). Flanking regions of rpoSB13 were subsequently recovered from a draft genome sequence of P. knackmussii B13 (Miyazaki, unpublished data), which showed that the rpoS region of strain B13 is syntenic to that in P. aeruginosa PAO1 with a gene for a lipoprotein (nlpD) upstream of rpoS, and an rsmZ-like gene and a gene for a ferredoxin (fdxA) downstream (Figure 1). We therefore concluded that this region in B13 most likely encodes a similar stationary phase sigma factor as in P. aeruginosa. A single crossover rpoS mutant was produced by marker insertion (strain B13-2671, Figure S3, Table 1). Despite repeated attempts we were not successful in producing a double recombinant with an internal rpoS deletion. However, it was possible to replace rpoSB13 by a gene for a RpoSB13-mCherry fusion protein (see below). Maximum specific growth rates of strain B13-2671 (rpoS) on MM with 5 mM 3CBA were similar as B13 wild-type (0. 22±0. 01 versus 0. 26±0. 01 h−1, respectively), but the onset of exponential growth was slightly delayed in B13-2671 (rpoS) (Figure S4A). Reversion to the wild-type allele occurred in less than 1% of cells in stationary phase (Figure S4B). The fact that most of the core genes of ICEclc are solely expressed in stationary phase P. knackmussii B13 cells [36], and the presence of sequence features typical for RpoS in the PinR promoter [33] had suggested an implication of RpoS in controlling ICEclc stationary phase expression. Inactivation of rpoS in B13 indeed resulted in reduced expression of both PinR and Pint promoters. This was evident, first of all, from a reduced proportion of cells in a B13-2673 (rpoS) compared to B13 wild-type population expressing eCherry and eGFP above detection threshold from single copy transcriptional fusions to PinR and Pint, respectively (Figure 2B, Table 2). Secondly, stationary phase cells of B13-2673 (rpoS) produced a lower average reporter fluorescence signal than wild-type cells (Table 2). In most individual cells the magnitudes of eGFP and eCherry expression correlated, confirming that PinR and Pint were expressed in the same cell (Figure 2B). Both eCherry and eGFP were not visibly expressed in B13-2673 (rpoS) cells examined after 24 h in stationary phase, but after 72 h a small fraction of cells still developed eGFP and eCherry fluorescence (Figure 2B). This delay (48 h) is much longer than would be expected from the slight growth delay (5 h) of B13-2673 (rpoS) compared to B13-78 wild-type to reach stationary phase (Figure S4). Late (72 h) expression of Pint and PinR in B13-2673 (rpoS) was not due to reversion of the rpoS mutation (Figure S4B). To confirm that the effect on Pint and PinR expression was caused by a disruption of rpoS, we complemented strain B13-2673 with a single copy mini-Tn5 insertion containing rpoSB13 under control of its own promoter (PrpoS, Figure 2A). Both the proportion of cells and their average fluorescence levels of both fluorescent markers from PinR and Pint were restored to wild-type levels in the rpoS-complemented strain P. knackmussii B13-2993 (Figure 2B, Table 2). The number of cells expressing autofluorescent proteins from both promoters was even slightly higher in the rpoS complemented strain than in B13 wild-type after 96 h in stationary phase, although this was not a statistically significant difference (Table 2). We can thus conclude from this part that, because both the expression level of eGFP and eCherry in single cells and also the percentage of cells that expressed both markers in strain B13-2673 (rpoS) was significantly lower than in B13 wild-type and the rpoS-complemented strain (B13-2993), RpoS is necessary for achieving native transcription levels from the PinR promoter (i. e. , within 48 h of stationary phase). On the other hand, RpoS is not absolutely essential, since cells with interrupted rpoS gene eventually (96 h) express PinR and Pint, which was not due to reversion of the rpoS mutation (Figure S4B). Since the observed lower expression from the integrase promoter (Pint) in the rpoS mutant could be the result of either less InrR being formed from PinR, or of a direct control by RpoS of Pint, we compared eGFP expression from a single copy Pint-egfp transcriptional fusion in B13, the B13 rpoS mutant (B13-2976) and a B13 lacking both inrR copies (B13-2979, Table 1). Interestingly, the proportion of cells expressing eGFP and their average fluorescence were much lower in a strain lacking both inrR copies than in the strain missing RpoS (Figure S5, Table 3), suggesting that the major influence of RpoS is indirectly via InrR. Since the proportion of cells expressing eGFP from Pint in an inrR−/− background was already so low, it was not possible to detect statistically significant differences to a strain that would carry the triple rpoS and inrR−/− mutations (Table 3). For this reason, we produced the triple rpoS inrR−/− mutation in a B13 strain containing a dual reporter of Pint-egfp and PinR-echerry (B13-3091), and correlated eGFP to eCherry expression. Since this strain would be devoid of InrR-mediated expression of Pint, we expected that expression of egfp from Pint in absence of rpoS would be lower than expression of echerry from PinR. Indeed, there was a slight tendency for the mean proportion of cells expressing eGFP (from Pint) in strain B13-3091 (rpoS, inrR−/−) to be lower than that expressing eCherry (from PinR), although this was only poorly significant after 96 h (P = 0. 04), again because of the very low subpopulation sizes (<0. 5%, Table 2). Purified and reconstituted RpoS-RNA polymerase from E. coli bound DNA fragments encompassing Pint in in vitro electrophoretic mobility shift assays (K. Globig and J. van der Meer, unpublished data). This suggests that transcription from Pint is both indirectly (via InrR) and directly dependent on RpoS. Whereas expression of the reporter gene fusions was interpreted as being representative for the behaviour of the native Pint and PinR promoters on ICEclc, we also determined ICEclc core gene expression and transfer frequencies from B13 wild-type or derivatives as donor and P. putida UWC1 as recipient. Expression of the ICEclc core genes in stationary phase cells measured by microarray analysis was lower (up to 27-fold) for both B13-2671 (rpoS) and B13-2201 (inrR−/−) compared to B13 wild-type (Figure S6). Interestingly, expression of the inrR operon was not only downregulated in B13-2671 (rpoS) but also in B13-2201 (inrR−/−) (Figure S6), suggesting autoregulation by InrR. Not only ICEclc core gene expression but also transfer frequencies were significantly lower at all time points from B13-2673 (rpoS) or B13-3091 (rpoS, inrR−/−) than from B13-2581 wild-type or the rpoS-complemented B13 rpoS mutant (B13-2993) as donor (Figure 3, Table S1). ICEclc transfer frequencies from the complemented B13 rpoS mutant were not significantly different than those from B13 wild-type. Transfer frequencies from B13-2673 (rpoS) as donor were significantly higher than from B13-3091 (rpoS, inrR−/−) as donor, but only after 96 h mating time (Table S1). These results thus corroborated that RpoS is favorable (but not essential) for expression of ICEclc core genes and thus for conjugative transfer. RpoS exerts its control mainly via its interaction with the inrR promoter, with InrR relaying the activation further to other ICEclc core genes, but also via direct interaction at Pint. Since in the absence of RpoS the proportion of cells expressing Pint or PinR in the population diminishes but not completely disappears, we wondered whether the levels of RpoS or the magnitude of rpoS expression in individual B13 cells are a precondition for cells to become locked in the PinR - Pint bistable ‘ON’-state. Expression from PrpoS is maximal at the end of the exponential phase and in stationary phase, as shown by the appearance of mCherry fluorescence from single copy PrpoS-mcherry and rpoS-mCherry fusions in B13-3165 or B13-3564, respectively (Figure S7), which coincides with the timepoint of activation of PinR and Pint. To correlate expression from rpoS with that of Pint or PinR in individual cells we created B13 derivatives with single copy PrpoS-mcherry and Pint-egfp or PinR-egfp fusions (B13-3183 and B13-3189, respectively). mCherry expression from PrpoS in stationary phase is normally distributed among all cells with a mean around 50 RFU (Figure 4A). In contrast, simultaneous eGFP expression from Pint-egfp or PinR-egfp in B13-3183 and B13-3189, respectively, occurs highly skewed in only 3% of cells (Figure 4A). However, there was no particular correlation between expression of mCherry and eGFP in single cells. To better account for post-transcriptional effects on RpoS expression we repeated the experiment with B13 derivatives expressing RpoS translationally fused to mCherry at its C-terminal end (RpoS-mCherry) from the original rpoS locus. This was done by substituting the native rpoSB13 by the rpoSB13-mcherry allele. Similar as B13 wild-type RpoS also RpoS-mCherry was expressed during stationary phase in all cells with normal distribution (Figure 4B), and eGFP was again expressed in 3–6% of cells in the population from either the Pint or PinR promoter (strains B13-3564 and B13-3555, respectively). RpoS-mCherry but not an N-terminal mCherry-RpoS fusion protein complemented B13-rpoS for bistable Pint or PinR-dependent eGFP expression (data not shown). This indicated that the RpoS-mCherry fusion protein functionally replaces B13 wild-type RpoS. Significantly, only B13-3564 and B13-3555 cells expressing the highest RpoS-mCherry levels had also activated eGFP from Pint or PinR, respectively, although not all cells with high RpoS-mCherry levels expressed high levels of eGFP (Figure 4C). This suggests that the RpoS level per se is not sufficient to elicit PinR or Pint expression but is a precondition for PinR- or Pint-expression to occur. To artificially increase RpoS expression more globally across all cells in the population, with the idea that this would precondition more cells to activate PinR and Pint, an additional rpoSB13 copy under control of its own promoter was introduced by mini-Tn5 transposition (B13-3260, Figure 5A). Strikingly, ∼18% of all cells in stationary phase cultures of B13-3260 (rpoS+) expressed eGFP from Pint and eCherry from PinR compared to 5% in B13-2581 wild-type (Figure 5B–5E). ICEclc transfer from B13-3260 (rpoS+) as donor to P. putida UWC1 as recipient was twice as high as with B13 wild-type after the same mating contact time, although this was not a statistically significant difference (48 h, Table S2). In contrast, B13 with an extra copy of inrR (strain B13-3257) did not significantly differentially express both reporter genes from PinR and Pint (Figure 5B–5E). To determine whether the higher subpopulation of cells expressing both Pint and PinR-promoters was due to a generally higher level of RpoS in cells, we compared the RpoS-mCherry fluorescence levels in B13 with native rpoSB13 replaced by the rpoSB13-mcherry allele (B13-3564) and in the same strain into which another single copy of rpoSB13-mcherry was transposed (B13-3712). Indeed, the mean mCherry fluorescence in B13-3712 was almost twice as high as in B13-3564 (Figure 5F), suggesting that in double-copy rpoS strains on average more cells became permissive and could induce PinR and Pint. In order to further examine how variability in RpoS levels would be linked to bistable expression of PinR and Pint, we measured the contribution of intrinsic and extrinsic noise on both promoters in individual cells. Noise was deduced from intra- and intercellular variations of reporter gene expression (eGFP and eCherry) from two individual single copy transcription fusions to Pint or PinR, placed at different positions of the B13 chromosome as suggested in Elowitz et al. [39]. Fluorescence intensities from eGFP and eCherry were recorded in three independent clones with different insertion positions of the reporter fusion constructs to avoid positional effects as much as possible. Both markers essentially expressed in the same subpopulation of cells (Figure 6). Interestingly, the total noise was significantly higher on the Pint promoter than on PinR (Table 4). Moreover, Pint expression was dominated by intrinsic rather than by extrinsic noise, which suggests that the variation in expression from Pint depends more strongly on variations in small numbers of regulatory molecules in individual cells, such as would be expected when Pint is at the end of a cascade involving InrR. Adding an extra copy of rpoS or of inrR under control of their own promoters into the double-Pint reporter strain resulted in a significant decrease of intrinsic and total noise compared to wild-type (Table 4), which was insensitive to the size of the sampled subpopulation (Table S3). This indicates that the relative contribution of the extrinsic noise on Pint expression becomes more dominant, as would be expected from the increase in a global transcription factor (since RpoS is also directly acting on Pint). Also adding an additional copy of inrR resulted in a lowering of the total noise, although the proportion of cells expressing eGFP and eCherry in the inrR+ strain was not increased compared to wild-type (Figure 6, Table 4). One of the mysteries in ICE gene transfer among bacteria is the mechanism that controls the (typically low) frequency by which they become excised in clonally identical populations of donor cells. ICE conjugation must start with its excision and therefore the cellular decision that determines conjugation is binary: ICE excision or not. Low transfer frequencies (e. g. , below 1% per donor cell in a population) suggest that the binary ‘ON’-decision is only made in a very small proportion of donor cells. Indeed, our previous results on ICEclc in P. knackmussii B13 using stable fluorescent reporter gene fusions at single-cell level had indicated that 3% of cells in stationary phase after growth on 3-chlorobenzoate (3CBA) as sole carbon and energy source measurably express PinR and Pint [8], [33]. Moreover, single cell activation frequencies are in the same order as measured ICEclc excision and transfer at population level [33]. Our results presented here show for the first time how the expression level of the global transcription factor RpoS in individual cells across a population can modulate the frequency of cells activating excision of the ICEclc element. By gene interruption and complementation we first establish that RpoS in P. knackmussii is a stationary phase sigma factor controlling transcription of the PinR- and Pint-promoters and thus, indirectly, transfer of ICEclc to P. putida. Addition of an extra rpoSB13 gene copy led to an increased proportion of stationary phase cells in which the PinR- and Pint-promoters are activated, which suggested that the expression level of RpoS is important for controlling the bistable switch leading to ICEclc activation. Indeed, by expressing an RpoS-mCherry fusion instead of RpoS wild-type protein in strain B13 we showed that PinR- or Pint-egfp expression in stationary phase preferably occurred in individual cells with the highest levels of RpoS-mCherry fluorescence (Figure 4C). Moreover, strains with two rpoS-mCherry gene copies produced on average twofold higher RpoS-mCherry protein fluorescence levels in cells, leading to an increase of up to 20% of cells expressing eGFP from PinR or Pint. This showed that an incidentally high RpoS level in an individual cell is a prerequisite for leading to PinR- or Pint-expression. On the contrary, having a high RpoS-level is not sufficient and an as yet unknown other ICEclc-encoded factor (s) must be responsible for the activation or derepression of PinR (Figure 7). We conclude that RpoS levels are a precondition for a cell or, in other words, a threshold, to activate the ICEclc bistable promoters during the first 2 days of stationary phase. This conclusion is further supported by noise measurements on the PinR or Pint-promoters (Figure 6). Intrinsic noise is dominant on the Pint promoter in wild-type B13, which would be in agreement with the major role played by (a low abundant) InrR and the relatively minor role of (a widely abundant) RpoS directly on Pint-expression. This effect may actually have been overestimated by a bias introduced by the measurement technique (i. e. , adding two extra Pint-copies with egfp or mcherry to two Pint from both ICEclc copies in the B13 chromosome, in the presence of two inrR copies). In contrast, and in the same ‘biased’ setting (two extra PinR-copies on a total of four), the total noise is significantly lower on the PinR-promoter and the relative contribution of the extrinsic noise is higher (Table 4), which is indicative for the more important contribution of RpoS on this promoter. Doubling the rpoS copy number resulted in a significant decrease of the total noise on Pint and a more important relative contribution of extrinsic noise (RpoS). This would make sense since individual cells would overall contain higher levels of RpoS permitting more direct interaction with Pint. Adding a third copy of inrR also reduced the level of intrinsic noise on Pint, but in this case because such cells would produce more InrR, diminishing the noise effect by ‘small numbers’ of regulatory factors (i. e. , InrR). Noise in individual cell RpoS levels is thus not propagated to noise in expression of downstream regulons, as was shown recently for global transcription factors in yeast [40], but rather is ‘captured’ in those cells having high RpoS levels and transduced by ICEclc factors to a precise activation cascade leading to ICEclc excision and transfer. Intriguingly, doubling rpoS copy number strongly increased the proportion of cells in the population expressing Pint and PinR from 3% to almost 20%, although the transfer frequency of ICEclc only doubled (Table S2). In contrast, adding a third copy of inrR to B13 did not statistically significantly increase the proportion of cells expressing Pint and PinR. To explain this, we propose the following model for ICEclc bistability generation (Figure 7). In this model cells that by chance have the highest RpoS levels are preconditioned to activate ICEclc, although another factor is needed for the actual activation mechanism. Available data suggest that activation starts at PinR, leading to synthesis of InrR, which, by an as yet unknown mechanism precisely relays activation (i. e. , within the same individual cell) to Pint and other ICEclc core genes. Microarray analysis confirmed the important role of InrR for the overall activation of ICEclc core functions, and indicated a possible feedback loop on its own expression (Figure S6). Importantly, RpoS but not InrR levels determine the proportion of cells that may become ICEclc activated. The feedback loop of InrR on PinR expression may be necessary to obtain sufficiently high InrR levels to act as co-regulator for the different ICEclc core gene operons [36]. Increasing inrR copy number, therefore, can decrease the noise in the expression of the ICEclc genes downstream of PinR, but does not influence the proportion of cells in culture activating ICEclc. The fact that a double rpoS gene copy increases the number of cells expressing Pint and PinR to 20% but only doubles transfer frequency suggests that there may be another component that is not under RpoS or InrR control that further limits conjugation rates. How may RpoS be accomplishing such a ‘thresholding’ control? One hypothesis is that RpoS has a relatively poor affinity for the PinR-promoter and that, therefore, on average only cells with by chance high RpoS levels can activate PinR. The inrR promoter bears a potential RpoS-motif in the −10 box (TGTCGATCCT), although it is not highly conserved [41]. As far as we are aware, this is the first time that RpoS has been implicated in controlling horizontal gene transfer of a conjugative element. RpoS homologs are part of a large protein cluster called the σ70 family, which is widely distributed among prokaryotes, although RpoS regulons can be quite different in individual species [42]. The only other report detailing a role for RpoS dealt with stationary phase regulation of Tn4652 activity in P. putida [43]. Interestingly, in that case RpoS downregulates tnpA transposition frequency since Tn4652 becomes at least 10 times more activated in an rpoS-defective strain [43]. Study on effects of stochastic fluctuations in sigma factors at the single cell level are extremely limited. Perez-Osorio documented highly heterogeneous rpoS mRNA levels in P. aeruginosa biofilms, but this occurred rather as a consequence of physico-chemical gradients within the biofilm [44]. Stochastic stress-induced fluctuations control the rbsV-rbsW-sigB operon for the stress response sigma factor SigB in Bacillus subtilis. Interestingly, sigBp expression proceeds in a ‘burst-like’ fashion with a higher pulse frequency under stress than under normal growth condition [45]. Bursts are initiated by stress-dependent fluctuations in phosphatase levels, then first amplified and subsequently terminated by sigB operon feedback on itself and on its anti- and anti-anti-sigma factors RbsW and RbsV, respectively. Gene expression noise is ubiquitous and plays an essential role in a variety of biological processes, triggering stochastic differentiation in clonal populations of cells [46]. Noise can provide a selective advantage by increasing phenotypic heterogeneity and increasing the chance of individuals to survive [46]. Evidence exists that more noisy systems can become selected under specific conditions [47]. In that sense, our data implicate that specific evolutionary elements such as ICEclc are wired within noise in a global transcription factor but can transduce this noise to a precise activation cascade, and thus may have been selected for their capacity to successfully exploit the noise. Escherichia coli DH5α (Gibco Life Technologies, Gaithersburg, Md.) was routinely used for plasmid propagation and cloning experiments. E. coli HB101 (pRK2013) was used as helper strain for conjugative delivery of mini-transposon constructs [48]. P. knackmussii strain B13 [49] is the original host of the clc element (ICEclc), of which it carries two copies [50]. All further B13 derivatives are listed in Table 1. Luria-Bertani (LB) medium [51] was used for cultivation of E. coli, whereas LB and type 21C mineral medium (MM) [52] were used for cultivation of P. knackmussii. 3-Chlorobenzoate (3CBA) was added to MM to a final concentration of 5 or 10 mM. When necessary, the following antibiotics were used at the indicated concentrations (µg per ml): ampicillin, 500 (for P. knackmussii) or 100 (for E. coli); kanamycin, 50 and tetracycline, 100 (for P. knackmussii strain B13 derivatives) or 12. 5 (for E. coli). P. knackmussii strain B13 was grown at 30°C; E. coli was grown at 37°C. Self-transfer was tested by mixing 500 µl suspension of around 109 donor cells (P. knackmussii B13 or one of its derivatives) and 500 µl suspension of around 109 recipient cells (P. putida UWC1) on membrane filters for 24,48,72 or 96 h, as described earlier [53]. Transconjugants (P. putida UWC1 with ICEclc) were selected on MM plates with 5 mM 3CBA as sole carbon and energy source (to select for ICEclc) and 50 µg per ml rifampicin (resistance marker of the recipient). Transfer frequencies were expressed as number of transconjugant colony forming units (CFU) per number of donor CFU. Polymerase chain reaction (PCR), reverse transcription RT-PCR, plasmid and chromosomal DNA isolations, RNA isolation, DNA fragment recovery, DNA ligations, transformations into E. coli and restriction enzyme digestions were all carried out according to standard procedures [51] or to specific recommendations by the suppliers of the molecular biology reagents (Qiagen GmbH; Promega; Stratagene). Sanger-type DNA sequencing was performed on an automated DNA sequencer using a 3. 1 Big-Dye kit (Applied Biosystems, ABI PRISM, 3100 DNA sequencer). Sequences were aligned and verified with the help of the Lasergene software package (Version 7, DNASTAR Inc. , Madison, Wisc.). Sequence databases were interrogated by using the BLAST program [54]. Primers were designed for conserved regions obtained in a nucleotide sequence alignment among rpoS genes of P. aeruginosa, P. putida and P. fluorescens (Table S4, Figure S1). A single 1-kb PCR product was obtained using these primers and B13 genomic DNA as template. This fragment was cloned and sequenced on both strands by primer walking. Surrounding regions of the rpoS gene of P. knackmussii were retrieved from draft genome sequence of P. knackmussii B13 (R. Miyazaki and J. R. van der Meer, unpublished). The B13 rpoS gene region was submitted to GenBank under accession number AB696604. An internal fragment of the rpoSB13 gene was amplified with a forward primer (080304) carrying a BamHI, and reverse primer (080303) carrying an EcoRI restriction site (Table S4). The amplified fragment was digested and cloned into the suicide plasmid vector pME3087, which carries a tetracycline resistance [55]. The plasmid was then mobilized from E. coli into P. knackmussii strain B13 via conjugation. Potential B13 transconjugants with a single recombination into rpoS were selected on MM with 5 mM 3CBA as carbon source plus 100 µg per ml tetracycline, further purified by replating and verified by PCR for accuracy of homologous recombination. In this manner a mutant of strain B13 was obtained in which rpoS was replaced by two incomplete and separated rpoS fragments (Figure S3). This mutant was named B13 rpoS (strain 2671). Separate experiments to delete rpoS by using recombination with a DNA fragment in which rpoS was fully deleted were not successful either (not shown). The same strategy was then used to produce a single recombinant disruption of rpoS in P. knackmussii strain B13 that lacked both inrR copies [33]. Reversion of the rpoS-pME3087 allele to wild-type rpoS in stationary phase cultures was tested by specific PCR (Table S4, Figure S4B). A 2. 2-kbp fragment containing the rpoS gene and its presumed promoter (PrpoS) was amplified from strain B13 purified genomic DNA using primers 091206 and 090902 (Table S4). The amplified material was first cloned into the vector pGEM-T-Easy (Promega). From here, the PrpoS-rpoS fragment was recovered by NotI digestion and inserted into the mini-Tn5 delivery plasmid pCK218, which was used to place the construction in single copy on the chromosome of strain B13-2673 (rpoS, mini-Tn[PinR-echerry-cat, Pint-egfp, Km], see below). As this strain carried a mini-Tn5 insertion already it was necessary to remove the Km gene cassette associated with it. Hereto the strain was transformed with plasmid pTS-parA [56], a temperature-sensitive replicon transiently expressing the ParA resolvase. B13 transformants were selected on LB plus ampicillin and subsequentially grown in the absence of kanamycin for twelve generations. Clones that had lost the Km cassette were screened by replica plating and the absence of the gene was verified by PCR. Finally, the temperature sensitive replicon was cured by growing the strain in LB at 37°C for 16 h and ensuring ampicillin sensitivity. The resulting strain was then used to introduce the mini-Tn5 containing the PrpoS-rpoS fragment, which was designated B13-2993 (rpoS, mini-Tn[PrpoS-rpoS, Km], mini-Tn[PinR-echerry-cat, Pint-egfp]). Three independent clones with possible different mini-transposon insertion sites were examined for ICEclc transfer and reporter gene expression. A 1700-bp fragment containing orf95213 and inrR genes plus PinR was amplified by PCR using primers (060605+080502, Table S4) carrying EcoRI and SpeI restriction sites, respectively. The PinR-orf95213-inrR fragment was digested with EcoRI and SpeI and cloned into the mini-Tn5 delivery plasmid pBAM1 [57]. In the same way, a 2. 2-kb fragment containing PrpoS-rpoS was amplified with primers (091206+090902) and cloned in pBAM1 using SphI and EcoRI. The resultant suicide plasmids were introduced into B13 or its derivatives by electroporation, from where the transposition was selected by plating on MM plus 3CBA and kanamycin. Bona fide single copy transposition was verified by PCR. At least three independent clones with possibly different insertion positions were used for further experiments. Transcriptional fusions between the Pint promoter in front of intB13 and the egfp gene, or Pint and a promoterless echerry gene have been described previously [33], [58]. Transcriptional fusion between the promoter of the orf95213, inrR, ssb gene cluster (PinR) and either egfp or echerry have been detailed elsewhere [33]. To examine expression of both PinR and Pint promoters simultaneously, we used a previous construct with PinR-echerry in one and Pint-egfp in the opposite direction [33]. Fusions were inserted in single copy into the chromosome of strain B13 or its mutant derivatives via mini-Tn5 delivery using pCK218 [59]. To measure activity of the rpoS promoter (PrpoS), a 1200-bp fragment upstream of rpoS including the nlpD gene was amplified from strain B13 by PCR (Figure 1). This fragment was purified and digested with NotI and EcoRI, and unidirectionally fused to a promoterless mcherry gene in the mini-Tn5 vector pBAM1 [57]. Transposon insertion mutants were selected on MM with 3CBA plus kanamycin or tetracycline and purified, upon which the correctness of the mini-Tn5 insertion was verified by PCR. For all mini-transposon insertions at least three independent clones were purified and examined for induction. To produce a C-terminal fusion of RpoS to mCherry, a ∼750 bp fragment containing the mcherry open reading frame was amplified using pMQ64-mcherry (kindly obtained from Dianne Newman, CalTech) as a template and primers (101003 and 101004), in which the start codon of mcherry was replaced by a short nucleotide sequence encoding 15 amino acids (KLPENSNVTRHRSAT) as a linker peptide. The fragment was then cloned in HindIII and SpeI sites on the mini-Tn5 delivery plasmid pBAM1, resulting in pBAM-link-mCherry. A 2. 1 kb region containing PrpoS and rpoS lacking its stop codon was amplified using B13 genomic DNA and primers 101001 plus 010102. This fragment was digested wtih EcoRI and HindIII, and cloned into the same sites on pBAM-link-mCherry (designated pBAM-rpoS-mcherry), After transformation in E. coli and purification, this plasmid was introduced into strain B13 or its derivatives by electroporation. Single copy transposon insertions of the rpoS-mcherry fusion construct were selected by plating cells on MM plus 3CBA and kanamycin. If required for introduction of subsequent mini-transpositions the kanamycin gene cassette was removed by ParA resolvase action (see above). At least three independent clones with possibly different insertion positions were used for further experiments. To replace rpoS of B13 by the gene for the RpoS-mCherry fusion protein we used double recombination by crossing-over. Hereto, a ∼1 kb downstream region of rpoS was first amplified using B13 genomic DNA and primers 110524 plus 110525, which was digested using XbaI and SalI and ligated wtih pJP5603-ISceIv2 [60]. Next, the gene for the RpoS-mCherry translational fusion protein on pBAM-rpoS-mcherry was recovered by digestion with EcoRI and SpeI, an inserted upstream of the amplified fragment in pJP5603-ISceIv2 which was hereto digested with EcoRI and XbaI. After transformation in E. coli and purification, the resulting plasmid was electroporated into strain B13-78 (Table 1). Single and double recombinants were selected according to a previously described strategy [9], obtaining an allelic exchange mutant that has the gene for RpoS-mCherry instead of the original rpoS. P. knackmussii strain B13 or B13 rpoS carrying the PrpoS-mcherry fusion were grown in 96-well black microtiter plates (Greiner Bio-one) with a flat transparent bottom. Each well contained 200 µl of MM medium with 5 mM 3CBA and was inoculated with 2 µl of a bacterial preculture grown overnight in LB medium. Microtiter plates were incubated at 30°C with orbital shaking at 500 rpm. At each given time point both culture turbidity (A600) and fluorescence emission (excitation at 590 nm and emission at 620 nm) were measured from triplicate cultures using a Fluostar fluorescence microplate reader (BMG Lab Technologies). Cultures of P. knackmussii strain B13-78 wild-type served for background fluorescence correction. To image eGFP, eCherry or mCherry expression in single cells, culture samples of 4 µl were placed on regular microscope slides, closed with a 50 mm long and 0. 15 mm thick cover slip, and imaged within 1–2 minutes. Fluorescence intensities of individual cells were recorded on image fields not previously exposed to UV-light to avoid bleaching. For most imaging series, except data shown in Figure 2, a Zeiss Axioskop2 upright epifluorescence microscope was used, equipped with Spot Xplorer 1. 4MPixel cooled CCD camera (Visitron Systems GmbH, Puchheim, Germany), and 100×/1. 30 oil immersion Plan-Neofluar lens at an exposure time of 500 ms. Filters used for eGFP and for eCherry/mCherry were eGFP HQ470/40 and Cy3 HQ545/30, respectively (Chroma Technology Corp, VT, USA). Images were digitally recorded using VisiView software (version 2. 0. 4, Visitron Systems GmbH). For data shown in Figure 2 and Figure S3 a Leica DMI6000B inverted epifluorescence microscope was used, equipped with a cooled black-and-white charge-coupled device camera (DFC320, Leica Microsystems CMS GmbH, Wetzlar, Germany), a 100/1. 30 oil immersion lens (HCX PL FLUOTAR; Leica), at an exposure time of 800 ms. Filters used for eGFP, and for eCherry or mCherry were GFP BP470/40 and Y3 BP535/50, respectively (Leica). Images were digitally recorded as 8-bit TIFF-files using the Leica AF6000 software. The mean pixel intensity for every individual object in an image was quantified by an automatic subroutine in the program MetaMorph (version 7. 7. 5; Visitron Systems GmbH) as described previously [33]. Fluorescence intensities per cell were expressed as cellular average gray values (AGVs) in which background intensities of each image were subtracted. Subpopulation expression was determined from cumulative ranking of all objects according to their AGV. The ‘breakpoint’ between subpopulations on cumulative distribution curves (Figure S8) was determined by manually placing slope lines to the linear parts of the curve. The point where both slope lines crossed was used to determine the corresponding percentile for the largest subpopulation with lowest AGVs. The relative size of the subpopulaton with highest AGVs (indicative for bistable promoter expression of Pint and PinR) was then calculated as 100% - the percentile of the breakpoint. The average expression intensity over the highest expressing subpopulation was calculated as the mean AGV over the percentile range between that of the breakpoint and 100%. Fluorescence images for display were adjusted for brightness to a level +143, cropped to their final size and saved at 300 dpi with Adobe Photoshop (Version CS4). Corresponding phase-contrast images were ‘auto contrasted’ using Photoshop. To identify and quantify noise in expression of the Pint and PinR promoters, two identical copies were fused to distinguishable reporter genes (i. e. egfp and echerry) and integrated into separate locations on the chromosome of B13 or its derivatives using mini-Tn5 delivery. Three independent clones with different insertional positions were maintained. Stationary phase cells of such double-reporter strains grown in MM with 3CBA were examined in epifluorescence microscopy, and their eGFP and eCherry fluorescence intensities were measured as outlined above (AGVs). AGVs of both markers in each cell were scaled to subtract background AGV of digital EFM images and normalized to the highest AGV in a population (100%). Only cells belonging to the subpopulations of having higher eGFP or eCherry fluorescence than the breakpoint in the respective cumulative curves (e. g. , Figure 6) were used for noise calculation. Intrinsic noise (ηint), extrinsic noise (ηext), and total noise (ηtot) were then calculated according to previous definitions given in Elowitz et al. [40] as follows: where g and c denote the normalized eGFP and eCherry AGV, respectively, observed in the nth single cell. Angled brackets denote a mean over the sample population. Significance of different treatments was examined by pair-wise t-test or ANOVA followed by a Tukey post hoc test. To test the effect of subpopulation size on noise calculations, data sets were randomly resampled using bootstrap procedures (1000 times), upon which the intrinsic, extrinsic and total noise were calculated and finally, averaged over all resampled populations of the same data set. Total RNA was isolated from P. knackmussii B13-78 (wild type), B13-2671 (rpoS) and B13-2201 (inrR−/−) cultures after 48 h in stationary phase after growth on 3CBA as sole carbon and energy source, by using the procedure described previously [37]. Briefly, cDNA was synthesized from total RNA, labeled with cyanine-3, purified and hybridized to a 8×15K custom-made Agilent microarray chip (Agilent Technologies, Santa Clara, CA). Data analysis was performed as described previously [37]. Microarray data and design have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GPL10091.

Horizontal gene transfer is one of the amazing phenomena in the prokaryotic world, by which DNA can be moved between species with means of a variety of specialized “elements” and/or specific host cell mechanisms. In particular the molecular decisions that have to be made in order to transfer DNA from one cell to another are fascinating, but very little is known about this at a cellular basis. Here we study a member of a widely distributed type of mobile DNA called “integrative and conjugative elements” or ICE. ICEclc normally resides in the chromosome of its bacterial host, but can excise from the chromosome and prepare for conjugation. Interestingly, the decision to excise ICEclc is made in only 3%–5% of cells in a clonal population in stationary phase. We focus specifically on the question of which mechanism may be responsible for setting this threshold level of ICEclc activation. We find that ICEclc activation is dependent on the individual cell level of the stationary phase sigma factor RpoS. The noise in RpoS expression across a population of cells thus sets the “threshold” for ICEclc to excise and prepare transfer.

Abstract Introduction Results Discussion Materials and Methods

microbial evolution biology microbiology

2012

Cellular Variability of RpoS Expression Underlies Subpopulation Activation of an Integrative and Conjugative Element

Exposure to ultraviolet (UV) radiation from sunlight accounts for 90% of the symptoms of premature skin aging and skin cancer. The tumor suppressor serine-threonine kinase LKB1 is mutated in Peutz-Jeghers syndrome and in a spectrum of epithelial cancers whose etiology suggests a cooperation with environmental insults. Here we analyzed the role of LKB1 in a UV-dependent mouse skin cancer model and show that LKB1 haploinsufficiency is enough to impede UVB-induced DNA damage repair, contributing to tumor development driven by aberrant growth factor signaling. We demonstrate that LKB1 and its downstream kinase NUAK1 bind to CDKN1A. In response to UVB irradiation, LKB1 together with NUAK1 phosphorylates CDKN1A regulating the DNA damage response. Upon UVB treatment, LKB1 or NUAK1 deficiency results in CDKN1A accumulation, impaired DNA repair and resistance to apoptosis. Importantly, analysis of human tumor samples suggests that LKB1 mutational status could be a prognostic risk factor for UV-induced skin cancer. Altogether, our results identify LKB1 as a DNA damage sensor protein regulating skin UV-induced DNA damage response. Ultraviolet (UV) radiation represents the number one leading cause for skin cancer. UV radiation can cause genetic mutations to DNA that if not repaired can lead to skin cancer. Elucidation of the mechanisms involved in UV-induced DNA damage response is important to understand the human disease, its treatment and prevention. LKB1/STK11 is a ubiquitously expressed and evolutionary conserved serine-threonine kinase. LKB1 was first identified as a tumor suppressor gene through its association with the Peutz-Jeghers syndrome [1] and is involved in a number of biological processes such as cell cycle control [2], [3], cellular energy metabolism [4], [5] and cell polarity [6]. The sub-cellular localization and activity of LKB1 is controlled through its interaction with the STE20-related adaptor (STRAD) and the armadillo repeat-containing mouse protein 25 (Mo25) [7], [8], regulating the activity of at least 14 downstream kinases-related to the AMPK family [9] and also, phosphorylating other substrates including STRAD and PTEN [10], [11]. LKB1 is phosphorylated on at least 8 residues, and evidence suggests that LKB1 auto-phosphorylates itself on at least four of these, whereas the other four are phosphorylated by upstream kinases [10], [12]. Among these residues Thr-366 is conserved in mammalian, Xenopus and Drosophila LKB1, and is located on a C-terminal non-catalytic moiety of the enzyme [13]. ATR and ATM phosphorylate LKB1Thr366 in response to ultraviolet irradiation (UV) and γ-radiation respectively, suggesting a role for LKB1 in response to DNA damage [14]. Although its function in DNA damage response has not been elucidated, mutation of Thr-366 to Ala or Asp partially inhibits the ability of LKB1 to suppress cell proliferation and it does not affect the nuclear cellular localization of LKB1. Moreover, phosphorylation of LKB1 at Thr-366 does not directly regulate LKB1 kinase activity [13], [14]. In addition to this, it has been suggested that LKB1-AMPK signaling controls non-homologous end joining (NHEJ) contributing to genome stability [15]. LKB1 appears to be mutated or inactivated in sporadic cancers whose spectrum of tumor types, suggest cooperation with exposure to environmental carcinogens. Thus, LKB1 has been found mutated in non-small cell lung carcinomas [16], [17], head and neck squamous cell carcinoma (SCC), pancreatic cancer [18] and melanomas [19]. It should be noted that hemizygous loss of chromosome 19p, spanning the LKB1 locus, is observed in many cancer types. This observation together with the data generated from mouse models suggests that LKB1 can behave as a haploinsufficient tumor suppressor [17], [20]. Indeed, Lkb1 deficiency sensitizes mice to DMBA-induced skin and lung SCC [21], and its inactivation in the context of RAS pathway activation facilitates the expansion of melanoma prometastatic tumor cell subpopulations [22] and progression of lung adenomas into carcinomas [23]. Cyclin-dependent kinase inhibitor 1A (CDKN1A) has an important role modulating DNA repair processes, inhibiting cell cycle progression and apoptosis. It competes for PCNA binding with several PCNA-reliant proteins that are directly involved in DNA repair processes including mismatch repair (MMR), base excision repair (BER) and translesion DNA synthesis (TLS) [24]–[29]. Evidence also suggest that CDKN1A may regulate nucleotide excision repair (NER), although its exact role has been controversial [30]. It has been showed that CDKN1A is proteolytically degraded in response to low-dose UV radiation by a mechanism that requires the physical interaction of CDKN1A with PCNA [31], [32]. Furthermore, the ability to degrade CDKN1A under this condition is critical for optimal DNA repair and to preserve genomic stability [24], [33]–[35]. The Hgf (hepatocyte growth factor) transgenic mouse (HgfTg) is a useful experimental model for determining the consequences and elucidating the mechanisms of exposure to UV radiation [36]–[38]. Here, we show that LKB1 haploinsufficiency sensitizes HgfTg mouse to UVB-induced skin cancer through a mechanism that involves CDKN1A protein accumulation. Interestingly, LKB1 and its downstream kinase NUAK1 bind and phosphorylate CDKN1A contributing to its physiological regulation. LKB1 deficiency leads to CDKN1A accumulation in response to UVB radiation, promoting both defects in DNA repair and protection from apoptosis. Our findings suggest that the mutational status of LKB1 can serve as a novel risk factor for UV-induced skin tumors. We previously demonstrated that LKB1 is involved in HGF signaling [4]. However, its in vivo role in response to UVB radiation has not been assessed. We examined the role of LKB1 in suppression of UVB-induced skin cancer using the HGF transgenic mouse model [38] by generating the HgfTg; Lkb1+/− mouse. Exposure of neonatal mice (3. 5 days old) to a single suberythemal dose of UVB radiation was sufficient to induce robust development of skin tumors only in HgfTg; Lkb1+/− mouse (Figure 1A). Early lesions appeared on the albino FVB background as persistent, discolored spots between 3 and 4 weeks of age, giving rise to frank and ulcerated tumors with a median onset age of 45 days (Figure 1A and B (i, ii, iii) ). Histologic analysis and staining for Involucrin, Cytokeratin-14 and β-Catenin revealed that these skin tumors were all malignant SCC (Figure 1C). Tumors also showed high amounts of p-c-MET as an indication of HGF activity, were positive for Cyclin D1 and showed a heterogeneous staining of LKB1 (Figure 1C). HgfTg; Lkb1+/− mice also showed an altered tumor spectrum relative to either Lkb1+/− or HgfTg mice (Figure S1A). Ten out of twelve UVB-irradiated HgfTg; Lkb1+/− mice developed SCCs. Tumors did not appear in non-irradiated animals or irradiated wild type or Lkb1+/− animals, and just one irradiated HgfTg mouse out of twelve developed an SCC (Figure 1D). As expected four HgfTg irradiated mice developed three nevi and one melanoma, albeit these mice were over 12 months old (Figure S1A). The tumor incidence in the UVB-irradiated HgfTg; Lkb1+/− mice was 83% showing variable multiplicity between animals (Figure S1B and C). Hence, Lkb1 heterozygosity in an HgfTg background sensitizes mice to single-dose UVB-induced skin SCC. Histopathological examination of mouse tumors revealed a remarkable similarity to lesions found in SCC patients. Human and mouse tumor SCC samples showed atypical proliferative keratinocytes forming irregular nests invading the stroma. These anastomosing growths of cords and nests were composed of cells that have nuclear atypia with irregular, large nuclei with one or more nucleoli and abundant eosinophilic cytoplasm. Mitotic figures are noted occasionally. (Figure 1E). Papillomas were rarely observed prior to SCC development in serially monitored UVB-induced HgfTg; Lkb1+/− mice, and we did not detect papillomatous changes adjacent to carcinoma in our histologic analyses. Finally, the incidence of papillomas (1 of 25 mice) was comparable in the wild type and single mutant cohorts (2 of 23 HgfTg mice and 1 of 22 Lkb1+/− mice developed papillomas) (Figure S1B). Consistent with this and the lack of papilloma-SCC progression, no H-Ras mutations were detected in the UVB-induced SCC arising in the HgfTg; Lkb1+/− mice. However, these tumors showed high levels of p-c-Met that activates RAS and PI3K pathways. Tumors also exhibited undifferentiated and malignant regions characterized by a decrease in the expression levels of LKB1, β-Catenin, E-Cadherin and α6-Integrin (Figure S1D). In agreement with the high tumor growth rate, the proliferation markers cyclin D1 and Ki67 (Figure 1C and S1E) indicated that these tumors were highly proliferative. They also showed low levels of apoptosis measured by counting cleaved caspase-3 positive cells (Figure S1E). In agreement with previous studies [20] and the heterogeneous LKB1 tumor staining, LKB1 was not expressed in SCC primary tumor-derived cell lines (Figure S1F), suggesting that the Lkb1 wild-type allele (Figure S1G) could be inactivated by multiple mechanisms in SCC, including deletion and possibly point mutation or promoter hypermethylation. We next investigated mice skin integrity. Immunohistochemical analysis of Cytokeratin-14, E-Cadherin and β-Catenin revealed comparable staining in the epidermis of wild type, HgfTg, Lkb1+/−, and HgfTg; Lkb1+/− mice, indicating that keratinocyte differentiation is not compromised neither with the half genetic dose of LKB1 nor overexpression of HGF (Figure S2A). As expected, skin of HgfTg and HgfTg; Lkb1+/− mice showed high levels of p-c-Met and based on p-Erk1/2 staining, an increased activation of the RAS pathway (Figure S2A). Ki67 staining indicated that in response to UVB irradiation (2 h and 48 h post irradiation) a large number of keratinocytes in the epidermal basal layer of Lkb1+/− and HgfTg; Lkb1+/− mice were recruited into cell cycle (Figure S2B). HgfTg; Lkb1+/− mice also demonstrated aberrantly dividing cells in the epidermal suprabasal layers and evidence for the lose of cell division polarity (Figure S2B). Since UVB-irradiation triggered skin tumorigenesis, we quantified the number of basal keratinocytes showing elevated levels of p-CHK2 after UVB irradiation, as an indicator of DNA damage. Two hours post-irradiation HgfTg and Lkb1+/− mice did not show significant differences in the number of p-CHK2 positive cells (Figure S3A). It is known that CDKN1A plays an important role in DNA repair [24], [30], [31]. In response to low doses of UV-irradiation CDKN1A is proteolytically degraded by a mechanism that requires the physical interaction of CDKN1A with PCNA [39], [40] allowing the recruitment of PCNA to the damaged DNA regions and optimal DNA repair [35]. Interestingly, Lkb1+/− and HgfTg; Lkb1+/− mice, showed an atypical response to UVB irradiation, presenting a significant accumulation of CDKN1A in basal keratinocytes in response to UVB-induced DNA damage (Figure S3B). Thus, although there were small differences in the total number of cells damaged among the different genotypes, there was a significant accumulation of CDKN1A in Lkb1+/− and HgfTg; Lkb1+/− mice (P<0,0001 WT vs. Lkb1+/−; or WT vs. HgfTg; Lkb1+/−) (Figure 2A), suggesting a DNA damage repair deficiency upon Lkb1 haploinsufficiency. In fact, a global genomic DNA repair analysis [41] of mouse skin confirmed that Lkb1+/− and HgfTg; Lkb1+/− mice had significant UVB-induced DNA damage repair deficiencies (Lkb1+/− mice repair 30% of cyclobutane pyrimidine dimers (CPD) and 31. 25% of 6-4 photoproducts (6-4pps) relative to WT mice; HgfTg; Lkb1+/− mice repair 65% of CPD and 68% of 6-4pps relative to WT mice) (Figure 2B and S3C). Hence, UVB irradiation in the context of Lkb1 haploinsufficiency leads to the accumulation of CDKN1A and impaired DNA repair. Next we sought to determine the molecular mechanism (s) that underlie the response to UVB-induced DNA damage. CDKN1A proteolytic degradation after low doses of UV is known to be critical for PCNA release and optimal DNA repair [24], [31], [32]. Indeed, pretreatment of normal human keratinocytes with the proteasome inhibitor MG132 induced the accumulation of CDKN1A in response to UVB irradiation (Figure 2C) evidencing the fine-tune regulation of CDKN1A amounts upon low doses of UVB irradiation. To investigate the role of LKB1 in response to UVB irradiation regulating CDKN1A protein levels, we knocked down (mRNA) LKB1 in wild type immortalized keratinocytes and in normal human epidermal keratinocytes (NHEK). In the absence of LKB1, UVB irradiation induced the accumulation of CDKN1A (Figure 2D and S3D) together with PCNA (Figure S3E). qRT-PCR analysis demonstrated that UVB-induced CDKN1A accumulation in the absence of LKB1 was not due to its transcriptional up-regulation. In agreement with the previously described role of LKB1 regulating CDKN1A expression [3], [42], [43], LKB1 knockdown cells showed a significant decrease in the UVB-induced transcriptional regulation of CDKN1A (Figure 2D). Moreover, the total amounts of LKB1 decreased overtime in response to UV irradiation (Figure 2D). Accumulation of CDKN1A in response to UVB was also observed in mouse keratinocytes generated from Lkb1+/− and HgfTg; Lkb1+/− animals compared to cells isolated from wild type and HgfTg mice (Figure 2E). We next investigated whether CDKN1A accumulation in LKB1 knockdown cells interfered with the repair of the UVB-damaged DNA. A global genomic DNA repair assay [41] showed that parental cells fully repair specific UVB-induced DNA damage 48 h after irradiation. However, two different clones of LKB1 knockdown cells repaired 35% and 20% of CPDs and 6-4pps, respectively, at the same time point (Figure 2F and S3F). Thus, these new evidence support the role of LKB1 in UVB-induced DNA damage repair, regulating the amount of CDKN1A protein. Altogether, these data suggested that the UVB-induced DNA damage response mediated by CDKN1A stability and/or transcriptional regulation was compromised in the context of Lkb1 haploinsufficiency. We next investigated whether LKB1 kinase activity was necessary for UVB-induced CDKN1A degradation. We reconstituted the system in HeLa cells (deficient for LKB1) and expressed the different LKB1 isoforms (wild type LKB1 or LKB1KD (kinase dead) ) in normal human epidermal keratinocytes (NHEK). Expression of CDKN1A together with either wild type LKB1 or LKB1KD (kinase dead) in HeLa cells showed that in response to UVB radiation there was an accumulation of CDKN1A in LKB1KD transfected cells, suggesting that LKB1 kinase activity was involved in the regulation of CDKN1A protein amounts in response to UVB irradiation. Similar response was observed in NHEK transfected cells (Figure 3A). Additionally, expression of mouse wild type Lkb1 but not the kinase dead mutant Lkb1KD in LKB1 knocked down HaCaT cells (HaCaT shLKB1), promoted the UVB-induced degradation of CDKN1A (Figure 3B). Interestingly, LKB1 and CDKN1A form part of the same immunocomplexes (Figure 3C). This association was also observed with the endogenous proteins at basal levels and in response to UVB radiation (Figure 3D and E) and it appeared to be specific to CDKN1A since CDKN1B (p27), a related CDK inhibitor family member, did not bind to LKB1 (Figure S4A). LKB1 and CDKN1A protein-protein interaction was confirmed by Bimolecular Fluorescence Complementation (BiFC) [44] (Figure 3F and Figure S4B). Construction of two different LKB1 mutants lacking C-terminal 20 (Flag-LBK1Δ416) and 113 (Flag-LBK1Δ323) amino acids, showed that carboxy-terminal region of LKB1 (Figure S4C) seemed to be involved in the binding to CDKN1A and, in a lesser extent to HSP90, a known LKB1 binding protein (Figure S4C). Altogether, these results suggest that LKB1 physically interacts with CDKN1A immunocomplexes and its kinase activity is involved in the CDKN1A UVB-induced degradation. Next, we investigated the functional consequences of this interaction and examined whether LKB1 was able to phosphorylate CDKN1A. In vitro kinase assays using recombinant His-LKB1/GST-STRADα/GST-Mo25 heterotrimeric complex and recombinant human GST-CDKN1A demonstrated that LKB1 phosphorylates CDKN1A (Figure 4A). Mass spectrometry analysis of the phosphorylated CDKN1A identified Thr80 as the residue phosphorylated by LKB1 in vitro (Figure 4A and S5A). In vivo labeling of cells with [32P]-orthophosphate followed by the immunoprecipitation of CDKN1A revealed that CDKN1A becomes phosphorylated in the presence of LKB1, STRADα and Mo25α Furthermore under these conditions CDKN1A bound to LKB1 immunocomplexes was also phosphorylated (Figure 4B). However, sequence alignment analysis of mouse, rat and human CDKN1A revealed that Thr80 is not conserved in mouse and rat proteins. Instead, mouse and rat proteins exhibit a Serine at position 78 not existing in the human orthologue (Figure S5B). Thus, we investigated whether LKB1 or any of its downstream AMPK family kinases were involved in the regulation of mouse CDKN1A. Results showed that LKB1 only phosphorylates human CDKN1A at Thr80 and not mouse CDKN1A, however, NUAK1, a downstream kinase of LKB1, phosphorylated human CDKN1A at Thr146 and mouse CDKN1A at Ser78 and Thr141, the equivalent residues in human CDKN1A (Thr80 and Thr146, respectively) (Figure 4C). Although, LKB1 in vitro phosphorylation of human CDKN1A (0. 32±0. 08 pmol [32P]/pmol protein) was less efficient than phosphorylation of a known substrate such as AMPKα (1,2±0. 11 pmol [32P]/pmol protein), both, mouse and human CDKN1A were efficiently phosphorylated in vitro by NUAK1 (0. 8±0. 18 pmol [32P]/pmol protein and 0. 9±1. 2 pmol [32P]/pmol protein, respectively) (Figure S5C). We identified by mass spectrometry phosphorylation of Ser78 in endogenous CDKN1A upon UVB irradiation in mouse melanoma cells (Figure S5D), Phosphorylation on Ser78 was significantly decreased in LKB1 depleted cells (30% vs. 1% of peptide phosphorylated respectively; p<0. 0001) (Figure S5E). In agreement with the role of LKB1 and NUAK1 regulating CDKN1A degradation upon UVB irradiation, non-phosphorylable human CDKN1A mutants T80A, S146A and double mutant T80A; S146A were accumulated after UVB treatment as compared to the wild type protein. Interestingly, mutation of both residues (T80A; S146A) caused a synergistic accumulation compared to the single mutations (Figure S6A). Besides the low amounts of NUAK1 within HaCat cells, we found NUAK1 and CDKN1A form part of the same immunocomplexes (Fig. S6B). Depletion of NUAK1 partially reproduced the accumulation of CDKN1A in response to UVB observed in the absence of LKB1 (Figure 4D and Figure S6C), and induced phosphorylation of CDKN1A Ser146 upon UVB radiation was absent in NUAK1 knockdown cells (Figure S6C). Moreover, expression of mutant HA-NUAK1T211A that cannot be activated by LKB1, led to the accumulation of CDKN1A, upon UVB treatment (Figure 4E) and expression of NUAK1 in LKB1 depleted cells almost totally reconstituted the normal response to UVB (Figure 4F). Altogether these results show evidence indicating that LKB1 and its downstream kinase NUAK1 phosphorylate CDKN1A and are involved in its regulation in response to UVB radiation. It has been suggested that LKB1 plays a role in genotoxic stress [45], [46]. However the molecular mechanism (s) involved are not fully understood. LKB1T366 becomes phosphorylated in response to UV irradiation [14] (Figure S7A and B). We observed that LKB1T366 was phosphorylated more efficiently in the skin of WT and HgfTg mouse than in the skin of Lkb1+/− and HgfTg; Lkb1+/− animals (Figure 5A). In a reconstituted system LKB1 wild type promoted the degradation of CDKN1A in response to UVB, however, LKB1T366A and LKB1KD mutants did not promote this effect (Figure 5B). Analysis of CDKN1A immunocomplexes from the same samples showed that LKB1T366A mutant has a diminished affinity for CDKN1A (Figure S7C). We identified PCNA as part of the immunocomplex, supporting the role of LKB1 in DNA damage response. In response to UVB the number of PCNA molecules bound to CDKN1A decreased in Flag-Lkb1WT transfected cells, while in Flag-Lkb1T366A transfected cells were unmodified (Figure S7C). The effect of Flag-Lkb1T366A mutant on CDKN1A stability in response to UVB, was also partially observed with endogenous protein (Figure 5C), and the number of phospho-LKB1T366 molecules recruited to the CDKN1A immunocomplexes increased in response to UVB (Figure S7D). Moreover, expression of mutant Flag-Lkb1T366A also impaired the cells ability to repair UVB-induced DNA damage supporting the role of LKB1 and CDKN1A degradation in DNA repair (Figure 5D). In agreement to this, depletion of CDKN1A in UVB-irradiated LKB1 knockdown cells allows them to repair DNA more efficiently (Figure 5E) Thus, these results suggest that UVB-induced phosphorylation of LKB1T366 regulates CDKN1A stability, which is linked to the response to UVB-induced DNA damage repair. Treatment of HgfTg; Lkb1+/− mice with a single neonatal dose of UVB radiation led to the development of SCC. Under LKB1 haploinsufficiency, UVB treatment promoted an accumulation of CDKN1A followed by a deficiency in DNA damage repair. In addition to the suggested role of CDKN1A in response to low doses of UV-induced DNA damage, accumulation of CDKN1A after different genotoxic insults protect cells from apoptosis [47]–[49]. Thus, we investigated the consequences of LKB1 loss in UVB-induced apoptosis. As expected LKB1 behaved as a tumor suppressor [3]. Knockdown of LKB1 in HaCat cells increased proliferation and cells (cell-cell) contact inhibition (Figure S7E). UVB irradiation induced the accumulation of CDKN1A in the absence of LKB1. We noticed that there was a significant (P<0. 001) higher number of viable cells after UVB irradiation in LKB1 depleted cells than in parental cells (Figure S7F). In agreement to this LKB1 knockdown cells were significantly (P<0. 001) more resistant to UVB-induced apoptosis than parental cells (2. 8% LKB1 knockdown cells vs. 10. 3% parental cells at 48 hours post-irradiation (Figure 6A). Resistance to apoptosis correlated with lower amounts of pro-apoptotic proteins BIM and PUMA (Figure 6B). Knockdown cells also showed lower amounts of CDKN1A at 72 h than parental cells (Figure 6B). This result was also observed in skin keratinocytes of Lkb1+/− and HgfTg; Lkb1+/− mice compared to wild type animals at 72 h–80 h post-irradiation. (P = 0. 004 and P<0. 0001 respectively) (Figure S8A) Moreover, Lkb1+/− and HgfTg; Lkb1+/− mice showed significant lower amounts of keratynocytes staining positive for Bim and cleaved-Caspase 3 at 48 hours post-irradiation, than wild type mice (P<0. 0001, Figure 6C). Thus, the data show evidence supporting the contribution of the loss of LKB1 and accumulation of CDKN1A to UVB-induced apoptosis resistance, leading to malignancy. To evaluate the relevance of LKB1 in human skin SCC we examined the expression of LKB1 by immunohistochemistry in 54 human skin SCC samples (Table S1). Samples were comprised of anatomical localizations compatible with UV-exposed and non-UV-exposed regions. Roughly 50% of the samples showed either very low or no staining for LKB1 (Figure 7A and B). We found that the lack of expression of LKB1 was independent of the differentiation stage of the tumor samples (n = 18 differentiated, n = 30 moderately differentiated and n = 8 poorly differentiated) (Figure 7B). Interestingly, there was a tendency where samples showing low or no staining for LKB1 localized preferentially in UV-exposed areas (52,3% of (n = 45) vs. 33,3% in non-UV-exposed areas (n = 9) ) (Figure 7C). Moreover, samples from UV exposed areas and low LKB1 expression amounts felt into any tumor stage category, while all samples from non-UV exposed areas and low expression of LKB1 were poorly differentiated (Figure 7D). Interestingly, analysis of a curated data set of 225 tumors from another relevant UV-induced skin tumor such as cutaneous melanoma (c-Bioportal, MSKCC) [50] showed alterations in LKB1 or NUAK1 in 22. 2% of cases that were mutually exclusive (odds ratio 0. 625 (no association); 95% Confidence Interval: 0. 138438–2. 821652; P-value: 0. 412752 (Fisher' s Exact Test) ). In fact staining of human skin tumor SCC samples with LKB1 and NUAK1 showed an inverse Hscore correlation (95% confidence interval, P = 0. 0033) between LKB1 and NUAK1 expression (Figure S8B). This mutual exclusivity of LKB1 or NUAK1 alterations is observed other tumor types including head and neck squamous cells carcinomas (19. 7% of data set from 295 tumors), (95% Confidence Interval: 0. 552751–5. 723118 P-value: 0. 250381 (Fisher' s Exact Test), cervical squamous cell carcinoma (30. 6% of data set from 36 tumors), (95% Confidence Interval: 0. 095179–10. 506562 P-value: 0. 695155 (Fisher' s Exact Test) and lung squamous cell carcinoma (15. 3% of data set from 177 tumors), (95% Confidence Interval: 0. 109684–7. 534791 P-value: 0. 703561 (Fisher' s Exact Test). Hence, this additional data suggest that the loss of LKB1 expression at early stages could contribute to UV-induced skin cancer development (Figure 7E). Genotoxic environmental insults such as UV are associated with the development of skin cancer. DNA damage repair has been proven to be crucial in fending off detrimental effects such as mutagenesis and cell death. Here, we show that LKB1 tumor suppressor is a DNA damage sensor, and together with its downstream kinase NUAK1, contributes to UVB-induced CDKN1A degradation, allowing DNA repair and genomic integrity. LKB1/STK11 is mutated in sporadic human cancers whose spectrum of tumor types suggests cooperation with exposure to environmental carcinogens [17]–[19]. In humans, skin-SCC is associated with chronic rather than intermittent intense exposure to UV radiation [51]. However, under conditions of LKB1 haploinsufficiency in an HgfTg background, a single neonatal suberythemal dose of UVB was sufficient to induce skin-SCC bypassing the papilloma-SCC sequence. This result highlights the in vivo role of LKB1 in response to genotoxic insults, in particular to UVB irradiation. In contrast to human SCC samples, in mice we did not detect any mutations in HRAS. This could be the reason why our mice do not develop papillomas. However, it is likely that the requirement for RAS pathway activation for tumor development and progression in humans is achieved in the mouse through the activation of c-MET by HGF over-expression. Most of the SCC tumors showed a heterogeneous expression of LKB1. In this matter, lack of expression of LKB1 was observed often associated to undifferentiated tumor regions and in mouse tumor-derived cell lines. Hence, contrary to the previously published DMBA-induced SCC mouse model [21] and in agreement with the benign gastrointestinal polyposis associated with Lkb1 deficiency; in our model malignant SCC pathogenesis seems not to require biallelic inactivation of Lkb1. It is known that ATM/ATR are important kinases involved in DNA damage response [52]. Previous work suggested that a yet unidentified kinase would be likely to be involved acting between ATR and CDKN1A in response to low doses of UV irradiation [24]. CDKN1A is induced after ionizing radiation, but degraded after UV exposure [24]. Thus, UV-induced degradation of CDKN1A is necessary for optimal DNA repair and to preserve genomic stability. This is accomplished by CDKN1A ubiquitylation and degradation via the CRL4 (Cdt2) ubiquitin ligase complex, setting free proliferating cell nuclear antigen (PCNA) from the CDKN1A-PCNA complexes and controlling translesion DNA synthesis [24], [35], [39], [53]. In agreement with previous publications our results show that lack of LKB1 compromised the transcriptional regulation of CDKN1A (Figure 2D) [43], however, it also promoted accumulation of CDKN1A protein in response to UVB irradiation. We show that LKB1 deficiency impedes physiological UVB-induced CDKN1A degradation, impairing DNA damage repair and consequently contributes to mutagenesis and tumor development. It is known that, multitask Ser/Thr kinase LKB1 becomes phosphorylated by ATR at Thr366 in response UV [14]. However, the physiological role for this modification is unknown. We show that mutation of LKB1 Thr-366 to Ala impaired the cells ability to repair UVB-induced DNA damage by affecting CDKN1A UVB–induced degradation. Furthermore, in humans, LKB1 and its downstream kinase NUAK1 bind and phosphorylate CDKN1A (at Thr80 and Ser146, respectively) contributing to its degradation in response to UVB and DNA repair. Although LKB1 it is known to phosphorylate AMPK family members, the amount of pmol of phosphate incorporated per pmol of CDKN1A compared head to head to the in vitro efficiency toward AMPK, suggested its capability to phosphorylate other substrates different to the AMPK family members. In this matter, NUAK1 was very effective phosphorylating both human and mouse CDKN1A. Of note is that, Thr80 is not conserved in mouse CDKN1A sequence, instead, there is a Ser at position 78. Interestingly, NUAK1 phosphorylates mouse CDKN1A at Ser78 and Ser141, the homologous residues in the human orthologue, and also conserved in the rat protein. Although, the data suggest that NUAK1 contribution is mediated by LKB1, these results do not exclude its LKB1-independent effect. In fact NUAK1 has been previously involved in DNA damage response phosphorylating p53 and participating in the transcriptional regulation of CDKN1A promoter [45]. We hypothesize that this redundancy in humans (LKB1 and NUAK1) compared to mouse (NUAK1) provides biological robustness to a mechanism involved in a UV genotoxic response. This could be particularly relevant to humans which skin is clearly more exposed to environmental insults such as UV radiation. Several lines of evidence support the biological role of LKB1 in DNA damage response. First, LKB1 becomes phosphorylated at Thr366 in response to UVB. Second, LKB1 kinase activity seems to be necessary for CDKN1A degradation in response to UVB radiation. Third, LKB1 binds and phosphorylates CDKN1A. Fourth, NUAK1, the LKB1 downstream kinase, rescues the LKB1 knockdown phenotype in response to UVB. Fifth, LKB1 binds to CDKN1A and upon UVB treatment there is an increased association of CDKN1A molecules to phopho-LKB1T366. Furthermore, LKB1T366A mutant has a diminished binding to CDKN1A compared with LKB1WT and LKB1KD mutant, it does not promote CDKN1A degradation in response to UVB radiation, and impairs DNA damage repair. In addition to all these, there is concomitant degradation of LKB1 and CDKN1A in response to UV. Although, we cannot fully explain this observation, it is tempting to speculate that these two molecules are simultaneously proteasome-degraded, permitting the liberation of PCNA and DNA repair. The later, is also supported by the increment in UVB-induced DNA damage repair in LKB1 depleted cells when CDKN1A is knocked down. Although the link between CDKN1A degradation and DNA repair has been extensively demonstrated and our data, and other recent work [35] confirm this connection, how UVB-induced CDKN1A phosphorylation leads to its degradation and whether the concomitant LKB1 degradation is connected needs to be further investigated. From the pathogenic point of view in addition to the UVB induced mutations, the loss of LKB1 tumor suppressor would also contribute to deregulate cell proliferation and cell-to-cell contact inhibition. Furthermore, LKB1 deficient cells were resistant to UVB-induced apoptosis, probably through the accumulation of CDKN1A [47]–[49]. Altogether this would ultimately favor the fixation of UVB-induced mutations and tumor development. All these data suggest that in humans silencing a single copy of LKB1 would be sufficient to increase the risk of the acquisition and accumulation of UV-induced mutations, placing LKB1 as an important player in response to environmental insults associated to the acquisition of skin cancer. Indeed, analysis of human samples showed that 50% of skin-SCC lack or showed very low amounts of LKB1 expression. The absence of expression of LKB1 was independent of the differentiation stage of the tumor and had a tendency to be more frequent in SCC from UV–exposed areas. This suggests that the loss of LKB1 expression is an early event in tumor development and/or progression. Since our animal model demonstrates that LKB1 haploinsufficiency is sufficient to cause the accumulation of UVB-induced DNA damage, we posit that the mutational status of LKB1 is a prognostic risk factor for UV-induced skin cancers. In agreement to this, in melanoma and squamous cell carcinomas, LKB1 is mutated in 2% and 11% of tumor samples, respectively (COSMIC-Wellcome Trust Sanger Institute). Furthermore, our data and results from other studies (c-Bioportal, MSKCC) show that tumors with a clear environmental component including, melanoma, head an neck squamous cell carcinoma, lung squamous cell carcinoma and endometrial squamous cell carcinoma, alterations in LKB1 or NUAK1 are mutually exclusive, reinforcing the role of this molecular axis in DNA damage and genomic instability. In summary, here we unveil a novel role for LKB1 as a UV-induced DNA damage sensor protein. Reduced amounts of LKB1 are enough to impair UVB-induced DNA repair and cooperate with HGF signaling to promote skin cancer. At the molecular level the results indicate that we have identified the missing link between ATR and the physiological regulation of CDKN1A in response to UVB. In this matter, following UVB irradiation LKB1 becomes phosphorylated by ATM/ATR and then, LKB1 and its downstream kinase NUAK1 phosphorylate CDKN1A contributing to its physiological regulation. Thus, deficiencies in LKB1 promotes fixation of UVB–induced mutations, resistance to UVB-induced apoptosis contributing to tumor development. HgfTg and Lkb1+/− strains and UV treatment have been previously described [20], [38]. Data from our survival analysis was performed using Prism 6 (GraphPad Software Inc.). All animal work have been conducted according to relevant national and international guidelines and approved by the Animal Ethics Committee from the Institution (Institut de Recerca Vall d' Hebron (Barcelona, Spain). 293T, HeLa and HaCat cells were obtained from ATCC. NHEK (Normal juvenil Human Epidermal Keratinocytes) were obtained from Promo-Cell (Heilderberg, Germany) and cultured in Keratinocyte growth medium 2 (Promo-Cell). Mouse keratinocytes were isolated as described in [54] MG132 was from Sigma-Aldrich (Saint Louis, MO, USA) Cf = 200 nM. γ–32P-P-ATP and γ–32P-Orthophosphate were purchased from PerkinElmer (Waltham, Massachusetts, USA). Plasmids pCMV5-human CDKN1A, pCMV5-human CDKN1A T80A and pCMV5-human CDKN1A T80D were generated using QuickChange Site-Directed Mutagenesis (Stratagene, Cedar Creek, TX, USA). pCMV5-Flag-mouse-Lkb1WT and pCMV5-Flag-mouse-Lkb1KD (kinase dead) were a generous gift from D. Alessi, Univ. Dundee, UK; pCMV5-Flag-mouse-Lkb1T366A was generated using Quick-Change Site-Directed Mutagenesis (Stratagene, Cedar Creek, TX, USA). pcDNA4-Flag-STRADα and pKCFP-MO25α were a gift from M. Sanchez-Céspedes (PEBC-IDIBELL, Barcelona, Spain). pEYFP-p27wt was a gift from G. Mills (MD Anderson Cancer Center, Houston, USA). For LKB1 silencing five different lentiviral pLKO. 1-shLKB1 constructs were obtained from Sigma-Aldrich (Saint Louis, MO, USA). For NUAK1 and CDKN1A siRNA were purchased from Invitrogen. All transfections and lentiviral infections were performed as described [4]. All pCMV5-Flag-mouse-Lkb1 isoforms were co-transfected with equimolar amounts of pcDNA4-Flag-STRADα and pKCFP-MO25α. Total amount of transfected DNA was compensated using an empty vector (E. V.). Constructs were transfected into cells with Lipofectamine 2000 Transfection Reagent (Invitrogen), following the manufacturer' s recommended protocol. Immunoprecipitation was performed in RIPA buffer using M2-agarose (Sigma-Aldrich) 24 h post-transfection and after UVB treatment. Keratin-14, Involucrin, E-cadherin, β-catenin and α6-Integrin were obtained from HG. Palmer, VHIO, Spain, while LKB1 (D60C5), phosphor-ERK1/2 and total ERK1/2, phospho-Met (Tyr1234/1235), phospho-ATR (Ser428), cleaved caspase-3, PUMA, Bim and phospho-CHK2 antibodies were from Cell Signaling (Danvers, MA USA). Ki67 was from Master Diagnostica (Granada, Spain). G3PDH (GAPDH) was from Trevigen Inc. (Gaithersburg, MD USA). Cyclin D1, p27 (C-19), anti-HA (Y-11), and LKB1 were from Santa Cruz (Santa Cruz, CA, USA). Anti –NUAK1 was from Proteintech (Proteintech Group, Inc. Chicago, IL, USA). Anti–BIM for IHC-P was from Thermo scientific (Thermo Fisher Scientific Inc. , Waltham, MA USA). Anti-CDKN1A Ab-11 (Clone CP74) was from Thermo Fisher Scientific (Runcorn, Cheshire, UK). Anti–PCNA was from Abcam, (Cambridge, UK). Phospho-LKB1 (T366) was purchased from MRC, University of Dundee, Glasglow, UK. β-Actin was from Millipore, Madrid, Spain. Secondary antibodies included Alexa Fluor 488, Alexa Fluor 563 (Invitrogen, Carlsbad, CA USA), anti-rabbit and anti-mouse linked to horseradish peroxidase (GE Healthcare, Barcelona, Spain), MOM kit, and ABC Vector kit (Vector-Labs, Burlingame, CA, USA). Cells were irradiated with UVB (30 J/m2) at 50%–70% confluency without medium nor the lid. After treatments, cells were lysed in RIPA buffer and immuno-blots were performed as previously described [4]. Has been performed as previously described in [4], [55]. Paraffin-embedded tumor samples were subjected to immunocytochemistry according to the manufacturer' s antibody protocol. The samples used in this Project were provided by the Tumor Bank of the Vall d' Hebron University Hospital Biobank with appropriate ethical approval (supported by the Xarxa de Bancs de Tumors de Catalunya sponsored by Pla Director d' Oncología de Catalunya (XBTC); supported by the RETICS de Biobancos (ISCIII). All cases were evaluated independently by an expert dermatopathologist (BF) and one trained Molecular Biologist (JHL) blinded for patient groups, taking into account the percentage of positive cells and intensity of the staining, which was assessed semiquantitatively. Final results were obtained utilizing the average of the two values. Whenever a major discrepancy was observed between both observers, the cases were discussed using a multi-headed microscope. LKB1 was evaluated using Histoscore (Hscore) there was calculated: Hscore = (1× % weak staining cells) + (2× % moderate-strong staining cells) with results ranging from 0 to 200. Samples with an Hscore<25 were classified as low expression samples. pCMV-CDKN1A-mRFP1-YFP-C and pCMV-LKB1-CFP-YFP-N constructs were generated, introducing wild type human LKB1 (EMBL-EBI: AF035625) and human CDKN1A (EMBL-EBI: L25610) sequences into pCMV-R-YC and pCMV-C-YN vectors (obtained from Brack-Werner, Institute of Molecular Virology, GSF-National Research Center for Environment and Health, Neuherberg, Germany) [44]. HeLa cells were transiently transfected with these constructs for 24 h and YFP, CFP and mRFP1 fluorescence was analyzed by confocal microscopy (Espectral FV1000 Olympus). Cells were harvested 0,24,48 or 72 h after UV irradiation. Unirradiated control cells were also harvested. Genomic DNA was isolated using the DNeasy kit (Qiagen Mississauga, Ontario) according to the manufacturer' s protocol. DNA (100 ng in 0. 5 M NaOH and 10 mM EDTA) was denatured by boiling for 10 min. Ice-cold ammonium acetate (2 M) was added to a final concentration of 1 M. Denatured DNA was spotted onto a nitrocellulose membrane pre-wetted with 6× SSC using a slot-blot apparatus (Bio-Dot SF, Bio-Rad, Mississauga, Ontario). The filter was baked at 80°C for 2 h. Thymine dimers were quantified using the monoclonal antibody MC-062 (clone KTM53, Kamiya Biomedical, Seattle, WA). Bound antibody was detected by ECL plus (Amersham, Baie d' Urfè, Quèbec), and quantified by autoradiography. The membrane was re-probed with radiolabeled mouse genomic DNA to quantify the amount of the sample DNA per slot. The antibody signal was normalized to the amount of DNA per lane, and the rate of lesion removal was calculated [41] and graphed. In vitro LKB1 (Millipore) kinase protein assay were performed as described in [8] using recombinant human Hs-GST-CDKN1A (Abcam) or mouse Mm-GST-CDKN1A. In vivo γ–32P metabolic labeling was described in [56]. Quantification of kinase activities was done as in [10]. Samples were separated on a 10% SDS-PAGE gel, and the gel stained with colloidal Coomassie blue. Protein bands of interest were processed as described in [57].

Environmental insults are directly involved in cancer development. In particular, Ultraviolet (UV) radiation has been associated to the acquisition of different types skin cancer and premature skin aging. UV radiation causes modifications in the genetic material of cells (DNA) that if not repaired properly will lead to a mutated DNA (mutated genes) which might trigger the development of cancer. Understanding the molecular basis of the UV-induced DNA damage response is important to elucidate the mechanisms of skin homeostasis and tumorigenesis. Here we provide a UVB-induced skin cancer animal model showing that LKB1 tumor suppressor is also a DNA damage sensor. Importantly, the data suggest that reduced amounts of LKB1 protein in skin could be a risk factor for UV-induced skin carcinogenesis in humans.

Abstract Introduction Results Discussion Materials and Methods

dna damage biochemistry dermatology skin neoplasms medicine and health sciences skin tumors biology and life sciences dna repair dna biochemical activity

2014

A Mouse Model Uncovers LKB1 as an UVB-Induced DNA Damage Sensor Mediating CDKN1A (p21WAF1/CIP1) Degradation

Focal adhesions are protein complexes that anchor cells to the extracellular matrix. During migration, the growth and disassembly of these structures are spatiotemporally regulated, with new adhesions forming at the leading edge of the cell and mature adhesions disassembling at the rear. Signalling proteins and structural cytoskeletal components tightly regulate adhesion dynamics. Paxillin, an adaptor protein within adhesions, is one of these proteins. Its phosphorylation at serine 273 (S273) is crucial for maintaining fast adhesion assembly and disassembly. Paxillin is known to bind to a GIT1-βPIX-PAK1 complex, which increases the local activation of the small GTPase Rac. To understand quantitatively the behaviour of this system and how it relates to adhesion assembly/disassembly, we developed a mathematical model describing the dynamics of the small GTPases Rac and Rho as determined by paxillin S273 phosphorylation. Our model revealed that the system possesses bistability, where switching between uninduced (active Rho) and induced (active Rac) states can occur through a change in rate of paxillin phosphorylation or PAK1 activation. The bistable switch is characterized by the presence of memory, minimal change in the levels of active Rac and Rho within the induced and uninduced states, respectively, and the limited regime of monostability associated with the uninduced state. These results were validated experimentally by showing the presence of bimodality in adhesion assembly and disassembly rates, and demonstrating that Rac activity increases after treating Chinese Hamster Ovary cells with okadaic acid (a paxillin phosphatase inhibitor), followed by a modest recovery after 20 min washout. Spatial gradients of phosphorylated paxillin in a reaction-diffusion model gave rise to distinct regions of Rac and Rho activities, resembling polarization of a cell into front and rear. Perturbing several parameters of the model also revealed important insights into how signalling components upstream and downstream of paxillin phosphorylation affect dynamics. In multicellular organisms, cell migration is key to proper development and maintenance of physiological processes such as embryogenesis, axonal outgrowth in neurons, and wound healing [1–5]. Additionally, aberrant migration can lead to pathological effects such as cancer metastasis [1,3–7]. To identify key factors that lead to these physiological and pathological functions, a better understanding of the biochemical regulatory pathways governing the dynamics of motility is required. Regulation of cell migration occurs through several different mechanisms, and involves changes in protein activities that occur both globally (i. e. across the entire cell) and locally [8–11]. Polarization, for example, has historically been attributed to a cell-wide gradient in the activities of the Rho family of GTPases, including Cdc42, Rac1 (Rac), and RhoA (Rho), and their cycling between the cytoplasm and membrane binding [8,9, 12–15]. Specifically, the activities of Cdc42 and Rac, known to promote actin polymerization, membrane protrusion and membrane ruffling [16–20], are thought to be high at the cell front compared to the rear, whereas the activity of RhoA, responsible for actomyosin contraction, is low at the cell front and high at the rear [8,12–14]. On a smaller scale, mechanosensitive proteins (such as talin) reside within adhesions and facilitate local regulation [21,22]. These proteins are bound to both the adhesion and the actin cytoskeleton, and can be stretched in response to actomyosin contractile force to reveal binding sites that are normally concealed under low tension [21]. Proteins such as vinculin can subsequently bind to these exposed sites and alter processes such as adhesion assembly [22]. Local regulation is not exclusively initiated by mechanosensitive proteins, however. Signalling cascades, often initiated by protein phosphorylation, may take place at adhesions without a direct dependence on tension [10,23–27]. For example, in Chinese Hamster Ovary K1 (CHO-K1) cells, p21-Activated Kinase 1 (referred to hereafter as PAK) -mediated phosphorylation of the scaffold protein paxillin at serine residue 273 (S273) increases adhesion dynamics [25] and raises the level of active Rac through the activity of a trimeric protein complex consisting of G protein-coupled receptor kinase InteracTor 1 (referred to hereafter as GIT), beta-PAK-Interacting eXchange factor (referred to hereafter as PIX), and PAK that binds to paxillin subsequent to its phosphorylation [25]. Previous studies [25] have shown that high levels of S273 phosphorylation are associated with an increase in the number of small adhesions located near the cell periphery. These adhesions assemble and disassemble relatively quickly, allowing for fast membrane protrusion and cell migration. Conversely, cells with low levels of paxillin phosphorylation exhibit relatively large adhesions with slow assembly and disassembly rates, causing cell protrusion and migration velocities to be slow as well. These results were obtained by using either phosphomimetic or nonphosphorylatable mutants of paxillin in which the serine 273 residue was replaced with aspartic acid (S273D) or alanine (S273A), respectively. In addition to phosphorylation, downregulation of paxillin dephosphorylation, through the serine/threonine phosphatase PP2A, has also been shown to increase cell spreading, motility, and metastasis [28]. Therefore, it would be interesting to investigate whether restoring levels of PP2A activity to a normal range could attenuate cancer metastases. The combined effects of the multiple regulatory pathways involved in cell motility are thus complex and difficult to consolidate. This makes identifying key relationships challenging and predicting new relationships unintuitive. Through the use of mathematical modeling techniques, however, these processes may be integrated to explain biological phenomena, make predictions, and motivate new experimental studies. Many quantitative models capture migration, or aspects of migration, using a balance of physical forces within the cell, including the polymerization force of actin filaments, the inward force of actomyosin contraction, and the traction force generated by cell-matrix adhesions [29–33]. Models of adhesion assembly and disassembly have been also used to predict distributions of adhesion sizes [34]. Another route of modeling has been through studying the molecular interactions of signalling proteins [13,14,33,35–39], including Cdc42, Rac, and Rho. Their interactions cause cell polarization and directed cell motion [13,14,38,39]. Analysis of these interactions revealed that polarization is achieved through bistability, a commonly observed feature in biological systems in which one of two states can be attained, depending on initial conditions, but switching between them can also occur (though a hysteresis) upon perturbations [40]. In this system, bistability results from mutual inhibition between Rac and Rho [13,14,41,42]. In similar studies, the activities of Cdc42, Rac, and Rho have also been combined with the effects of extracellular matrix-dependent protrusion velocity, contractile force, and feedback of adhesions onto Rac and Rho activities [33,35,36] to predict bistability in both the protrusion velocity and the density of stable adhesions (as determined by matrix stiffness). Variations of such models revealed that bistability persists in models of Rac and Rho that assume Rac is inactivated by Rho [37]. Using IPA-3, an inhibitor of PAK activity, the presence of bistability was confirmed experimentally [37]. Here, we employed similar modeling techniques to predict the role of paxillin S273 phosphorylation and the GIT-PIX-PAK complex in determining the dynamics and steady state behaviour of active Rac and Rho. Western blot quantification was performed using Fiji [64]. Numerical simulations and bifurcation analysis were done using MATLAB (MathWorks, Natick, MA) and xppauto (a freeware available at http: //www. math. pitt. edu/~bard/xpp/xpp. html). Data were digitized with WebPlotDigitizer (Ankit Rohatgi, Austin, TX, available at https: //automeris. io/WebPlotDigitizer). Steady state expressions for intermediate complexes (see S1 Text) were derived using Wolfram Mathematica (Wolfram, Champaign, IL). It was previously shown with CHO-K1 cells that the D and A mutants of paxillin S273 are associated with more dynamic or more stable adhesions and more motility or less motility, respectively, whereas cells expressing wild type paxillin exhibited features that are intermediate to these two mutants [25]. To first determine whether the initial level of phosphopaxillin is a factor in reproducing the phenotype associated with wild type cells at steady state, we plotted the projected intersections of the ρ- and R*-nullsurfaces (black lines) along with the projected P*-nullsurface (gray line), onto the P*-R*-plane (Fig 2A). The three intersections between the two resulting curves mark the projected steady states of the system onto the same plane. When the binding between Paxp and the GIT-PIX-PAK complex (defined by the rate constant kC) was sufficiently large (= 5 s-1), the initial value of P* played a crucial role in determining the long term behaviour of ρ and R*. Fig 2A shows that there was a range of initial values of R* where a low initial P* value (to the left of the dashed black line) would cause R* to converge to its lower steady state (referred to hereafter as the uninduced state), and a higher initial P* value (roughly to the right of the dashed black line) that would cause R* to converge toward the elevated steady state (referred to hereafter as the induced state). In other words, the middle steady state is a saddle whose stable manifold acts as a separatrix (boundary) demarcating the basins of attraction of the induced and uninduced states. These results suggest that, depending on the initial level of phosphopaxillin (P0*), the system may converge to one of two steady state levels of active Rac: reduced or elevated, in a phenomenon referred to as bistability. It also shows that the slope of the stable manifold of the saddle (which happens to be close to the dashed black line) determines how sensitive the system is to perturbations in P0*. To experimentally verify whether bistability in the expression level of active Rac and Rho is present within the cell, we examined the assembly and disassembly rates of adhesions, which are regulated in part by the activities of Rac and Rho, in wild type CHO-K1 cells or in cell lines overexpressing one of the two mutant paxillin proteins (see Fig 2B). In cells expressing wild type paxillin (left panels), there were two subpopulations of adhesions that appeared to exist, one with smaller more dynamic adhesions with high rates of assembly (top) /disassembly (bottom), and another with larger more stable adhesions with slower rates of assembly (top) /disassembly (bottom). In contrast, both the A (middle panels) and D (right panels) mutants exhibited only one population of adhesions. In the A mutant, all adhesions assembled (top) /disassembled (bottom) slowly, whereas in the D mutant, all adhesions assembled (top) /disassembled (bottom) quickly regardless of adhesion size. For the list of all assembly and disassembly rates, see Table 2. This suggests that in either case, one population of adhesions is conserved while the other is absent. The existence of the two subpopulations of adhesions in wild type cells, with significantly different assembly/disassembly rates (Fig 2C), is indicative of the presence of two distinct levels of Rac and Rho activities, where a state with high Rac and low Rho activity would be capable of inducing fast adhesion assembly/disassembly rates, whereas a state with low Rac and high Rho activity would cause slow adhesion assembly/disassembly rates to occur. Together with our model, these experimental results suggest that, in wild type paxillin-expressing cells, an intermediate range of initial levels of phosphopaxillin within the cells, can lead to both (i) high Rac/low Rho activity causing fast adhesion assembly/disassembly, and to (ii) low Rac/high Rho activity leading to slow adhesion assembly/disassembly. In other words, there is a mixture of adhesion assembly/disassembly rates in these wild type cells. Interestingly, the disappearance of one of the two subpopulations of adhesions in the A and D mutants also indicates that paxillin phosphorylation rate plays a crucial role in regulating bistability. In summary, we have shown that Rac/Rho activity depends on both the initial level of phosphopaxillin in cells expressing wild type paxillin and on paxillin phosphorylation rate in the A and D mutants. In this section, we studied with our model how varying the rates of paxillin S273 phosphorylation (B) and dephosphorylation (δP) rates (see Eq (9) in Methods Section) affect Rac and Rho activities. Downstream of paxillin phosphorylation, the binding between Paxp and an intact GIT-PIX-PAK complex were shown to be required for the effects of paxillin S273 phosphorylation to be detected [25]. Disruption of the Paxp-GIT, GIT-PIX, or PIX-PAK interactions in D mutant cells caused a decrease in protrusiveness and a decline in adhesion assembly and disassembly rates. These phenotypes would be expected to arise from decreasing Rac activity, an outcome that was confirmed by the model described by Eqs (7) – (11) (see Methods Section). Indeed, after eliminating the Paxp-GIT, GIT-PIX, or PIX-PAK interactions in the model, through setting their respective association constants kC = 0 s-1, kG = 0 s-1, or kX = 0 s-1, we found that the induced state cannot be reached, even after setting initial level of phosphopaxillin P0*=1. GIT-PIX binding could be disrupted experimentally in two different ways: by either generating a GIT binding-deficient mutant of PIX (labeled PIX-ΔGBD) or by generating a PIX binding-deficient mutant of GIT (labeled GIT-ΔSHD). Transfecting CHO-K1 cells with the PIX-ΔGBD mutant did not cause any changes in the general behaviour of the cells (S3 Fig), as their migration speed (A), average adhesion size (B) and adhesion assembly/disassembly rates (C and D, respectively) remained effectively unchanged when compared to those of cells expressing wild type PIX. These results were consistent with previous observations showing that adhesion disassembly rate remained unchanged when D mutant cells were transfected with PIX-ΔGBD, in contrast to cells transfected with GIT-ΔSHD which exhibited a significantly slower adhesion disassembly rate [25]. The observed effect of the GIT-ΔSHD mutant was supported by the model, which showed that the induced state became unattainable when kG = 0 even if P0*=1. The apparent discrepancy between the effects of the PIX-ΔGBD and GIT-ΔSHD mutants is likely due to the fact that CHO-K1 cells, transfected with PIX-ΔGBD, also express endogenous PIX, making these cells overexpress this protein (unlike the GIT-ΔSHD transfected cells). That is, the overexpression of PIX could lead to high PIX activity across the cell, and as a result cause more dynamic adhesions in spite of the lack of PIX binding to GIT. To test this latter hypothesis, we plotted in Fig 3 the two-parameter bifurcation of the model, described by Eqs (7) – (11) (see Methods Section), with respect to GIT-PIX binding rate (kG) and PIX concentration ([PIX]), to trace the location of the two saddle nodes delimiting the bistable regime of Fig 2D, as both of these parameters are changed simultaneously. Our results revealed that the boundary of the bistable regime (gray area, bounded to the left by a very small monostable regime of uninduced states (white) and to the right by a large monostable regime of induced state (white) ), are gradually dominated by the monostable regime of induced states, shrinking as the value of the maximum paxillin-phosphorylation rate increased from B = 2 s-1 (A) to its default value of B = 4. 26 s-1 (B) and finally to B = 50 s-1 (C). In all cases, increasing [PIX] always mediated switching to the monostable induced state when kG = 0, independent of the current state of the system (i. e. , being induced or uninduced) and of the initial level of phosphopaxillin (P0*). This suggests that PIX overexpression induces Rac activation even if GIT-PIX binding is impaired, an outcome that is in agreement with the PIX-ΔGBD results of S3 Fig. Given that paxillin S273 phosphorylation/dephosphorylation rates appeared to play a key role in determining the steady state levels of active Rac and Rho, we asked next whether varying the level of phosphopaxillin in space could cause spatial gradients of Rac and Rho activity to form. To do so, we simulated our full spatiotemporal model of Rac/Rho activation and paxillin phosphorylation, given by Eqs (1) – (6) (see Methods Section), starting from spatially uniform initial levels of active Rac R*, and active Rho ρ, and a gradient of phosphopaxillin P*, that decreased sigmoidally from the cell front to the rear (see the heat-maps of Fig 6A). These initial conditions led to polarization-like effects in active Rac R* (right) and active Rho ρ (left), generating a region of high Rac/low Rho activity near the front of the cell at x = 0 μm, and a second region of low Rac/high Rho activity near the cell rear at x = L μm. This polarization-like effect closely resembled those previously seen and characterized in computational models of Rac-Rho cycling, in which decelerating (post transient) stationary waves were formed in a phenomenon termed wave-pinning [13,14,66]. Defining the boundary between the two regions of high Rac/low Rho and low Rac/high Rho as the value of x where the levels of the active forms of both GTPases were halfway between their maximum and minimum levels, we saw that as the maximum paxillin-phosphorylation rate B increased from B = 0 s-1, the position of the boundary initially increased sharply from x ≈ 4. 25 μm, followed by a plateauing phase and then by a second sharp increase towards x ≈ 6. 5 μm near B ≈ 22 s-1 (see Fig 6C). In other words, the boundary between the two polarized regions of Fig 6A gradually shifted to the back of the cell as B increased. Beyond B ≈ 22 s-1, the difference between the levels of these two GTPases in the two regions of polarization dropped significantly and became very small and visually indistinguishable from each other (Fig 6D). Effectively, this suggests that when B > 22 s-1, there is a relatively uniform level of both R* and ρ throughout and a loss of polarization in Rac and Rho activities. Since spatial differences in the initial levels of P* were indeed capable of inducing polarization-like effects, we next examined whether a spatial difference in the maximum paxillin-phosphorylation rate, B, could generate a similar gradient of paxillin S273 phosphorylation and, in turn, have similar effects to those seen in Fig 6A. By starting with spatially uniform levels of ρ, R*, and P*, and values of B which decreased sigmoidally from the cell front to rear, the heat-maps of Fig 6B revealed that it was indeed possible for the model to generate distinct regions of Rac R* (right) and Rho ρ (left) activities that remained stable over time. It is important to point out that since the system was very sensitive to increases in B, we found that this polarization could only occur when the spatially uniform initial level of P* was very low (at most P* ≈ 0. 1). If the initial level of P* was too high, the system became uniformly induced (an expected outcome in view of Fig 2A). Paxillin phosphorylation at S273 has been shown to play a crucial role in determining cell movement. This was shown in CHO-K1 cells, expressing two different paxillin-S273 mutants (A and D), that exhibited distinct phenotypes in terms of adhesion dynamics and cellular motility (slow versus fast, respectively). Such dichotomy in the dynamics of adhesions along with paxillin suggests the presence of bistability, a common feature in many biological systems. To investigate this very feature and the underlying dynamics of this system, we developed a molecularly explicit mathematical model to examine how paxillin phosphorylation and its binding to the GIT-PIX-PAK complex affect Rac activation. The model was based on previous models of crosstalk between GTPases that exhibited bistability [13,14]. In the presence of several positive and negative feedback loops that directly and indirectly affect the Rac-Rho subsystem, the model preserved its bistable switch. Indeed, for the parameter values detailed in Table 1, we found that the model resulted in two stable steady states, one corresponding to the induced state (high Rac/low Rho activity) and another corresponding to the uninduced state (low Rac/high Rho activity). The bistable switch exhibited very pronounced difference between the induced and uninduced states in active Rac and Rho, but not in phosphopaxillin, and displayed features that were more in line with observed experimental data. In this study, we analyzed the steady state properties of the model and showed how perturbing various components of it affects outcomes. That allowed us to compare different model variations and identify the role of paxillin, PIX, PAK, γ = [PAKtot]/[Ractot] and other molecular intermediates (such as GIT-PIX, GIT-PIX-PAK, PIX-PAK and PIX-PAK-RacGTP complexes) in defining dynamics, particularly bistability. The bistable switch produced by the model was shown to be not only dependent on the paxillin S273 maximum phosphorylation rate, but also on its dephosphorylation rate. The bifurcation diagrams of active Rho and Rac with respect to the phosphorylation (dephosphorylation) rate exhibited a regime of bistability flanked to the left (right) and right (left) by the monostable regimes of uninduced and induced states, respectively. The existence of this bistable regime (bounded by two saddle nodes) produced a memory effect, where the final state of the system (whether induced or uninduced), obtained by varying one of these two parameters, was not necessarily the same as that obtained at default value. Furthermore, contrary to the bistable switch obtained by the model presented in [37], the levels of active Rac and Rho in the induced and uninduced states varied marginally within each state in the bistable switch, allowing for the level of active Rac (along with active Rho) to show a plateauing profile in the induced and uninduced states similar to that seen when inhibiting PAK activation with different concentrations of IPA-3 [37]. Although the two models had several common features, they differed in three main aspects: their parameter values, their detailed molecular nature, and their description of how RhoGTP affects Rac activity. For the latter, the model presented here assumed (as in [8,44,45]) that the active forms of Rac and Rho mutually inhibit each other by downregulating each other’s GEFs, but in [37], active Rho was assumed to activate Rac inactivation through GAP. Replacing Rho-dependent inhibition of Rac-GEF by Rho-dependent activation of Rac-GAP in the model could produce similar or related outcomes to those obtained here [14], but not necessarily within the same parameter regime identified as being physiologically-relevant (due to the dependence of several reactions of the model on the steady state expressions of other intermediates). Thus, it would be interesting to study how changing the effect of active Rho on Rac in the model alters/shift dynamics within the parameter space. In agreement with experimental results, the effects of PAK activation/inactivation on the dynamics of Rac and Rho qualitatively mimicked those of paxillin S273 phosphorylation/dephosphorylation. We also discovered that, unlike with the maximum paxillin-phosphorylation rate, it was possible to recover the induced state by increasing the PAK activation rate even if the Paxp-GIT or GIT-PIX interactions were disrupted. The formation of the PIX-PAK-RacGTP complex was shown to be necessary for this recovery, highlighting the importance of this complex in ways not previously identified experimentally. Interestingly, if the value of total PAK-to-total Rac ratio (i. e. , γ) was increased from its default value of 0. 3 to the unphysiological value of 0. 5, we saw a more robust recovery to the induce state, even if the S273 paxillin signalling pathway was impaired. In fact, we found that, at higher γ, this recovery no longer required the existence of the PIX-PAK-RacGTP complex, although recovery was more difficult in the absence of this complex. The results obtained for both cases (i. e. , γ = 0. 3 and γ = 0. 5), led to the idea that PAK could recover Rac activation even when there were deficiencies in either paxillin S273 phosphorylation or formation/binding of the GIT-PIX-PAK complex, achieved through hyper-activation of PAK. All these model predictions involving PIX-PAK-RacGTP and γ remain to be validated experimentally. Simulation of the full spatiotemporal (reaction diffusion) model suggested that a spatial difference in the level of phosphopaxillin or in the rates of paxillin S273 phosphorylation/dephosphorylation could produce gradients in Rac and Rho activity across the cell. Based on this, we concluded that regulation of paxillin S273 phosphorylation may aid in establishing polarity gradients and cell directionality through Rac. The polarization-like effects we observed were similar to those seen in minimal models of Rac-Rho interplay [38]. Using these minimal models, Holmes et al. demonstrated the existence of parameter regimes corresponding to cell spreading, rounding, and polarization. It is likely that these regimes also exist for the model presented here, and is a possible avenue of further pursuit. Interestingly, experimental evidence has previously shown that phosphopaxillin appeared in adhesions near the leading edge of the cell [25], accompanied by Rac activation prior to and during lamellipodial protrusion [9]. These results, along with the findings of the current study, suggested the possibility of a relationship between paxillin S273 phosphorylation and lamellipodial Rac activity that may lead to the propagation or maintenance of lamellipodial protrusions. In addition to implications in establishing polarity gradients, phosphorylation of paxillin S273 may be related to certain cancerous phenotypes through deregulation of its phosphatase PP2A [4,67]. Inhibition and truncation of PP2A led to increased motility and metastasis [28,68], leading to the conclusion that certain cancerous phenotypes may be reversed by increasing PP2A activity to subsequently decrease paxillin phosphorylation and thus Rac activity. Fig 2D suggested, however, that while inhibiting and inducing paxillin dephosphorylation could increase and decrease the levels of Rac activity, respectively, switching from a pathological state to a physiological state may not be as intuitive as returning PP2A activity to normal physiological levels. If the range of normal dephosphorylation rates were within the bistable regime, for example, then the dephosphorylation rate would have to be increased beyond normal levels (past the right saddle node in Fig 2D) in order for cells to return back to the uninduced state. In order to validate the predictions presented here, the activities of Rac and/or Rho (or signaling molecules downstream of Rac and Rho) must be measured while B and αR are experimentally varied. One of the techniques used to achieve this is to measure the activities of FRET-based Rac and Rho biosensors while using chemical potentiators or inhibitors that activate or inhibit protein activities [41]. Here, we administered OA to decrease the paxillin dephosphorylation rate through the inhibition of PP2A. Although the cells had varying initial levels of active Rac, incubation of these cells with OA led to increased Rac activity, followed by partial recovery to normal levels after washout. We hypothesized that the absence of complete recovery to the normal level (after washout) was due to both the the presence of hysteresis and the close proximity of the left saddle node of Fig 2D to the vertical axis of that bifurcation diagram, limiting the recovery through this pathway. Experiments that detect the memory predicted in this system can also be done by measuring the activities of Rac and Rho at different concentrations of OA in a manner similar to what was done in [37] when measuring Rac and Rho activities in response to IPA-3. When the concentration of OA is increased from zero, Rac activity would be expected to increase as more cells enter the induced state. If, in the same experiment, the concentration of OA is gradually decreased, we would expect a decrease in the level of Rac activity. To demonstrate bistability, the EC50 (Rac) /IC50 (Rho) of the OA-dependent dose response curves obtained from the two experiments should be different. Bistability in our model was verified by considering the assembly and disassembly rates of adhesion in wild type cells as well as in paxillin S273 A and D mutant cells. These rates were taken to be reflective of Rac and Rho activities, as faster (slower) assembly/disassembly rates were associated with high (low) Rac/low (high) Rho activities. Our results revealed that when plotting these rates in terms of adhesion sizes, wild type cells exhibited bimodality in their assembly/disassembly rates. Smaller adhesions were more dynamic while larger adhesions had slower assembly and disassembly rates. Interestingly, these rates were unimodal for the A and D mutants; the former exhibited only slow rates and the latter exhibited only high rates independent of adhesion size. We suggested that differences in initial level of phosphopaxillin (i. e. , all molecular species containing Paxp) was responsible for generating the two subpopulations of adhesions in cells expressing wild-type paxillin, while differences in the maximum phosphorylation rate of paxillin were responsible for the appearance of only one subpopulation for both the A and D mutants. Shifts and changes in the bistable switch obtained by the model when introducing certain binding deficiencies similar to those tested here (e. g. , disrupting Paxp-GIT, GIT-PIX, or PIX-PAK interactions), could be qualitatively validated experimentally by checking for frequencies of adhesions with fast vs. slow assembly and disassembly rates when such deficiencies were introduced. This was done with the GIT-PIX intermediate (but not for PIX-PAK intermediate), in which CHO-K1 cells were transfected with the GIT binding-deficient mutant of PIX. These cells did not exhibit any changes in their adhesion assembly/disassembly, in contrast to those transfected with the PIX binding-deficient mutant of GIT [25]. Using the model, we showed that the overexpression of PIX (a known Rac-GEF) in the former case caused these cells to become induced regardless of the state of the system prior to transfection and of the initial level of phosphopaxillin (P0*). This study clearly demonstrates the value of combined molecular modeling and experimental work. The model can now be used to guide future experiments, which in turn can help refine the model. In our future efforts, we aim to extend these approaches to further explore the molecular details of other intermediates regulating cell migration, and to develop phenomenological, but physiologically relevant, low dimensional model (based on the interactions between the three key variables: active Rho, active Rac and phosphopaxillin, incorporated into the model presented here) to study analytically cellular polarization and the phenomenon of wave-pinning [66].

Cellular migration is crucial in both physiological and pathological functions. Maintenance of proper migration and development of aberrant migration are effectuated by cellular machinery involving protein complexes, called adhesions, that anchor the cell to its environment. Over time, these adhesions assemble at the leading edge, as the cell extends forward, anchoring the front of the cells to its substrate, while those at the cell rear disassemble, allowing detachment and forward movement. Their dynamics are controlled by a number of regulatory factors, occurring on both cell-wide and adhesion-level scales. The coordination of these regulatory factors is complex, but insights about their dynamics can be gained from the use of mathematical modeling techniques which integrate many of these components together. Here, we developed several molecularly explicit models to explore how local regulation of paxillin, an adhesion protein, interacts with the activities of Rac and Rho to produce cell-wide polarization associated with motility and directionality. By altering paxillin phosphorylation/dephosphorylation within such models, we have advanced our understanding of how a shift from a non-motile state to a highly motile state occurs. Deciphering these key processes quantitatively thus helped us gain insight into the subcellular factors underlying polarity and movement.

Abstract Introduction Methods Results Discussion

cell binding phosphorylation cell physiology cell motility ecology and environmental sciences chemical compounds atmospheric science enzymes enzymology organic compounds fluorophotometry developmental biology serine amino acids research and analysis methods fluorescence resonance energy transfer proteins chemistry guanosine triphosphatase atmospheric chemistry spectrophotometry physics biochemistry mass diffusivity environmental chemistry hydrolases cell biology post-translational modification organic chemistry earth sciences cell migration carbon dioxide biology and life sciences hydroxyl amino acids physical sciences greenhouse gases chemical physics spectrum analysis techniques

2018

Paxillin phosphorylation at serine 273 and its effects on Rac, Rho and adhesion dynamics

Active and repressed ribosomal RNA (rRNA) genes are characterised by specific epigenetic marks and differentially positioned nucleosomes at their promoters. Repression of the rRNA genes requires a non-coding RNA (pRNA) and the presence of the nucleolar remodeling complex (NoRC). ATP-dependent chromatin remodeling enzymes are essential regulators of DNA-dependent processes, and this regulation occurs via the modulation of DNA accessibility in chromatin. We have studied the targeting of NoRC to the rRNA gene promoter; its mechanism of nucleosome positioning, in which a nucleosome is placed over the transcription initiation site; and the functional role of the pRNA. We demonstrate that NoRC is capable of recognising and binding to the nucleosomal rRNA gene promoter on its own and binds with higher affinity the nucleosomes positioned at non-repressive positions. NoRC recognises the promoter nucleosome within a chromatin array and positions the nucleosomes, as observed in vivo. NoRC uses the release mechanism of positioning, which is characterised by a reduced affinity for the remodeled substrate. The pRNA specifically binds to NoRC and regulates the enzyme by switching off its ATPase activity. Given the known role of pRNA in tethering NoRC to the rDNA, we propose that pRNA is a key factor that links the chromatin modification activity and scaffolding function of NoRC. Nucleosomes present a major obstacle for the binding of sequence-specific DNA-binding factors, the interaction of positively charged histone tails with DNA and the masking of DNA binding sites that face in towards the histone octamer surface [1], [2]. As a result, all DNA-dependent processes, such as transcription, replication, repair and recombination, are affected by the positioning of nucleosomes on regulatory sites. ATP-dependent chromatin remodeling enzymes, which use energy from ATP hydrolysis to slide, evict or replace histones within nucleosomes, are key modulators of chromatin structure and DNA-dependent processes [3]. Thus, it is of particular importance to reveal their molecular mechanism of nucleosome remodeling, how these enzymes are targeted to their genomic loci and their role in defining nucleosome positions in vivo [4]–[10]. In mammalian cells, there are numerous types of remodeling enzymes that associate with different subunits to form remodeling complexes with distinct biological functions. Due to the high combinatorial complexity, it is estimated that several hundred different chromatin remodeling complexes exist in humans. These remodeling enzymes comprise several groups of ATPases classified into the Snf2, ISWI, Mi-2, Chd1, Ino80, ERCC6, ALC1, CHD7, Swr1, RAD54 and Lsh subfamilies [9], [11]. In addition to their diversity, chromatin remodeling enzymes are highly abundant, with approximately one enzyme for every 10 nucleosomes in yeast and human cells [5], [10]. Remodeling enzymes preferentially localise to specific genomic regions, raising the questions of which signals target the enzymes to these locations and what their functions are at these sites [12], [13]. Recently, the continuous sampling model was suggested for the abundant ISWI type remodeling enzymes. According to this model, the enzyme continuously samples all nucleosomes by transiently binding and dissociating without translocation. Only upon introducing additional signals, such as the direct interaction with sequence-specific DNA-binding factors, histone modifications and altered DNA/nucleosome structures, do these nucleosomes become marked as “to be translocated” by converting them to high-affinity substrates [13]. However, there is still a lack of mechanistic proof for the continuous sampling model. Active rRNA genes cover the promoter-bound nucleosomes from −157 to −2 (relative to the transcription start site), compatible with the binding of the UBF and TIF-IB/SL1 factors required for transcription initiation [8]. On repressed genes, the nucleosome is shifted 24 nt downstream, occluding the TIF-IB binding site [8], [14]. NoRC (nucleolar remodeling complex), which is an ISWI type remodeling enzyme that consists of two subunits, Tip5 (TTFI interacting protein 5) and the Snf2H ATPase, is required to establish the repressed rRNA genes and initiate heterochromatin formation [15], [16]. NoRC is recruited to the rRNA gene by the Transcription Termination Factor-I (TTF-I), which binds upstream of the gene promoter [17]. Recent studies have revealed that NoRC also interacts with the pRNA (promoter-associated RNA), a 150–200 nt long non-coding RNA that is complementary to the rRNA gene promoter sequences and is required for efficient rRNA gene silencing and subsequent DNA methylation [18], [19]. We addressed whether NoRC affects the architecture of the repressed rRNA gene, its mechanism of nucleosome positioning and how the enzyme is targeted to the promoter nucleosome. We demonstrate that, within arrays of nucleosomes, NoRC is capable of recognising the rRNA gene promoter nucleosome with a higher affinity than that for other nucleosomes and that it specifically repositions the nucleosome to the site that was observed in vivo. We show that the mechanism of positioning corresponds to a release model of nucleosome positioning, in which NoRC has a reduced affinity for the remodeled substrate. We further studied the role of the pRNA-NoRC interaction and observed that this RNA serves as a negative regulator of NoRC activity, indicating that tight regulation of these enzymes reduces the wasteful turnover of ATP when maintained within chromatin. The remodeling complex NoRC, consisting of the Snf2H and Tip5 subunits, was expressed using the baculovirus system and purified to apparent homogeneity (Figure 1A). The activity of NoRC was tested on mononucleosomal substrates reconstituted on the 601 nucleosome positioning sequence in the centre or at the border of the DNA fragment ([20], [21], Figure 1B and S1). The end-positioned nucleosomes were repositioned by NoRC to the central locations in an ATP-dependent remodeling reaction (Figure 1B, upper panel). In contrast, when the nucleosomes were located at the centre of the DNA fragment, only minor ATP-dependent effects were detected (Figure 1B, lower panel). The initial analysis indicated that the recombinant NoRC complex was active but required a specific nucleosomal substrate for its activity. One of the features of the nucleosomal substrate is the linker DNA. To test whether linker DNA is required for NoRC function, we analysed the ATPase activity of NoRC in the presence of nucleosomal arrays and mononucleosomes with and without linker DNA (Figure 1C and S1D). Interestingly, mononucleosomes lacking linker DNA stimulated the ATPase activity of NoRC significantly less than the linker-containing mononucleosomes or nucleosomal arrays. This experiment suggests that recognition of the nucleoprotein structures in the core nucleosome by NoRC activates its ATPase activity but that linker DNA is required for full stimulation. Next, to determine the minimal length of DNA required for NoRC binding, we carried out DNA-binding experiments using a mixture of DNA molecules with different lengths (from 10 to 130 bp in 10 bp increments, Figure 1D). Quantification of DNA: NoRC complexes in a competitive assay revealed that the DNA-binding affinity of NoRC strongly decreases with DNA lengths below 60 bp and that the remodeler does not significantly bind to DNA of 40 bp or shorter. Initial experiments did not demonstrate that Tip5 or NoRC have any sequence-specific DNA binding activity (data not shown). However, NoRC may recognise DNA with a particular structure. Therefore, initial binding of Tip5 to cruciform DNA was analysed. Cruciform DNA and two linear, double-stranded 40 bp DNA fragments (‘DNA sequence controls’) were prepared as described [22]. Increasing amounts of Tip5 were incubated with either the cruciform DNA or the linear DNA and analysed in an electromobility shift assay (EMSA). No binding of Tip5 to either of the linear DNA fragments was visible under the experimental conditions (Figure 1E, panels 1 and 3). In contrast, the incubation of Tip5 with the cruciform DNA resulted in the formation of protein/cruciform DNA complexes (panel 2). The experiment shows preferential binding of NoRC to structured DNA. To test whether linker DNA is required for a stable interaction of NoRC with the nucleosomes, EMSAs using reconstituted mononucleosomes containing linker DNA of 0 bp (146 bp template), ∼25 bp (171 bp template), ∼50 bp (247 bp template, centrally positioned nucleosome) and ∼100 bp (247 bp template, end-positioned nucleosome) and increasing amounts of NoRC were performed (Figure 1F). NoRC bound with similar affinity to the DNA molecules ranging in length from 146 bp to 247 bp, forming discrete NoRC: DNA complexes as expected from the previous experiment. However, when this DNA was reconstituted into nucleosomes, NoRC failed to form a stable complex with the nucleosomes containing 0 bp and 25 bp of linker DNA but formed discrete NoRC-nucleosome complexes with nucleosomes bearing 50 or 100 bp of linker DNA (Figure 1F). Thus, NoRC has a higher binding affinity for free DNA than nucleosomal cores, which suggests that linker DNA is required for efficient targeting of NoRC to remodeling sites. To determine the relative orientation of NoRC when bound to the nucleosome, we performed DNase I footprinting experiments. Nucleosomes were reconstituted on the central position of the radioactively end-labelled 247 bp mouse rDNA promoter fragment, a known target site of NoRC [15]. Free DNA, nucleosomes and NoRC-nucleosome complexes were incubated with DNase I, the reaction was stopped by the addition of EDTA and the reaction products were resolved by EMSA (Figure 2A, B). Free DNA, nucleosomes and the corresponding NoRC-nucleosome complexes were gel-purified and further analysed on sequencing gels. When compared to free DNA, DNase I digestion of the nucleosomal DNA resulted in a characteristic cleavage pattern, revealing sites of protection and a repeated pattern of DNase I-sensitive sites with a distance of approximately 10 bp, indicating a nucleosome positioned in the centre of the rDNA fragment (Figure 2C). Because a natural DNA sequence was used in this study, the nucleosome lacked precise positioning and a mixture of rotationally phased nucleosomes broadened the protected region [23]. To avoid the formation of multimeric complexes or template precipitation, NoRC was incubated with the nucleosomal substrates at concentrations that result in 50–70% complex formation. NoRC significantly protected the borders of the nucleosome and the adjacent linker DNA from DNase I digestion (Figure 2D). Our data suggest that the binding of NoRC to the nucleosome is bilateral, interacting with both exit and entry sites of the nucleosome, and confirms that NoRC binds to the linker DNA. To examine the ability of NoRC to reposition nucleosomes on its target site, we reconstituted mononucleosomes on a DNA fragment containing the rRNA gene promoter sequence in vitro (position −190 to +90, relative to the transcription start site). Nucleosomes reconstituted on the rDNA promoter region occupied multiple positions on the DNA, as demonstrated by native gel electrophoresis (Figure 3A, lane 1). NoRC dependent remodeling establishes a preferential nucleosome position that is located close to the center of the DNA (Figure 3A). This nucleosome position was characterised by Exonuclease III footprinting, showing that it protected the DNA from positions −120 to +27 (Figure S2). This position correlates well with the nucleosome position of the repressed rRNA genes in vivo [8]. This suggests that NoRC recognises specific DNA sequences or structures on the nucleoprotein complex that allow site-specific positioning. A common feature of ribosomal gene promoters is that they lack sequence homology but retain structural similarity and contain intrinsically distorted regions [24]. The relative DNA curvature of the mouse rDNA promoter was calculated with the Bolshoy algorithm using the ‘bandit’ program (Figure 3B, [25], [26]). The mouse rRNA gene promoter contains a region of high local DNA curvature ([25]; at about position −110) that is specifically bound by Tip5 (Figure 3C). This result agrees with the results of the previous experiment, which demonstrated the preferential binding of Tip5 to cruciform DNA (Figure 1E). Thus, these data indicate the specific recognition of structured DNA by the remodeling enzyme, suggesting a potential mechanism for targeting NoRC to the rRNA gene promoter. Two kinetic models were proposed to explain how chromatin remodelers are able to direct the nucleosome to a specific position on DNA [9]. The release model implies that remodelers bind with high affinity to nucleosomes positioned at the “wrong” sites and remodel the nucleosome until it reaches the final (correct) position. The nucleosome at the final position exhibits the lowest affinity for the remodeling enzyme and is thus the worst substrate for the remodeling enzyme. In contrast, the arrest model postulates that the nucleosome exhibits a much higher affinity for the remodeling enzyme at the final position, locking it on the nucleosome and reducing the catalytic conversion rate [9], [27]. To assign one of the kinetic models for a particular remodeler, the binding and remodeling of nucleosomes must be compared. Thus, we compared the differential binding affinities of NoRC to the individual nucleosome positions by EMSA. The incubation of rDNA −190/+90 reconstituted into nucleosomes with increasing concentrations of NoRC resulted in a stepwise binding of the different nucleosome species (Figure 3D). Free DNA and most of the nucleosomes were bound with similar affinities and retarded in the gel. However, the nucleosome occupying the −120/+27 position bound with the lowest affinity. This nucleosome position is the final position of the NoRC-dependent remodeling reaction (Figure 3A), revealing that NoRC has the lowest binding affinity for the “remodeled” nucleosome, therefore suggesting that NoRC remodels nucleosomes according to the release model. Differential local binding affinities are required to position nucleosomes on DNA. However, on a more global scale, differential binding affinities could also serve to target the remodeling enzymes to specific genes and regulatory regions. To test how NoRC and Snf2H select their remodeling targets, we used competitive binding and remodeling assays. Nucleosomes were reconstituted on a fluorescently labelled rRNA gene promoter fragment (Cy5 labelled) and the 601 nucleosome positioning sequence ([20], Cy3 labelled). Nucleosomes were mixed and binding or remodeling reactions were performed with increasing amounts of remodelers. Snf2H bound with similar affinity to both nucleosome substrates, and remodeled them with similar efficiency (Figure 3E, F). In contrast, NoRC showed preferential binding to the nucleosomes reconstituted on the rRNA gene promoter, preferentially binding the DNA and nucleosomes at lower NoRC concentrations when compared to the 601 substrate (Figure 3E, lanes 8 to 12). Binding with higher affinity was mirrored in the remodeling assay where NoRC was remodeling the rRNA gene promoter nucleosomes prior to the 601 nucleosomes (Figure 3F and Figure S3). As cellular nucleosomes are arranged in arrays, we tested whether NoRC is also capable of selectively recognising and repositioning the rRNA gene promoter nucleosome within nucleosomal arrays. Chromatin was reconstituted using the salt dialysis method on a circular DNA containing the rRNA gene promoter and incubated with NoRC or ACF in the presence of ATP. A partial MNase digestion of the nucleosomal DNA was performed and analysed in a primer extension reaction (Figure 3G). ACF did not qualitatively change the distribution of the nucleosomes within the analysed region of the rRNA gene promoter. However, NoRC induced a specific relocalisation of the promoter nucleosomes, placing the 3′ end of the nucleosome at position +22. NoRC-dependent nucleosome positioning at +22 perfectly corresponds to the cellular nucleosomal configuration of the repressed rRNA gene [8]. The 5 bp discrepancy between the mononucleosome remodeling and array remodeling assay could arise from internucleosomal interactions that influence the remodeling outcome. Our data strongly support the hypothesis that nucleosome remodeling complexes determine nucleosome positioning in vivo, thereby directly affecting gene expression. Previous studies have revealed a specific interaction between TTF-I and NoRC, suggesting that TTF-I recruits NoRC to the rRNA gene promoter [15], [17]. The results described here reveal an additional targeting signal, encoded by the high affinity of NoRC for nucleosomes positioned at “wrong” sites of the rDNA promoter. TTF-I improves the efficiency of NoRC recruitment to the rRNA gene promoter without affecting the outcome of the NoRC-dependent nucleosome remodeling reaction (Figure S4). Recent studies have demonstrated that NoRC binds to a non-coding RNA, which is initiated upstream of the rRNA gene promoter and contains promoter sequences in the sense orientation. It was suggested that promoter RNA (pRNA) is required to tether NoRC to inactive rRNA genes, where it establishes repressive epigenetic marks [18], [19], [28]. We studied two pRNA constructs that exhibit strong and weak binding affinities for Tip5, pRNA−143/−39 and pRNA−113/−39, respectively [19]. pRNAs were generated by in vitro transcription, re-natured and added to the remodeling reactions (Figure 4). First, the presence of the pRNAs did not influence the nucleosome positioning behaviour of NoRC. Second, we observed specific inhibition of the NoRC-dependent remodeling reaction with increasing levels of pRNA−143/−39 (Figure 4A). In contrast, Snf2H was similarly inhibited by both pRNAs, suggesting that the Tip5 subunit determines RNA-binding specificity and activity. Moreover, NoRC recognises the secondary structure of the pRNA, as inhibition of its nucleosome-remodeling activity was lost when the stem-loop structure was mutated (Figure 4B). We identified a regulatory role of the pRNA, demonstrating that the non-coding RNA serves as an inhibitor of the remodeling enzyme. To gain more insight into the inhibitory mechanism of pRNA, we investigated the effect of pRNA on NoRC ATPase activity. The incubation of NoRC with an increasing amount of DNA or pRNA only modestly stimulated the NoRC ATPase activity (Figure S5), whereas the presence of nucleosomes considerably accelerated ATP/ADP exchange. The incubation of NoRC with nucleosomes and increasing amounts of pRNA−143/−39 or pRNA−113/−39 resulted in a RNA concentration-dependent inhibition of the ATPase activity (Figure 5A). As in the remodeling reaction, pRNA−143/−39 inhibited the NoRC-dependent ATPase activity more efficiently than pRNA−113/−39, confirming the higher binding affinity of the remodeling complex for this RNA and explaining the inhibition of the nucleosome remodeling reaction. To reveal the mode of RNA-dependent inhibition, we studied the binding of NoRC to nucleosomes in the presence of RNA (Figure 5B). A competitive EMSA revealed that the pRNA competes with nucleosomes for NoRC binding, indicating that only exclusive NoRC: pRNA or NoRC: nucleosomes complexes exist. Again, competition of nucleosomes from the NoRC: nucleosome complex required less pRNA−143/−39 than pRNA−113/−39, indicating the higher binding affinity of pRNA−143/−39 for NoRC (Figure 5B). Both RNA species competed similarly with Snf2H, pointing to the specific role of Tip5 in NoRC (lanes 13 to 24). In summary, our data demonstrate that pRNA competes with nucleosomes for NoRC binding and therefore directly interferes with its ATPase activity and the nucleosome remodeling reaction. Ribosomal genes present an ideal model system for studying the dynamics and mechanism of chromatin remodeling, as the epigenetic marks, the chromatin structure of the active and repressed genes and the factors involved are well characterised. Active rRNA genes contain a nucleosome covering the gene promoter from positions −157 to −2, allowing the binding of UBF and TIF-IB/SL1 to their recognition sites at the nucleosomal borders [8]. In contrast, repressed genes have a nucleosome covering the positions from −132 to +22 relative to the transcription start site, masking the binding site of TIF-IB. The repression of rRNA genes is intimately linked with the recruitment of NoRC, which induces nucleosome remodeling, gene repression and the acquisition of heterochromatic marks [16]. We show that the activity of NoRC is sufficient for recognition of the promoter structure and nucleosome positioning in vivo. Nucleosomal arrays are required to establish the cellular nucleosome positioning pattern, suggesting that internucleosomal interactions influence the activity of remodeling enzymes. Our results are in good agreement with data demonstrating that ISWI machines are molecular rulers and potentially act in the context of di-nucleosomes [30], [31]. Although NoRC does not serve as a sequence-independent spacing factor, it is capable of recognising sequence features of the rRNA gene promoter, which serve as positioning signals. Several studies have demonstrated the importance of positioned nucleosomes in the genome [32]. However, irrespective of the ability of many sequences to position nucleosomes in vitro they fail to do so in vivo [33], [34], suggesting that there are additional mechanisms that structure chromatin. We show that NoRC positions nucleosomes according to the release mechanism [9], [13]. The enzyme binds with high affinity to nucleosomes positioned at “wrong” sites, which is the recruitment signal. The remodeling reaction is highly processive, with ACF moving a nucleosome for approximately 200 bp without leaving the nucleosomal substrate [35]. After initiation of the remodeling reaction, the endpoint of the translocation reaction is determined by a reduced affinity of the remodeler for the nucleosome at this site. As any remodeler with distinct binding affinities to nucleosomes at different but close positions on DNA could position nucleosomes, we suggest that chromatin remodeling enzymes serve to organise chromatin structure with respect to the underlying DNA sequence. The concentration and composition of the remodeling enzymes in combination with the specific targeting of those complexes to chromatin would determine a specific chromatin architecture and specify accessible regulatory sequences that determine the activity of DNA-dependent processes. We suggest that the combinatorial aspect of remodeling enzymes and complex constitution may determine cell types and their responses to the environment. There are a multitude of signals targeting remodeling enzymes to specific genomic regions, including direct recruitment by proteins, protein modifications, histone variants, coding and non-coding RNAs, as well as nucleosomes at “wrong” positions [13]. The continuous sampling model for chromatin remodeling enzymes suggests that high concentrations of remodeling enzymes and low binding affinities towards the non-signalling nucleosomes allow for efficient screening of the genome for signals that attract remodeling enzymes [10]. Here, we provide evidence for the continuous sampling mechanism of NoRC, where the remodeling enzyme selectively remodels the promoter nucleosome within an array of nucleosomes. Differential binding affinities guide the remodeling enzyme to these sites of action. However, on the genomic scale additional targeting signals help to further increase the local concentration of the remodeling enzymes at their sites of action. In the case of NoRC, interaction with TTF-I directly recruits NoRC and thereby improves the efficiency of the remodeling reaction, but does not influence the remodeling outcome [14], [17]. Previous studies have shown that the TAM domain in the Tip5 subunit interacts with pRNA and that this interaction is a prerequisite for maintaining NoRC in the nucleolus [18]. We show that pRNA competes with nucleosomes for NoRC binding, specifically inhibiting its ATPase activity. Therefore, we suggest that a ternary complex consisting of NoRC, nucleosomes and RNA does not exist, despite the fact that NoRC contains several DNA/nucleosome-binding domains and an RNA-binding TAM domain [36]. We suggest, that the pRNA serves three functions (Figure 5C). First, after replacing TTF-I at the rRNA gene promoter, it serves to maintain NoRC localisation at the promoter. Due to the release mechanism of nucleosome positioning, NoRC has a low affinity for the remodeled chromatin structure and most likely would dissociate from the promoter. Given, that Grummt and colleagues have shown that the 5′-end of the pRNA forms a triplex with the T0 site at the promoter region and that the 3′-end interacts with Tip5, we propose a tethering function for the pRNA. Switching off the ATPase activity of NoRC ensures that the nucleosome is stably maintained in the OFF position and that the enzyme does not waste ATP. The pRNA and NoRC recruit DNA methyltransferases, histone deacetylases and histone methyltransferases to silence the rRNA genes [37]–[39] and recruit the silenced genes to the heterochromatin environment of the nuclear matrix [36]. The proteins were expressed in SF21 cells. N-terminally His tagged Snf2H with or without Tip5 was purified via Ni-NTA (Qiagen) chromatography. Flag tagged Snf2H and Acf1 were purified using M2 beads (Sigma) [14]. Murine rRNA gene promoter fragments of 146 bp (−231 to −86; positions relative to the transcription start site), 171 bp (−231 to −61), 247 bp (−231 to +16) and 280 bp (−190 to +90) were amplified by PCR from a plasmid containing the genomic DNA isolated from the NIH3T3 cell line (genbank access #KC202874. 1). To radioactively label the DNA fragments, α-32P dCTP was added to the PCR reaction mix. The 601 DNA and the pRNA were prepared by PCR as described [19], [40]. PCR products were used for nucleosome assembly reactions as described [23]. Nucleosomes were assembled according to Rhodes and Laskey using the salt gradient dialysis technique [41]. A typical assembly reaction (50 µl) contained 5 µg DNA, varying amounts of histone octamer, 200 ng BSA/ml, and 250 ng competitor DNA in high salt buffer (10 mM Tris, pH 7. 6,2 M NaCl, 1 mM EDTA, 0. 05% NP-40,2 mM β-mercaptoethanol). The salt was continuously reduced to 200 mM NaCl during 16–20 h. The quality of the assembly reaction was analysed on a 5% PAA gel in 0. 4× TBE followed by ethidium bromide staining. Nucleosomes reconstituted on the 247 bp rDNA promoter fragment display two distinct positions that can be separated by native gel electrophoresis [21]. Nucleosome mobility was assayed as described [42]. Briefly, reactions contained 4 nM Cy5 labelled DNA reconstituted into nucleosomes, 1 mM ATP, 100 ng/µl BSA, 1 mM DTT, 70 mM imidazole in Ex80 buffer (20 mM Tris pH 7. 6,80 mM KCl, 1. 5 mM MgCl2,0. 5 mM EGTA, 1 mM β-mercaptoethanol, 10% glycerol, 200 ng/µl BSA) and recombinant remodeling enzymes. Nucleosomes were incubated with NoRC for 45 min at 30°C. The reactions were stopped by the addition of 1700 ng CMV14 plasmid DNA and incubated for 15 min on ice. The nucleosome positions were analysed by electrophoresis on 5% PAA gels in 0. 4× TBE and fluorescence scanning. Tip5 binding to cruciform DNA was performed as described [22]. NoRC binding to the DNA and nucleosomes was studied by electromobility shift assays (EMSA). The substrates used in the assay were either radioactively or fluorescently labelled as indicated in the legends. Reactions were performed in Ex80 buffer and the indicated amounts of NoRC. Reactions were incubated for 45 min at 30°C and then analysed by native PAGE. Competitive titration experiments were performed using identical reaction conditions, containing 25 nM NoRC, 4 nM fluorescently labelled mononucleosomes and the indicated amounts of the indicated pRNA constructs. The reactions were analysed on 5% polyacrylamide gels in 0. 4× TBE and subsequent fluorescence scanning. NoRC/nucleosome and nucleosome DNase I footprinting experiments were performed as described [29]. Essentially, radioactively end-labelled DNA was reconstituted into nucleosomes and incubated with NoRC using the same experimental conditions as in the remodeling reactions. DNase I digestions were stopped by the addition of EDTA to a final concentration of 5 mM. The complexes were resolved on native PAA gels and the DNA, nucleosome and NoRC/nucleosome complexes were excised from the gel. DNA was purified and analysed on 7% sequencing gels. Mapping nucleosomal boundaries on nucleosomal arrays before, or after remodelling with NoRC or ACF was performed as described [14]. An ATPase reaction contained 150 ng of DNA or chromatin in 10 µl of Ex75 buffer, 10 µM ATP and γ32P-ATP (0. 1 µl; 3000Ci/mmol, Hartmann Analytic), the indicated amounts of pRNA−143/−39 or pRNA−113/−39 and 10 units RNasin. The reactions were initiated by the addition of the remodeling enzyme and incubated for 60 min at 30°C. Aliquots of 1 µl were spotted on thin layer cellulose chromatography plates (Merck) and air-dried. The hydrolyzed phosphate was separated from unreacted ATP by thin layer chromatography in 0. 5 M LiCl/acetic acid buffer. The plates were dried at 65°C for 5 min and exposed to Phospho Imager plates (FujiFilm BAS-1500). ATP and hydrolyzed phosphate spots were quantified using the Multigauge software (Fuji). The percentage of hydrolyzed ATP was calculated according to the following equation: Pi/ (ATP+Pi) ×100%, where Pi: amount of hydrolyzed radioactive phosphate; ATP: amount of left γ32P-ATP. Nucleosome positioning on the Cy5 5′ end-labelled mouse rDNA fragment (from positions −190 to +90 relative to the transcription start site) was determined with Exo III mapping. Reactions were carried out in an initial volume of 50 µl with 30 nM nucleosomes and 2 U/µl of Exo III (NEB) in 10 mM Tris, 90 mM KCl, 1 mM MgCl2, and 1 mM DTT at 16°C. At different time points 7 µl of the reaction mix were removed and the reaction was stopped by the addition of EDTA (final concentration of 50 mM). Proteins were digested with Proteinase K after the addition of SDS to a final concentration of 1% and the DNA was subsequently purified by ethanol precipitation. DNA samples were analysed on 6% sequencing gels. The DNA ladder was prepared with the DNA Cycle Sequencing Kit (Jena Bioscience) using a Cy5 labelled oligonucleotide and the mouse rDNA promoter fragment (−190 to +90), with either ddTTP or ddCTP in the reaction mix. Results were imaged with a FLA-5000 imager (Fujifilm). As control, we carried out Exo III digestions with naked DNA in order to discriminate nucleosome positions from exonuclease pause sites on free DNA. To map NoRC dependent positions a remodeling reaction was performed prior to Exo III analysis. Remodeling was performed with 7. 4 ng/µl of NoRC and Cy5 labelled nucleosomes in the presence or absence of 1 mM ATP for 60 min at 30°C. The reaction was stopped with competitor plasmid DNA and used for native gel analysis and Exo III footprinting.

Tumour cells overexpress ribosomal RNA (rRNA), which is required for ribosome assembly and cell growth. rRNA gene repression is mediated by the chromatin remodeling complex (NoRC) and a non-coding RNA that binds to this enzyme. This study addresses the mechanism of nucleosome positioning by NoRC and the functional role of the non-coding RNA, which is termed pRNA because it corresponds to the promoter sequence. NoRC recognises the promoter nucleosome in a chromatin array with high affinity and uses a release mechanism to position the nucleosome over the transcription initiation site. The pRNA binds specifically to NoRC and inhibits its ATPase activity. We suggest that the RNA retains NoRC at the gene promoter after remodeling, linking its chromatin modification and scaffolding activity to inactive rDNA copies.

Abstract Introduction Results Discussion Materials and Methods

biomacromolecule-ligand interactions biochemistry nucleic acids proteins enzymes genetics gene expression biology biophysics gene function

2014

Chromatin Targeting Signals, Nucleosome Positioning Mechanism and Non-Coding RNA-Mediated Regulation of the Chromatin Remodeling Complex NoRC

It has been proposed that switching from annual to biannual (twice yearly) mass community-directed treatment with ivermectin (CDTI) might improve the chances of onchocerciasis elimination in some African foci. However, historically, relatively few communities have received biannual treatments in Africa, and there are no cost data associated with increasing ivermectin treatment frequency at a large scale. Collecting cost data is essential for conducting economic evaluations of control programmes. Some countries, such as Ghana, have adopted a biannual treatment strategy in selected districts. We undertook a study to estimate the costs associated with annual and biannual CDTI in Ghana. The study was conducted in the Brong-Ahafo and Northern regions of Ghana. Data collection was organized at the national, regional, district, sub-district and community levels, and involved interviewing key personnel and scrutinizing national records. Data were collected in four districts; one in which treatment is delivered annually, two in which it is delivered biannually, and one where treatment takes place biannually in some communities and annually in others. Both financial and economic costs were collected from the health care provider' s perspective. The estimated cost of treating annually was US Dollars (USD) 0. 45 per person including the value of time donated by the community drug distributors (which was estimated at USD 0. 05 per person per treatment round). The cost of CDTI was approximately 50–60% higher in those districts where treatment was biannual than in those where it was annual. Large-scale mass biannual treatment was reported as being well received and considered sustainable. This study provides rigorous evidence of the different costs associated with annual and biannual CDTI in Ghana which can be used to inform an economic evaluation of the debate on the optimal treatment frequency required to control (or eliminate) onchocerciasis in Africa. Human onchocerciasis or river blindness is a neglected tropical disease (NTD) caused by the parasitic filarial nematode Onchocerca volvulus and transmitted by the bites of Simulium blackflies [1]. In addition to ocular pathology (vision loss, blindness), and increased host mortality [2], [3], onchocerciasis also causes disfiguring skin lesions and severe dermal itching that can drastically impair an individual' s quality of life [4]. In 1987, ivermectin was registered for human use against onchocerciasis, and Merck & Co. , Inc. took the unprecedented decision to donate ivermectin for as long as needed to eliminate onchocerciasis as a public health problem [5]. Two major onchocerciasis control programmes have been launched in Africa. The former was the Onchocerciasis Control Programme in West Africa (OCP), which started in 1974 and closed in 2002, and was initially based solely on vector control until ivermectin was licensed for human use in 1987. For the most part, the OCP used an annual treatment strategy (alone or in combination with antivectorial measures), but in the Western extension, some foci were treated biannually in the absence of vector control [6], [7]. Currently, the former OCP countries undertake their own national onchocerciasis control programmes. The African Programme for Onchocerciasis control (APOC) was launched in 1995 and it has recently been extended to 2025 [8]. It targets the 19 onchocerciasis endemic countries in Africa that were not covered by the OCP (though three of them, Kenya, Rwanda, and Mozambique, were found not to be endemic) [9]. APOC' s predominant strategy involves annual, community-directed treatment with ivermectin (CDTI) in areas where the prevalence of onchocercal nodules is greater than 20%, for all those aged five years and older (excluding pregnant or breastfeeding women in the first week after delivery) [9], [10]. Based on the experience in Uganda [11], and the success achieved in most onchocerciasis foci in the Americas [12], there have been recent discussions of switching to biannual treatments (twice yearly) to increase the feasibility of elimination. In the past, only a small number of foci within the OCP (such as River Gambia in Senegal [7]) have received biannual treatment in Africa, and therefore there are no ground-truth data on the cost associated with increasing the treatment frequency to twice per year on a large scale. (In Uganda, the cost of biannual CDTI was simply estimated by doubling that of the annual treatment [11].) Motivated by ivermectin efficacy studies suggesting sub-optimal responses of O. volvulus to the drug [13], [14], [15], Ghana (an ex-OCP country), has recently adopted a biannual treatment strategy at a large scale [15]. In Ghana, onchocerciasis is endemic in 9 out of 10 regions with a total at-risk population of approximately 3. 2 million [16]. Responsibility for ivermectin distribution—which occurs in 73 districts—was devolved from the OCP to Ghana in 2002 (under the supervision of APOC). Since 2006, onchocerciasis control has been implemented in the context of the Neglected Tropical Diseases Programme (NTDP) [16], and in 2009,40 (55%) districts started using a biannual ivermectin distribution strategy. The decision regarding which areas should change to the biannual treatment strategy was based on the combined results of rapid epidemiological mapping of onchocerciasis (REMO) conducted in Ghana in 2009, parasitological evaluation via skin snipping and determination of microfilarial prevalence, and entomological evaluations (according to unpublished results of the Ghana onchocerciasis mapping exercise conducted in 2009, and the REMO report summarised in 2010). Areas with an infection prevalence in the adults above 20%, were allocated to a biannual treatment frequency considering also a buffer zone of 20 Km around these CDTI priority areas. Therefore, NTDP decisions as to whether to allocate districts to annual or biannual CDTI were not made on a priori criteria of associated costs but only based on transmission criteria. In this paper, we report the results of a study undertaken to estimate the costs associated with annual (the standard strategy) vs. biannual CDTI (the newly adopted strategy) in Ghana. We also assess some factors that may hamper the scaling up of treatment frequency at a large scale given that other countries in the region may consider switching from annual to biannual ivermectin distribution. Ethical approval for the study in Ghana was obtained from Imperial College Research Ethics Committee (ICREC) and the Ethics Review Committee (ERC) of the Ghana Health Service (GHS). The study focused on the Brong-Ahafo and Northern regions in Ghana. In the former, data were collected in the Wenchi district where CDTI takes place annually; the Pru district and the Kintampo North district, where CDTI is taking place biannually, and in the latter, data were also collected in the Kpandai district, where a mixed strategy (some communities being treated annually and others biannually) is used (Table 1). These districts were selected partly based on logistics at the time of the study, and partly because already established relationships with the GHS at the district and sub-district levels would ensure collection of accurate data via the purposely designed questionnaires (see below). Figure 1 shows the locations of the districts where the study was conducted. As stated earlier, decisions to switch to a biannual treatment frequency were based on infection and transmission criteria alone, so there were no obvious reasons why the decision to change treatment frequency would have been influenced by the local district-specific programme cost. Data were collected at various levels in the organization of the GHS. Firstly, information was gathered by conducting semi-structured interviews at the headquarters of the NTDP in Accra, and at the Regional Health Service directorates in the Brong-Ahafo region. Secondly, districts (and sub-districts where appropriate) were chosen to represent a range of geographical sizes, and population densities (Table 1). Thirdly, community drug distributors (CDDs) were interviewed in at least three communities in each district. In this study, the costs under investigation were those borne by the health care providers (such as the GHS, the major in-country partners, and the local communities). Therefore the cost of drug manufacture and transport to Ghana were excluded. Only data on the cost of CDTI were collected; costs associated with individual, clinic-based treatment with ivermectin were ignored. Data were collected on both the financial and economic costs of CDTI. Financial costs are those where a monetary transaction has taken place for the purchase of a resource. Economic costs also include, in addition to the financial costs described above, estimates of the monetary value of goods or services for which no financial transaction has taken place. Therefore, economic costs also account for the value of goods or services which could have been used for another purpose (opportunity costs). Examples of resources which have no financial costs but do have important economic costs are the ‘free’ use of building space provided by the Ghana Ministry of Health, the use of donated vehicles, and the time devoted to CDTI by unpaid CDDs. The costs associated with CDTI arise from various programmatic activities as outlined in Box 1. Data collection was organized at the national, regional, district, sub-district and community levels and involved interviewing key personnel and scrutinizing national records. Data collected at the national level included records of funds provided by non-governmental organizations (NGO) such as Sightsavers (http: //www. sightsavers. org/), and others such as APOC (managed by the World Bank and implemented by the World Health Organization) (http: //www. who. int/apoc/en/), among others. Given these multiple sources, it would have been most interesting to obtain a detailed breakdown of the relative contribution of each organization to the funding of onchocerciasis control in Ghana. Unfortunately, even at the national sampling level, it was rarely possible to separate the costs by their funding source. This, however, did not affect the study, which focused on the aggregate cost of onchocerciasis control. The costs collected were incurred in the year 2011. At each level, costs were collected according to different resource types (Box 2) using an approach based on methods described by McFarland et al. [17] and the UNAIDS guidelines for costing studies [18]. First, the total gross expenditure on a resource (per year) was calculated from national records and/or questionnaires. Second, the most appropriate person (s) to answer questions on how the resource is used for activities relating to onchocerciasis control was selected for interview. Third, the interviewee was asked to indicate what fraction of time the resource was used for onchocerciasis control over the year (this was corroborated by multiple sources where possible). Multiplication of the total gross cost and fraction of time attributable to onchocerciasis control yielded an estimate of the recurrent yearly cost for a resource (such as an employee). The cost of capital resources—goods that last for more than one year, such as cars and computers—were estimated in a similar fashion, but the gross cost was spread over the average useful lifetime of the resource (a technique known as ‘annualization’) to arrive at an average yearly cost [18]. (An annualization and discount rate of 3% was used to calculate the economic costs of capital resources [19].) The average useful lifetime of all capital goods was assumed to be five years, in line with the value estimated by McFarland et al. [17] and corroborated by study participants at the national level. However, the sensitivity of the results to this assumption was investigated by varying the average useful lifetime between 5 and 8 years [20]. The annual cost of building space was estimated as the equivalent market rental value for the space being used for the control programme [18]. The interviewee was also asked to estimate the fraction of time that the resource was used for itemized onchocerciasis control programmatic activities (Box 1). In addition, in districts receiving ivermectin biannually, the interviewees were asked to describe how their time spent on different CDTI activities had changed since increasing the treatment frequency to twice per year, and to indicate which of the CDTI activities are repeated for both treatment rounds. At each level, and where relevant, interviewees were given the opportunity to express whether they had encountered any specific difficulties with the increasing of treatment frequency. Costs collected at the national and regional levels, were factored down and costs from the sub-district and community levels factored up, with the aim of arriving at a value for the cost per person treated per year in each district (Figure 2). This is described for each of the levels below. Figure 3 depicts the cost of onchocerciasis control by CDTI disaggregated by resource type in the four sampled districts. The largest proportion of the total cost was associated with the payment of personnel. Recurrent transportation costs, such as the costs of fuel and vehicle maintenance, were the next most costly resource and showed the most variation among districts. Figure 4 depicts the cost of CDTI-based onchocerciasis control disaggregated by programmatic activity in the four sampled districts. Surveillance and evaluation incurred the highest cost, followed by the drug distribution chain. For Pru and Kintampo North districts, the data show a noticeable increase in the reporting cost compared to Wenchi district. From the pooled community data, it was estimated that there is one CDD for every 390 people and they spend an average of 61 hours distributing ivermectin each treatment round. The above value was used with data on the number treated in each district (Table 1) to estimate the total amount of time CDDs spend distributing the drug across the whole district. This increased the economic cost by USD 0. 046 per person per year when treating annually, and by USD 0. 092 when treating biannually (Table 2). This result was robust to the assumed daily wage of a hired farmland worker, which when increased or decreased by GHC 1. 00, only changed the economic cost of CDD per treatment by plus or minus USD 0. 012. The CDDs reported receiving an average equivalent of USD 3. 17 in compensation for attending the distribution training sessions (which are conducted before each treatment round), and between USD 3. 17 and USD 9. 52 after distributing the drug. In this analysis, it was assumed that each distributor received the average (arithmetic mean) of the reported values (a total of USD 9. 96 in compensation for both training and distribution for each treatment round). The implementation of a large-scale, mass biannual ivermectin treatment strategy was reported at the district and sub-district level as being well received and perceived as sustainable in the future. However, the disease control officers at the district health centres in the sampled districts in which biannual treatment is being implemented, reported that increasing the treatment frequency to twice per year substantially increased the workload by increasing the amount of time they spent on reporting activities (the percentage of the economic cost at the district, sub-district, and community levels attributed to reporting activities increased from 6% in the districts (Wenchi) treated annually to 15% in the districts treated biannually (Pru and Kintampo North) (Figure 4) ). The costs disaggregated by resource type were consistent among the sampled districts. These data are also similar to those presented by McFarland et al. [17]. The recurrent transportation cost was notably higher in Kpandai compared with the other districts. This may in part be due to the poorer quality of roads in the area, resulting in higher vehicle maintenance and fuel costs (although many other factors, including the spread of the communities, also affect transportation costs). The costs disaggregated by programmatic activity showed slightly more variation among districts than among the different resource types. It is noteworthy that in the Pru and Kintampo North districts (and to a lesser extent in the Kpandai district), the percentage of the economic cost attributed to reporting activities at the district, sub-district, and community levels is substantially higher than that in the Wenchi district (15% in Pru and Kintampo North compared to 6% in Wenchi) (Figure 4). This was attributed to the increase in treatment frequency and is discussed in further detail in the section on Reported Obstacles Associated with Switching from Annual to Biannual CDTI. The compensation system for CDDs has recently been implemented in Ghana to cover their transport costs, to facilitate attendance of training days, and to help serve as an added incentive. The amount received by CDDs per treatment round was corroborated at the district health centres. Generally, the reported amount received by the community distributors was very consistent across communities and districts. Accounting for the volunteer CDDs' time added approximately USD 0. 05 per person per treatment round. The is consistent with the value reported by Onwujekwe et al. [25], who found that taking into account volunteer CDD time in two Nigerian communities added approximately USD 0. 07 (2011 prices) per person per treatment round (using the Nigerian minimum wage to value the volunteer CDDs' time). However, both our and the Onwujekwe et al. [25] estimates are substantially lower than that reported by McFarland et al. [17], who estimated that accounting for volunteer CDDs' time added an average of USD 0. 19 (2011 prices) per treatment round (valuing volunteer time based on the average per capita Gross National Income (GNI) for each of the three countries studied in [17], namely, Cameroon, Nigeria and Uganda). However, this estimate was highly variable between the different study sites (USD 0. 05–0. 54 (2011 prices) per treatment round). The use of different methods to value donated CDDs' time (see below) could partly explain the difference (i. e. estimation using the country' s minimum wage, or using the country' s per capita GNI). Other possible explanations include the occurrence of cultural differences affecting the time it takes to distribute the drug. As mentioned above, the method used to value the volunteer CDD' s time has marked effects on the cost output. For example, we assumed the market value of the volunteer CDD' s time to be USD 2. 36 per day (the minimum wage in Ghana of GHC 3. 73 divided by the 1. 58 exchange rate [23]) based on the wage that a farmland worker would receive (i. e. the wage received for the most common alternative occupation) [21], [30]). However, had we valued the volunteer CDDs' time using the per capita GNI method (as used by McFarland et al. [17]), this figure would have increased to USD 4. 96 per day [21], [30]. This difference may seem relatively small but when these costs are factored up to the district level, they can become substantial. Disease control officers at the district health centres reported that increasing the treatment frequency to twice per year increased substantially the amount of time they spent on reporting activities. This is consistent with the costs disaggregated by programmatic activity (Figure 4), which indicate that the time spent on reporting activities increased more than any other project activity when comparing biannual and annual treatments. This may potentially lead to delays in ivermectin being delivered to the districts, if the necessary reports for the next dispatch of drugs are not completed on time (delivery of the next batch of ivermectin being contingent on reporting). Delays in the delivery of treatment to communities not only will have administrative implications, but more importantly, transmission implications. Treating individuals every 6 months is highly important for transmission suppression, as it has been estimated that adult O. volvulus female worms start recovering from the temporary sterilising effects of ivermectin approximately between three and four months after treatment, and by six months microfilarial production has recuperated to a substantial degree [31]. Therefore, delays in treatment will permit more transmission, ultimately making the disease harder to eliminate and diminishing the benefit of treating biannually. National onchocerciasis control programmes which consider increasing CDTI frequency may need to support reporting activities at the district level and potentially at the drug donation programme level to encourage timely reporting but also to allow greater flexibility in deadlines to minimize delays in drug distribution. In Ghana, onchocerciasis control is under the remit of the NTDP and therefore different disease control programmes are often integrated. For example, onchocerciasis and lymphatic filariasis control activities are often carried out simultaneously. Potentially, this can lead to difficulties in obtaining accurate costs for a single disease intervention. In addition, this study was retrospective, and therefore, to a certain extent, the data obtained were subject to some degree of recall bias. In order to reduce the time and logistical complexity involved in collecting the cost data, our sampling strategy was not random, as we purposely visited local government offices and communities in districts where CDTI was annual, biannual, or a combination of the two. However, we were only able to obtain data in one district that implements annual treatment and one sub-district in each of the districts. Also, the selected districts may have been more accessible by road from Accra, the capital of Ghana, than other more remote locations. Nonetheless, there is no reason to assume that the costs reported for the sites included in this study (either delivering annual or biannual CDTI) are not representative of other sub-districts in the area, nor is there a treatment cost-associated reason as to why an area switched from annual to biannual CDTI other than the parasitological criteria listed above. This is confirmed by the similarity of cost estimation of annual treatment between the districts delivering only annual CDTI and the sub-districts also delivering yearly treatment within districts implementing both strategies. Due to logistic reasons, the regional level costs in the Northern region were assumed to be the same as those estimated from Brong-Ahafo region. However, due to differences between the regions (such as road networks and community scattering), the costs incurred in the Northern region may be higher. Nevertheless, this assumption will not affect the main conclusions of the study regarding the relative costs of annual vs. biannual treatment. Our estimate of the cost of annual CDTI is consistent with the range of values previously reported in the literature [17], [25], [26]. Our results indicate that the cost of biannual ivermectin treatment was approximately 50–60% higher than the cost of annual treatment, and that simply doubling the cost of annual CDTI does not yield a correct estimate as some studies have assumed [11]. This is higher than estimates for increasing treatment frequency obtained at smaller scales and when targeting specific age groups, such as those associated with school-based anthelmintic treatment programmes [29], which are not truly relevant for onchocerciasis, but similar to estimates for the more comparable lymphatic filariasis control programme [28]. Our study will be beneficial in informing economic evaluations regarding cost-effectiveness analyses of increasing CDTI frequency from annual to biannual in the African context for the control and elimination of human onchocerciasis.

The African Programme for Onchocerciasis Control (APOC) has recently been extended until 2025, with renewed commitment towards onchocerciasis elimination. This aim is aligned with the goals stated by the World Health Organization and the London Declaration on Neglected Tropical Diseases in January 2012. Switching from annual to biannual (twice yearly) ivermectin distribution might increase the feasibility of onchocerciasis elimination in some African foci. However, relatively few communities have received biannual treatments in Africa, and there are no cost data associated with increasing ivermectin treatment frequency at a large scale, essential pre-requisites to provide reliable information for evidence-based decision making regarding adoption of a biannual treatment strategy. Therefore, we undertook a study to estimate costs associated with biannual compared to annual ivermectin delivery in Ghana, which since 2009 has implemented a biannual treatment strategy in selected priority areas. Our results indicate that the cost of biannual ivermectin treatment per year is approximately 60% higher than the cost of annual treatment. This study provides tangible evidence of the different costs associated with annual and biannual ivermectin treatment, which can be used to inform economic evaluations and policy decisions regarding the optimal treatment frequency required to eliminate onchocerciasis in Africa.

Abstract Introduction Methods Results Discussion

2013

The Cost of Annual versus Biannual Community-Directed Treatment of Onchocerciasis with Ivermectin: Ghana as a Case Study

The modulation of fitness by single mutational substitutions during environmental change is the most fundamental consequence of natural selection. The antagonistic tradeoffs of pleiotropic mutations that can be selected under changing environments therefore lie at the foundation of evolutionary biology. However, the molecular basis of fitness tradeoffs is rarely determined in terms of how these pleiotropic mutations affect protein structure. Here we use an interdisciplinary approach to study how antagonistic pleiotropy and protein function dictate a fitness tradeoff. We challenged populations of an RNA virus, bacteriophage Φ6, to evolve in a novel temperature environment where heat shock imposed extreme virus mortality. A single amino acid substitution in the viral lysin protein P5 (V207F) favored improved stability, and hence survival of challenged viruses, despite a concomitant tradeoff that decreased viral reproduction. This mutation increased the thermostability of P5. Crystal structures of wild-type, mutant, and ligand-bound P5 reveal the molecular basis of this thermostabilization—the Phe207 side chain fills a hydrophobic cavity that is unoccupied in the wild-type—and identify P5 as a lytic transglycosylase. The mutation did not reduce the enzymatic activity of P5, suggesting that the reproduction tradeoff stems from other factors such as inefficient capsid assembly or disassembly. Our study demonstrates how combining experimental evolution, biochemistry, and structural biology can identify the mechanisms that drive the antagonistic pleiotropic phenotypes of an individual point mutation in the classic evolutionary tug-of-war between survival and reproduction. The ability of a single mutational substitution to modulate fitness across environments is the most important consequence of natural selection under environmental change. Understanding the antagonistic tradeoffs of pleiotropic mutations that promote survival in changing environments is therefore essential for a complete understanding of evolution. However, the molecular basis of fitness tradeoffs caused by pleiotropic mutations is rarely determined in terms of how the mutations affect protein structure. Perhaps the main reason for this intellectual gap is because the fields of structural biology and experimental evolution do not often intersect. Structural studies tend to focus on proximate explanations for protein function stemming directly from structural features, without determining the ultimate consequences of evolved protein changes for fitness across environments at the system level. In contrast, experimental evolution studies have identified that point mutations can be consequential for determining fitness tradeoffs in independently evolving populations facing the same environmental change [1], [2], without elucidating the structural details of how such trade-offs are mediated by functional changes at the protein level. It has been argued that interdisciplinary approaches are necessary for the ‘functional synthesis’ that will advance our understanding of evolutionary biology [3], [4], especially to reveal the mechanistic details of evolutionary novelty and adaptive constraint; however, the necessary mergers between disciplines remain rare [5], [6], [7]. Perhaps the most fundamentally important tradeoff in evolutionary biology is that between survival and reproduction, the cornerstones of evolution by natural selection [8]. It is often assumed that natural selection is driven by genetic changes that promote relative differences in offspring production, or reproduction in close relatives [9]. However, the need for organisms to survive in the face of depleted resources or environmental stressors can be of equal or greater importance for dictating relative differences in fitness. It is evident that the functional properties of proteins could bridge tradeoffs in survival versus reproduction, because the genetic changes underlying a protein may simultaneously affect its stability (survival) as well as operational (reproductive) properties across environments. Thus, adaptive evolution in a changing environment provides a key context for studying how protein changes might mediate the interplay of survival versus reproduction, and for determining which variants are favored to evolve under natural selection. ‘Life-history’ tradeoffs between survival and reproduction have been invoked in the adaptive evolution in a variety of organisms [10], but these examples often hinge on statistical correlations between traits, without attempting to identify the molecular basis of changes in protein function that cause such tradeoffs to arise. Here we challenged populations of an RNA virus, bacteriophage Φ6 of the cystovirus genus [11], to evolve in a novel temperature environment where heat shock imposed extreme virus mortality. A single amino acid substitution in the viral lysin protein P5 favored improved stability (and hence, survival) of challenged viruses, despite a concomitant tradeoff that decreased viral reproduction. Lysins are lytic proteins that locally degrade the cell wall, either to provide access to the inner bacterial membrane during infection or to release virus progeny by cell lysis [12], [13]. An electron microscopy image reconstruction of the bacteriophage Φ12, a cystovirus related to Φ6, suggests that Φ12 P5 is part of the icosahedral nucleocapsid shell of Φ12 and that P5 may interact with the lipid membrane, which constitutes the outer virus layer [14]. It remains to be confirmed whether P5 has an analogous location in Φ6. Nevertheless, the selected mutation in the Φ6 P5 gene increased the thermostability of P5. Crystal structures of the wildtype and mutant P5 reveal the molecular basis of this thermostabilization and identify P5 as a lytic transglycosylase. We show that loss of P5 enzymatic activity is not the source of the viral reproduction tradeoff, which may instead result from inefficient capsid assembly or disassembly. Our study demonstrates how a combination of experimental evolution and biophysical approaches can be used to discover the mechanistic details that drive antagonistic pleiotropic effects of an individual point mutation in the classic evolutionary tug-of-war between survival and reproduction. The P. phaseolicola host bacteria for phage Φ6 cannot grow and survive at temperatures greater than 30°C. In contrast, virions of wildtype phage Φ6 can withstand exposure to temperatures between 30°C and 40°C, but suffer ‘high mortality’ (subsequent inability to productively infect cells) when subjected to 5 min heat shock at temperatures ranging between 40°C and 50°C [15]. Three populations of wildtype phage Φ6 were evolved independently for 20 days (100 generations) under selection involving 5 min heat shock at 50°C every fifth generation; three control populations were evolved identically, but experienced periodic ‘mock’ heat shocks of 25°C (Figure S1). Subsequently, we conducted repeated (n = 3) survival assays at 42. 5°C, 45°C, 47. 5°C and 50°C for each of the endpoint treatment and control populations. Each treatment population improved in survival in the 50°C selective environment, relative to wildtype phage Φ6 (independent samples t-tests with 5 df, P<0. 0001). Moreover, the treatment populations were significantly advantaged in survival at temperatures above 45°C relative to the controls, which did not differ in thermotolerance from the ancestor [15] (Figure 1A). These differing thermotolerance ‘reaction norms’ (Figure 1A) clearly demonstrated that evolutionary history affected the evolved ability for treatment versus control populations to withstand elevated temperatures. To examine whether particular molecular substitutions were associated with the differing thermotolerance phenotypes of the evolved populations (Figure 1A), we obtained the consensus genome sequence for each evolved population. Each evolved lineage differed from the wildtype ancestor by 1 to 6 substitutions; overall we observed 18 substitutions at 9 sites across the 3 genomic segments (Table S1). Interestingly, all three treatment populations showed an identical non-synonymous mutation (G2238T) that was the only molecular change on the small RNA segment, and which was not present in the controls. These data strongly suggested that the mutation was somehow beneficial for adaptation to withstand the 50°C heat shock environment, because it arose spontaneously and fixed in all of the independently evolved treatment (but not control) populations. The G2238T mutation corresponds to a V207F amino-acid substitution in the gene for the lysin protein P5, which locally degrades the cell wall, either to provide access to the inner bacterial membrane during infection or to release virus progeny by cell lysis [12], [13]. Hereafter, we refer to the wildtype protein as P5wt and the mutant protein as P5V207F. Laboratory culture of phage Φ6 on agar always occurs at 25°C, because this incubation temperature allows the P. phaseolicola host bacteria to produce a confluent lawn, which supports robust plaque formation of the virus. Intriguingly, we observed that the viruses from the treatment populations showed a novel plaquing phenotype, which is generally referred to as a ‘bull' s-eye’ plaque morphology (Figure 1B). Whereas plaques formed by the wildtype virus were clear, the bull' s-eye plaques appeared turbid due to residual bacterial growth within the plaque, indicating that mutant viruses were less efficient at killing bacteria at the ordinary growth temperature of 25°C. To confirm that the P5V207F mutation was antagonistically pleiotropic (i. e. , caused improved extracellular survival at 50°C but reduced intracellular growth at 25°C), we isolated viruses containing only the P5V207F mutation (Figure 1C). To do so, we conducted a classic genetic cross [1], [2], where an evolved strain bearing only this mutation on the small segment was ‘back-crossed’ with its wildtype Φ6 ancestor, to obtain a hybrid reassortant with a mutated small segment and the ancestral medium and large segments; genotype of the hybrid was confirmed through sequencing. Plaque assays at 25°C showed that the hybrid produced the same bull' s-eye plaque morphology that was characteristic of the treatment populations (Figure 1D). In addition, survival assays showed that the thermotolerance reaction norm for the hybrid was qualitatively similar to data observed in the evolved treatment populations (Figure 1A, 1C); at all elevated temperatures survival of the P5V207F mutant significantly exceeded that of the wildtype. Although percent survival of both strains was modest at the extreme 50°C temperature, survival of P5V207F (0. 355±0. 241 std. dev.) was still greater than the wildtype (0. 013±0. 007 std. dev.) (t-test with t = 3. 42, df = 14, P = 0. 004; Figure 1C). These results indicated that the P5V207F substitution caused both the unique bull' s-eye plaque morphology when grown at 25°C, as well as the improved extracellular survival at elevated temperatures. Because the bull' s-eye plaque morphology of the V207F mutant suggested that this genotype less efficiently killed bacteria (relative to the wildtype) at 25°C, we hypothesized that the mutation was deleterious for growth at 25°C. To examine this idea, we conducted paired-growth assays at 25°C, which measured reproduction of the hybrid strain and of the wildtype ancestor under benign conditions, relative to a genetically-marked common competitor virus; (hybrid: n = 28 replicates, wildtype: n = 18 replicates). After adjusting for the cost of the genetic marker on the common competitor, results showed that the mean log reproductive fitness of the mutant relative to the wildtype was −0. 341 (±0. 324 std. dev. ; Figure 1E). Reproductive fitness of the hybrid was significantly less than that of the wildtype based on a two-tailed t-test (with t = 2. 878, df = 44, P = 0. 006). Together, we observed that the P5V207F mutation caused a ∼1. 5-fold decrease in reproduction at 25°C (Figure 1E), but a ∼27-fold increase in extracellular survival at 50°C (Figure 1C). Results of repeated (n = 5) fitness assays modified to match the treatment conditions (i. e. , imposing 50°C heat shock prior to growth on agar at 25°C; Figure S1) also demonstrated the net positive effect of the V207F mutation in the selective environment: mean log fitness of the mutant relative to the wildtype was 3. 920 (±0. 280 std. err.), which significantly exceeded ancestral log fitness of zero (t-test with t = 13. 98, df = 5, P = 0. 0002). We conclude that the P5V207F mutation is an antagonistically pleiotropic allele that produces a survival/reproduction tradeoff across the two portions of the selective environment experienced by the treatment populations; although the mutation is beneficial for extracellular survival under the brief 50°C heat shock, it is deleterious for reproduction occurring at 25°C. The evolved thermotolerance of the viruses carrying the P5V207F mutation could result from an increase in the inherent thermostability of P5, or an increase in the stability of a protein-protein interaction directly or indirectly involving P5. To determine how the V207F mutation in protein P5 induced thermotolerance, we first determined the effect of the mutation on the thermostability of P5. We purified recombinant P5wt and P5V207F from Escherichia coli. Thermal melting curves of the two proteins were measured by circular dichroism (CD) spectrometry and differential scanning calorimetry (DSC). P5wt and P5V207F both began to unfold cooperatively as the temperature reached 50°C and 55°C, respectively, as judged from the sharp loss of CD signal from α-helical secondary structure at 220 nm (Figure 2A). Melting temperatures for P5wt and P5V207F calculated from the CD melting curves were 55. 3°C and 62. 9°C, respectively (Figure 2A). Similarly, the melting temperatures of P5wt and P5V207F determined by DSC were 52. 6°C and 58. 3°C, respectively (Figure 2). These data indicate that the V207F mutation increases the melting temperature of P5 by between 5. 7–7. 6°C. Additionally, the heights and areas of the DSC curves were greater for the V207F mutants than for their wildtype counterparts. Together, these data indicate that the higher thermotolerance of the mutant virus is due to increased thermostability of P5V207F relative to P5wt. Since melting temperature and DSC peak area depend on the change in entropy and enthalpy, respectively, we conclude that the additional free energy of stabilization from the V207F mutation (ΔG) derives from both the entropic term (TΔS) and the enthalpy (ΔH) in the free energy equation (ΔG = ΔH−TΔS). The P5 proteins eluted from a size-exclusion column as 40 kDa proteins despite a calculated molecular mass of 25 kDa. We concluded that P5 was either an elongated monomer or a compact monomer with disordered regions that increased the hydrodynamic radius of the protein. Φ6 P5 was previously reported to be a monomer in solution [16]. To determine whether P5 contained regions of disorder or internal flexibility, we subjected P5wt and P5V207F to limited proteolysis with various proteases. Treatment with V8 protease (Staphylococcus aureus endoproteinase Glu-C) produced three fragments, which were identified by mass spectroscopy as consisting of residues 1–39,40–47 and 48–220, respectively (Figure 2B). We will refer to the largest fragment, residues 48–220, as P5ΔV8. Additional proteolytic products were observed for P5wt but not for P5V207F (Figure 2B), revealing a higher overall protease sensitivity of the wildtype protein. P5ΔV8 eluted from a size-exclusion column as would be expected for a globular 19 kDa protein. To investigate the state of residues 1–47 further we calculated the difference CD signal between P5wt and P5ΔV8wt (Figure S2). The minimum at 190 nm indicates that residues 1–47 have essentially no secondary structure. The melting temperatures of P5ΔV8wt and P5ΔV8V207F were similar to the full-length proteins as determined by CD (56. 6°C and 61. 0°C) and DSC (54. 1°C and 59. 4°C; Figure 2C, 2D). CD melting curves of the truncated proteins also showed greater stability at temperatures below 50°C than the full-length proteins, suggesting that residues 1–47 are responsible for the observed non-cooperative unfolding of the full-length proteins in this temperature range (Figure 2A, 2C). Together, these data indicate that the first 47 residues of P5 are either disordered in solution or fold separately from the rest of the protein. In the virion, however, residues 1–47 may adopt a stable conformation upon binding other viral components. To understand the molecular basis of the thermostabilization of P5 by the V207F mutation we determined the crystal structures of P5ΔV8wt and P5ΔV8V207F at 1. 4 Å resolution (we could not obtain crystals of P5wt or P5V207F). Crystallographic data collection and refinement statistics are provided in Table S2. P5ΔV8 adopts a lysozyme superfamily fold consisting of an N-terminal lobe (NTL) and a C-terminal lobe (CTL) connected by a central helix (Figure 3A). The two lobes create a substrate binding cleft containing the predicted catalytic residue Glu95 (Figure 3A) [17]. The central helix and the CTL of P5ΔV8 have similar conformations as various other lysozyme structures. However, the NTL in P5 differs from other lysozyme structures in its secondary structure content and relative orientation to the CTL (Figure S3). We note that P5ΔV8 forms an unusual crystal-packing interaction in which three N-terminal residues (residues 48–50) of one of the two subunits in the asymmetric unit insert into a shallow groove in a molecule in the adjacent asymmetric unit (Figure S3F). Given that residues 48–50 make specific crystal contacts it was surprising that residues 53–59 were disordered and only residues 48–52 and 60–220 could be modeled. The structures of P5ΔV8wt and P5ΔV8V207F are almost identical, with a root mean square deviation (rmsd) of the Cα positions of 0. 135 Å (Figure 3B, Figure S3G). Residue 207 is located on helix α8, facing the hydrophobic core of the protein. In P5ΔV8wt, V207 and adjacent residues 153,176,179–180,194,199, and 203–204 create a small (30. 28 Å3) cavity lined with hydrophobic side chains (Figure 3B, Figure S4). The lack of electron density within the cavity suggests that it is unoccupied. The larger phenylalanine side chain in P5ΔV8V207F neatly fills the cavity, with no significant changes in the protein. The B-factors near residue 207 are similar in both structures (15–16 Å2). The only significant side chain shift observed in P5ΔV8V207F is an adjustment in the N179 side chain to accommodate the F207 side chain, resulting in a van der Waals interaction between these two side chains. By filling an unoccupied hydrophobic cavity, the V207F substitution increases the buried hydrophobic surface area in the protein. This provides a likely explanation for the increased thermostability of the mutant, because the hydrophobic surface area that is buried within a folded protein contributes directly to its free energy of stabilization [18], [19]. In support of this explanation, the hydrophobic cavity created by the replacement L99A in T4 lysozyme was large enough to bind benzene, and binding of benzene to the L99A mutant increased the melting temperature of T4 lysozyme by 6. 0°C [20]. The structure of Φ6 P5 bears the closest resemblance to G-type lysozymes and lytic transglycosylases (LTs) such as gp144 of phage ΦKZ [13] and the catalytic domain of E. coli slt70 with rmsds of 1. 2 Å and 1. 6 Å, respectively (Figure S3A–S3C). Like G-type lysozymes and LTs [13], [21], [22], [23], Φ6 P5 has just one glutamate, Glu95, in its predicted active site. To gain additional insight into the enzymatic activity of P5, we determined the structure of P5ΔV8wt in complex with the substrate analog chitotetraose (NAG4), at 1. 23 Å resolution. Chitotetraose is the most similar commercially available ligand to the natural substrates of LTs and ΦKZ gp144 binds chitotetraose [13]. The structure of ligand-bound P5ΔV8 is similar to that of the apo-enzyme with a few exceptions (Figure 4A, Figure S3G). The major difference is that residues 199–220 are missing from the ligand-bound structure. This region spans the last α-helix (α8) and includes the site of the selected V207F mutation. Helix α7 and the following linker are also in slightly different positions in the ligand-bound structure (Figure S3G). Additionally, the ligand displaces the side chain of Tyr196 out of the substrate-binding site, where the side chain is located in the apo-P5 structure (Figure 4). The ligand-bound P5 crystals belonged to a different space group with different crystal packing than the apo-P5 crystals. The substrate binding cleft is solvent-exposed in the ligand-bound crystals but mostly buried by non-crystallographic symmetry contacts in the apo-P5 crystals. Because the crystals of ligand-bound P5 took twice as long to grow as the apo-P5 crystals, we speculate that the lack of electron density for residues 199–220 in the ligand-bound structure is due to proteolytic cleavage by residual V8 protease or another contaminating protease in the P5 preparation. This cleavage may be favored by the increased solvent exposure of the C-terminal region in the ligand-bound crystals. The substrate-binding sites of LTs and lysozymes have six sugar binding subsites (A–F) and substrate cleavage occurs between an N-acetyl muramic acid (NAM) bound in subsite D and a N-acetyl glucosamine (NAG) bound in subsite E [24], [25], [26]. In the ligand-bound P5 structure, the four NAG residues of the ligand bind to subsites A–D (Figure 4A). A water molecule is observed between Glu95 and the NAG in subsite D (Figure 4B), supporting a role of Glu95 as the acid/base in the catalytic mechanism. The ligand binds in the same manner as the identical chitotetraose ligand in the structure of the ΦKZ gp144 LT [13]. In both the P5 and gp144 structures, the mode of chitotetraose binding is consistent with the presence of the additional lactic acid and peptidyl moieties that are present in the NAM residues of natural substrates at subsites B and D. Moreover, a superposition of ligand-bound P5 onto the structure of E. coli slt70 LT bound to a glycan product in subsites E and F shows that the geometry and electrostatics of the P5 surface should allow binding of such a product in the same manner (Figure 4C, 4D). Together, the structural features of the P5 active site and glycan binding site and their similarity to the ΦKZ and slt70 LTs strongly support that P5 is a lytic transglycosylase, as previously predicted [17]. To confirm that Φ6 P5 could lyse bacterial cell walls, we examined the cell lysis activity of P5wt and P5V207F using a turbidity assay at the normal growth temperature (25°C). Both proteins lysed cells with the same efficiency (Figure 5A). The maximal rates of cell lysis of P5wt and P5V207F were similar and were directly proportional to the enzyme concentration (Figure 5B). Merging experimental evolution and structural biology, we used RNA phage Φ6 as a model to demonstrate how a single antagonistically pleiotropic mutation caused a survival/reproduction tradeoff in evolving populations. The V207F mutation conferred better survival of viruses under 50°C heat-shock, despite reducing their reproduction at 25°C; these data demonstrated that thermotolerance was the most important fitness component for dictating the overall evolutionary success of the treatment populations. We are aware of few other studies that have examined survival/reproduction tradeoffs in viruses. de Paepe and Taddei compared lytic phages of E. coli and suggested a survival/reproduction tradeoff mediated by capsid structure [27]. The proposed mechanism was that denser-packaging of viral DNA within capsids affords increased stability, but tends to slow the rate of phage genome replication. Moreover, in experimental evolution studies where E. coli bacteria and phage ΦX174 or the related phage ID11 were exposed to elevated temperatures, results showed that changes in viral capsid proteins were likely stabilizing [28], [29]. Our results differed from these previous studies because we found the tradeoff was governed by an enzyme, rather than changes to the phage capsid. Also, we observed strong convergence evidenced by a single mutation that fixed in the independently evolved treatment populations, whereas the experiment with phage ID11 showed multiple possible first-step substitutions [28], [29]. One possibility is that our heat-shock regime selected strongly for structural stability, a single fitness component, whereas the phage ID11 study required growth of phage and bacteria at high temperature, thus selecting on multiple fitness components allowing various beneficial mutations to fix. Apparently, our selective regime was so stringent that the convergent mutation fixed despite strong antagonism for growth at ordinary temperature, which constituted environmental conditions aside from the 5 min heat shock. Further research could explore how this tradeoff may be lessened (or even eliminated) via further molecular change (s) resulting from fixation of mutations that compensate for the growth deficit. A previous report suggested that the Φ6 P5 lysin might have endopeptidase activity rather than the glycanhydrolase activity of lysozyme [16]. However, the cell lysis activity assay used in that study cannot distinguish between cell wall lysis due to endopeptidase activity from cell wall lysis due to glycosidase activity. Conversely, a bioinformatics study by Pei and Grishin classified Φ6 P5 as a distant relative of the lytic transglycosylase (LT) subfamily within the lysozyme superfamily [17]. LTs are enzymes that can degrade the peptidoglycan layer of the bacterial cell wall by cleaving a β (1,4) -glycosidic bond between NAM and NAG residues and forming a new glycosidic bond between the O6 and C1 atoms of the NAM residue [30], [31]. In general, lysozymes possess a catalytic dyad of glutamic and aspartic acid residues that catalyze the hydrolysis of the same substrate by using a water molecule from the solvent [32], [33], [34], [35]. However, in LTs, there is only one acidic residue, typically a glutamic acid, in the vicinity of the substrate cleavage site [36]. Φ6 P5 adopts a lysozyme superfamily fold and our structure of P5 bound to a tetrasaccharide substrate analog shows that the enzyme contains a single glutamate (Glu95) in its active site. Together, our structural data unambiguously identify Φ6 P5 as an LT. The lysozyme superfamily has been extensively studied as a model system for protein folding and stability, as reviewed in ref. [37]. Several hydrophobic cavities have been identified in lysozymes and it has been proposed that mutations filling these cavities should stabilize the protein by increasing the hydrophobic surface area buried within the fold [38]. However, attempts to fill two different hydrophobic cavities in T4 lysozyme by site-directed mutagenesis resulted unexpectedly in a slight decrease in the melting point of the protein, because stabilization from the increased hydrophobic contacts was offset by strain within the mutant side chains [38]. In contrast, the naturally selected V207F mutation in Φ6 P5 achieves the stabilization expected from the addition of four carbon atoms with a surface area of 35–40 Å2, ∼1 kcal/mol (or a 4°C increase in melting temperature), without the introduction of strain within the protein. Consistent with the average increase in the melting temperature of P5V207F of 5. 7°C reported here, binding of benzene to the cavity created by the L99A substitution in T4 lysozyme resulted in a 6. 0°C increase in the melting temperature of T4 lysozyme [20]. We conclude that the enhanced thermal stability of P5V207F is responsible for the survival of the mutant phages after the heat-shock challenge. The V207F mutation did not affect the structure of the P5 active site, allowing the mutant enzyme to fully maintain its functional role as a lysin, which is essential in the viral lifecycle [12], [13]. Thus, it is not clear why the mutation adversely affected viral reproduction. We speculate that the mutation may reduce the structural plasticity of P5, and may hence reduce the efficiency of the assembly or disassembly of the viral capsid. Indeed, the architecture of quasiequivalent icosahedral viral capsids, such as that of Φ6, necessarily depends on structural plasticity within the capsid proteins to form the contacts that hold the capsid together in multiple nonequivalent environments. In support of this hypothesis, an electron cryomicroscopy structure of phage Φ12 (a cystovirus closely related to Φ6) at 10 Å resolution suggests that P5 in an integral part of the viral capsid [14]. Moreover, the weak electron density for P5 in the Φ12 structure suggests that P5 has some flexibility relative to the rest of the capsid, and this flexibility of P5 has been proposed to allow other proteins to access the capsid during virus replication [14]. The selected V207F mutation may therefore impair viral replication, and hence reproduction, by reducing the flexibility of P5. Although it is widely recognized that infectious viruses can differ markedly in terms of their stability in the face of environmental stress, the associated effects of individual viral proteins remain largely unexplored. Medically and agriculturally important viruses sometimes show an inherent tendency to survive for extended periods outside of their hosts, suggesting that survivability should factor heavily in the relative transmission success of virus genotypes. This is likely to be true for variants of viruses that are transmitted between hosts via inert objects such as transmission of Hepatitis C Viruses between injection-drug users that share needle syringes [39]; better studied examples include differing ability of Influenza A Virus genotypes to withstand exposure to cold water when undergoing fecal-oral transmission in avian hosts [40]. Fever is generally assumed to be a useful innate defense against infecting viruses because these pathogens can degrade when exposed to elevated temperature; although this assumption is critical to the current debate of whether fever-reducing drugs ultimately harm or benefit infected hosts, it is perhaps surprising that evolution of temperature tolerance in viruses is seldom studied [28], [29], [41]. Our study indicated that evolved thermotolerance is rapidly acquired in RNA viruses selected under temperatures much higher than those they normally encounter, strongly suggesting that simple solutions (i. e. , point mutations) may govern this adaptation in other virus systems. Cultures of Pseudomonas syringae pathovar phaseolicola (ATCC #21781) host bacteria were initiated by a single colony grown at 25°C in LC medium: Luria-Bertani broth at pH 7. 5. Phage were grown by mixing ∼100 particles with 200 µl of overnight bacterial culture in 3 ml 0. 7% LC top agar, overlaid on a 1. 5% LC agar plate. After 24 h, phage lysates were prepared by harvesting viral plaques into LC broth, followed by centrifugation and filtration to remove bacteria. Viral stocks were stored at −20°C in 2∶3 glycerol/LC (v/v). Bacterial stocks were stored in 2∶3 glycerol/LC (v/v) at −80°C. Clones of wildtype Φ6 (strain #PT522) were used to found three treatment and three control populations. Treatment populations were incubated in the absence of cells for 5 min at 50°C followed by 24 h of growth (5 virus generations) on a lawn of P. phaseolicola at 25°C. Viral progeny were harvested as described above. This process was repeated for a total of 20 passages (100 generations [42]) while monitoring the bottleneck population size of evolving lineages to ensure that they experienced equal generation numbers. Control populations were maintained identically but experienced periodic mock heat shocks at 25°C. Relative reproduction of virus strains was estimated in paired-growth assays as described [43]. Reproduction was gauged relative to common competitor phage: wildtype phage Φ6 containing an engineered mutation (fragment of the Escherichia coli lacZ gene for beta-galactosidase) on segment L [15]. We mixed the test phage and marked competitor at a 1∶1 volumetric ratio, and then plated a dilution of this mixture containing ∼200 viruses onto a host lawn of bacteria. After 24 h incubation, the resulting plaques were harvested and filtered to obtain a cell free lysate. We tracked the ratio of test virus to marked competitor in the starting mixture (R0) and in the harvested lysate (R1) by plating on lawns of LM1034: P. phaseolicola containing the complementing fragment of the E. coli lacZ gene. LM1034 allows the marked competitor to produce blue plaques on agar containing X-gal (0. 4% w/v), whereas unmarked phage produce colorless plaques. We defined reproductive fitness (W) as the relative change in ratios, W = R1/R0. After log-transforming fitness estimates, mean log fitness of the wildtype strain was calculated and this value was subtracted from all fitness estimates to adjust for cost of the genetic marker on the common competitor. Fitness assays conducted under the treatment conditions used in experimental evolution (Figure S1) incorporated 5 min heat shock at 50°C. Plating a sample of the starting mixture onto a bacterial lawn confirmed the 1∶1 initial ratio (R0). An additional sample of the mixture was subjected to 5 min heat shock at 50°C, followed by plating on a lawn for plaque growth at 25°C; resulting plaques were harvested and titered to estimate the final ratio (R1). These data were analyzed as above, to estimate fitness in the treatment environment. Survival was assayed as described [15]; 120 µl of a virus lysate containing ∼108 particles was diluted onto a P. phaseolicola lawn to confirm the initial virus titer (Ni). The lysate was then heated for 5 min and the final titer (Nf) was measured. Percent survival equaled (Nf/Ni) * 100. Thus, survival under heat shock was gauged by tracking pfu viable for growth at 25°C. Genomic RNA was extracted (QiaAMP viral RNA extraction kit; Qiagen) and converted to cDNA by RT-PCR with Superscript polymerase and random hexamer primers (Invitrogen). Standard PCR methods were used to amplify 93. 2% of the genome excluding the single-stranded ends of each segment [2]. PCR products were purified for sequencing with ExoSAP-It (US Biological). Virus genomes were sequenced with double coverage of every nucleotide. Sequences were analyzed with CLC DNA Workbench 6 (www. clcbio. com). Genes encoding P5wt and P5V207F were cloned into the pET-28 vector (Novagen) in frame with an N-terminal six-histidine tag followed by a tobacco etch virus (TEV) protease cleavage site. P5wt and P5V207F were expressed in E. coli Rosetta (DE3) cells and (Novagen) purified by nickel-affinity and size-exclusion chromatography. The histidine tag was removed with 1∶100 (w/w) TEV protease (12 h at 16°C). Uncleaved P5 and TEV protease were removed with nickel-agarose beads. The proteins were stored at −80°C in 10 mM Tris pH 8,0. 1 M NaCl. P5ΔV8wt and P5ΔV8V207F were prepared by treating purified P5wt and P5V207F with 200∶1 (w/w) S. aureus V8 protease (Worthington) for 3 h. The protease was inactivated with 3 mM PMSF and Complete protease inhibitors (Roche). The P5ΔV8 proteins were then purified by size-exclusion chromatography. V8 (Glu-C) protease (Worthington) was added to 0. 3 g/l of P5wt or P5V207F to a molar ratio of 200∶1 P5∶V8, incubated on ice for 0. 5–18 h and heat inactivated (95°C, 5 min in SDS-PAGE loading buffer). Proteolytic products were purified by reverse phase chromatography with a C4 column (Vydac) in 0. 05% trifluoroacetic acid using a 10–80% acetonitrile gradient. Peak elution fractions were analyzed by MALDI mass spectroscopy at the Yale Chemical Instrumentation Center. Circular dichroism measurements were performed on an Aviv 202 spectrometer using 1mm path length cell. Protein samples were diluted to 0. 3 g/l in 5 mM sodium phosphate pH 8 to give a reading of approximately −30 millidegrees at 220 nm. For melting curves the temperature was increased from 4°C to 95°C in 1° increments. Readings were taken every degree and were averaged over 3 s after 3 min of temperature equilibration. Spectra were measured between 180 nm and 260 nm at scan rate of 1 nm/s. P5wt and P5ΔV8wt concentration was 5. 5 µM and 4 µM, respectively. The raw data were corrected by subtracting the contribution of the buffer to CD signal. Data were smoothed and converted to molar ellipticity. The measurements were taken at a constant temperature of 16°C. The signal of residues 1–47 was calculated by subtracting the signal of P5ΔV8wt from that of P5wt after correcting for concentration and number of amino acid residues in terms of molar ellipticity. For differential scanning calorimetry, P5 proteins were diluted to 20 µM in 10 mM Tris pH 8,0. 2 M NaCl and subjected to thermal scans from 10°C to 100°C at a rate of 60°C/h in a MicroCal VP calorimeter with a 15 min pre-equilibration time. Protein-free buffer was used as the reference. Data were collected in triplicate and analyzed with Origin 7 (OriginLab). Crystals were grown by vapor diffusion at 16°C. P5ΔV8 at 7 g/l in 10 mM Tris pH 8,0. 1 M NaCl was mixed with a half-volume of reservoir solution (1. 6 M sodium acetate, 0. 1 M sodium citrate pH 6. 5). Crystals were frozen in mother liquor. For the ligand-bound structure, P5ΔV8wt was co-crystallized with a 10-fold molar excess of chitotetraose (Sigma) added 3 h prior to mixing with a half-volume of reservoir solution (15% PEG 3350,0. 2 M KNO3 pH 6. 9). Two rounds of streak seeding into pre-equilibrated drops of reservoir solution were required to obtain single ligand-bound crystals, which were frozen in reservoir solution plus 25% (v/v) glycerol. P5ΔV8wt crystals were derivatized by soaking in reservoir solution plus 0. 2 M NaI for 45 s followed by freezing at 100 K. The structure was determined by single-wavelength anomalous diffraction with HKL2MAP [44]. The atomic model was built with ARP/wARP [45] and refined with Coot [46] and PHENIX [47]. The structure of ligand-bound P5ΔV8wt was determined by molecular replacement with PHENIX using P5ΔV8wt as the search model. Cavities were identified and analyzed with VOIDOO [48] using a 1. 1 Å probe radius. See Table S2 for data collection and refinement statistics. Atomic coordinates and structure factors for P5ΔV8wt, P5ΔV8V207F and ligand-bound P5ΔV8wt were deposited in the Protein Data Bank (ID codes 4DQ5,4DQ7 and 4DQJ). To assay the cell lysis activities of P5wt and P5V207F, the decrease in turbidity of a chloroform-treated E. coli culture was measured as described [16] by tracking absorbance at 450 nm and 25°C for 20 min after addition of 10–50 ng of protein. For additional details, see Extended Materials and Methods (Text S1).

The most fundamental mechanism of natural selection in a changing environment is the modulation of fitness by mutations. It is the tradeoffs offered by these mutations that drive evolution. However, fitness tradeoffs are rarely understood at the molecular level, in terms of how the selected mutations affect protein structure and function. Here, we merge experimental evolution and structural biology to study the fundamental tradeoff between survival and reproduction. We challenged populations of an RNA virus to evolve in a novel temperature environment where heat shock imposed extreme virus mortality. A single mutation in a specific viral protein increased the stability, and hence survival of challenged viruses, despite a concomitant tradeoff that decreased viral reproduction. This mutation increased the thermal stability of the mutant protein. Atomic structures of the wild-type and mutant protein reveal the molecular basis of this stabilization. The mutation did not reduce the enzymatic activity of the protein, suggesting that the reproduction tradeoff stems from other factors, such as inefficient virus assembly or disassembly. Our study uncovers the mechanism that drives the antagonistic effects of an individual point mutation in the classic evolutionary tug-of-war between survival and reproduction.

Abstract Introduction Results Discussion Materials and Methods

biomacromolecule-ligand interactions organismal evolution microbial mutation enzymes macromolecular assemblies population genetics evolutionary selection microbiology mutation mechanisms of resistance and susceptibility biocatalysis protein folding microbial evolution protein structure mucolytic enzymes enzyme classes proteins structural proteins biology biophysics evolutionary genetics viral evolution biochemistry enzyme structure adaptation natural selection virology evolutionary biology evolutionary processes

2012

Selective Pressure Causes an RNA Virus to Trade Reproductive Fitness for Increased Structural and Thermal Stability of a Viral Enzyme

Arabidopsis thaliana high-affinity potassium transporter 1 (AtHKT1) limits the root-to-shoot sodium transportation and is believed to be essential for salt tolerance in A. thaliana. Nevertheless, natural accessions with ‘weak allele’ of AtHKT1, e. g. Tsu-1, are mainly distributed in saline areas and are more tolerant to salinity. These findings challenge the role of AtHKT1 in salt tolerance and call into question the involvement of AtHKT1 in salinity adaptation in A. thaliana. Here, we report that AtHKT1 indeed drives natural variation in the salt tolerance of A. thaliana and the coastal AtHKT1, so-called weak allele, is actually hyper-functional in reducing flowers sodium content upon salt stress. Our data showed that AtHKT1 positively contributes to saline adaptation in a linear manner. Forward and reverse genetics analysis established that the single AtHKT1 locus is responsible for the variation in the salinity adaptation between Col-0 and Tsu-1. Reciprocal grafting experiments revealed that shoot AtHKT1 determines the salt tolerance of Tsu-1, whereas root AtHKT1 primarily drives the salt tolerance of Col-0. Furthermore, evidence indicated that Tsu-1 AtHKT1 is highly expressed in stems and is more effective compared to Col-0 AtHKT1 at limiting sodium flow to the flowers. Such efficient retrieval of sodium to the reproductive organ endows Tsu-1 with stronger fertility compared to Col-0 upon salt stress, thus improving Tsu-1 adaptation to a coastal environment. To conclude, our data not only confirm the role of AtHKT1 in saline adaptation, but also sheds light on our understanding of the salt tolerance mechanisms in plants. Although plants are sessile and are passively exposed to varying environments, they can successfully complete their life cycle and disperse across a range of environmental conditions. These capabilities are attributable to their physiological plastic and complex gene regulation networks, as well as to their affluent genetic diversities. Therefore, revealing the genetic variation that drives adaptation to particular environments is critical for understanding natural selection and evolutionary mechanisms. Numerous studies have elucidated the genetic changes in different plant species in their adaptation to extreme temperature [1], flowering time [2], photoperiod [3], and microbial attack [4]. Soil is a key environmental factor in maintaining plant growth and a major driving force in plant evolution. Nevertheless, little progress has been made in determining the genetic basis underlying plant adaptation to different edaphic conditions. Sodium is a non-essential element that is toxic to most plants. Saline soil is an important driver of genetic variation. A previous study has shown that genome duplication alters fitness to allow plants to adapt to saline environments [5]. Nonetheless, only very limited evidence supports the involvement of a specific gene in local adaptation to saline habitats. HKT1 (High-affinity Potassium Transporter) was first isolated from wheat and was shown to function as a K+/Na+ symporter [6]. Its homologs in various species, such as AtHKT1, OsHKT1; 1 and SKC1/OsHKT1; 5, are generally expressed in xylems of different organs and contribute to plant salt tolerance by unloading sodium from the transpiration stream [7–11]. Interestingly, natural variation in these HKT homologs has been observed in both A. thaliana and rice populations. Four amino acid substitutions in SKC1/OsHKT1; 5 significantly increase its sodium transport activity, restricting sodium flux to leaves and greatly enhancing the salt tolerance of the salt-tolerant rice variety Nona Bokra [9]. A. thaliana is widely distributed in all kinds of geographical environments, and many ecotypes have adapted to saline conditions [12]. Tsu-1 and Ts-1 are two coastal A. thaliana accessions that are highly tolerant to salt stress [13]. However, unlike the salt-tolerant rice variety Nona Bokra, they accumulate very high sodium in their leaves. Genetic analysis has revealed that the high leaf sodium phenotypes of Tsu-1 and Ts-1 are controlled by the AtHKT1 locus [13]. AtHKT1 is expressed at extremely low levels in the roots of these two accessions due to the absence of a distal enhancer [14]. Genome-wide association studies (GWAS) have confirmed that AtHKT1 controls the natural variation of leaf sodium content in A. thaliana populations worldwide and have revealed that A. thaliana accessions with weak AtHKT1 alleles are predominantly distributed in coastal or saline areas [15]. These results, together with other molecular ecology evidence, suggest that AtHKT1 plays a negative role in the salinity adaptation of A. thaliana [12]. However, this conclusion seemingly conflicts with the observation that the null hkt1-1 mutant is hypersensitive to salt stress [16–18]. Solving this puzzle is particularly important because it may uncover the role of AtHKT1 in local adaptation, improve our knowledge of the salt tolerance mechanisms of plants and help resolve the long debate about the relationship between shoot sodium content and salt tolerance. Edaphic salinity threatens plants via osmotic stress and ion toxicity. Halophyte plants generally accumulate sodium in their shoots, as high sodium helps cells resist osmotic stress, and the sodium is not toxic to the cell as long as it is sequestered in the vacuole [19]. Although A. thaliana is not a halophyte, the low expression of AtHKT1 and the high leaf sodium phenotype of coastal salt-tolerant accessions might suggest that A. thaliana may use a strategy similar to halophytes to adapt to saline habitats. It was therefore hypothesized that AtHKT1 might serve as a double-edged sword for salt tolerance, while weak expression of AtHKT1 could well balance the osmotic and ionic toxicity of high salt for the best salt tolerance [13,15]. Alternatively, AtHKT1 might contribute to saline adaptation in an uncharacterized mechanism. Otherwise, it should be independent of the saline adaptation. Seed production is a key index used for evaluating crops economic efficiency in agriculture and is an essential indicator in assessing the evolutionary adaptation of plants [20]. Therefore, in this study, we used seed yield as a central index for estimating the adaptability of plants. We found that AtHKT1 positively drives plant adaptation to a coastal environment, by changing its expression pattern to efficiently retrieve sodium from the stem-to-flower xylem sap and thus reduce floral sodium content to increase fecundity upon salt stress. To examine if there is an expression balance of AtHKT1 for the best salt tolerance, independent knockdown lines of AtHKT1 in the Col-0 background were established. We also isolated an athkt1 null mutant in the Col-0 background as a positive control by backcrossing hkt1-1 with Col-0. Under normal condition, no difference on fecundity was observed among different genotypes (Fig 1A and 1B). Though the coastal saline-adapted accession Tsu-1 produced twice as many seeds as Col-0, whereas the athkt1 knockout mutant yielded almost no seeds when they were treated with 100 mM NaCl, a mild salt stress condition that mimics the natural saline habitat of coastal A. thaliana accessions (Fig 1A and 1C). These data indicate that AtHKT1 is essential for salt tolerance and that Tsu-1 is more adapted to saline soil. Moreover, the number of seeds in AtHKT1 knockdown lines declined with decreases in AtHKT1 levels, while those expressing AtHKT1 to a similar level as Tsu-1 produced a similar number of seeds as athkt1 (Fig 1A, 1C and 1D). In addition, we found that the seed number is positively and linearly correlated with AtHKT1 expression when exposed to salt stress (Fig 1F), but not under normal condition (Fig 1E). To sum up, these data demonstrate that AtHKT1 linearly contributes to salt tolerance, and that weak expression of AtHKT1 does not really balance the osmotic and ionic toxicities for the best salt tolerance. A rough mapping has shown that the salt tolerance phenotype of Tsu-1 is linked to the AtHKT1 locus [13]. Nonetheless, it remained unclear whether AtHKT1 itself, or a locus linked to AtHKT1, controls the phenotype. To address this question, we tried to map the responsible locus or loci by QTL analysis. Epistatic interaction and additive effects generally cause a big problem for mapping QTLs. Fixing genotype of one causal locus could remove or minimize these problems, therefore we synthesized an F2 population derived from a cross between Tsu-1 and athkt1, as both alleles are hypo-functional. We first examined the productivity of the F1 progeny and observed that F1 plants produced an amount of seeds similar to Tsu-1 when exposed to 100 mM NaCl (Fig 2A), suggesting that the adaptive trait is probably controlled by one or more dominant loci. An analysis of 201 F2 individuals further confirmed that a single major locus controls saline adaptation, given that the ratio of salt tolerant F2s (153 individuals) and salt sensitive F2s (48 individuals) conforms to 3: 1 (Χ2 = 0. 08<Χ20. 05 = 3. 84) (Fig 2B). To identify the locus, we performed a QTL analysis using 36 polymorphic genetic markers covering all five chromosomes. As expected, we identified a single major QTL on chromosome 4 that contributed to 53. 1% of the variance in seed yield under salt stress (Fig 2C). Interestingly, this candidate region contains AtHKT1, but none of the NHX genes that could sequester sodium in the vacuole [21]. We employed an enlarged F2 population with 537 individuals and developed 10 polymorphic markers in the candidate region for fine mapping. Based on genotype and phenotype analyses of the F2 individuals or their F3 progenies, we narrowed down the causal gene to a 100-kb region between genetic markers GM635 and GM645 (Fig 2D). This region contains only 26 annotated genes; and only AtHKT1 is related to salt tolerance (S1 Table), suggesting that the AtHKT1 is the causal locus for adaptation of Tsu-1 to saline conditions. To verify this, we knocked out AtHKT1 in Tsu-1 using CRISPR (clustered regulatory interspaced short palindromic repeats) -Cas9 (CRISPR-associated protein 9) technology [22]. In the T2 generation, we obtained two homologous knockout mutants of AtHKT1 and named them hkt1-c1 and hkt1-c2 (c stands for alleles generated by CRISPR-Cas9 technology). In hkt1-c1, AtHKT1 has a thymine inserted between the 450th and 451st nucleotides of the coding sequence (S1A Fig), while, in the hkt1-c2, four nucleotides (from the 447th to the 450th nucleotides after the start codon) are deleted (S1B Fig), both of which result in frame shifts. Under normal conditions, there were no differences on seed production among hkt1-c1, hkt1-c2 and Tsu-1 (S1C Fig). Yet, when treated with 100 mM NaCl, the two mutants were totally sterile, whereas the wild control Tsu-1 was still productive (S1D Fig). The phenotypes of the two mutants were comparable to or even more severe than that of athkt1, as athkt1 still produced some seeds, although considerably less than Col-0 under the same conditions (Fig 3A and 3B). Considering that Tsu-1 produced significantly more seeds than Col-0 in the same salt treatment experiment (Fig 3A and 3B), we conjectured that AtHKT1 is responsible for the natural variation in saline adaptation between Tsu-1 and Col-0. To further confirm that the phenotypes of hkt1-c1 and hkt1-c2 were caused by AtHKT1 knock-out, we crossed them with athkt1 respectively. The F1 plants were all sterile when treated with 100 mM NaCl (S2A and S2B Fig), verifying that the phenotypes of hkt1-c1 and hkt1-c2 were not an off-target effect. To further determine whether a single AtHKT1 locus is sufficient to drive the variation in the salinity adaptation of A. thaliana, we cloned the genomic DNA of AtHKT1, including its long promoter from Col-0 and Tsu-1, and introduced each construct into the null athkt1 mutant separately. We tested the salt tolerance of three independent transgenic lines for each construct by assessing productivity. Transgenic lines with either the Col-0 AtHKT1 fragment or the Tsu-1 fragment could complement the salt-hypersensitive phenotype of athkt1 (Fig 3C). Nevertheless, the Col-0 AtHKT1 fragment could only restore the saline adaptability of athkt1 to the Col-0 level, whereas the Tsu-1 fragment was sufficient to restore athkt1 to the Tsu-1 level of adaptability (Fig 3D). These data demonstrated that a single AtHKT1 locus is sufficient to drive the natural variation in the salinity adaptation of Col-0 and Tsu-1. Root AtHKT1 expression drives leaf sodium content in both Col-0 and Tsu-1 [13]. Nevertheless, it was unclear in what tissue AtHKT1 drives salt tolerance in these ecotypes. To address this question, we performed two reciprocal grafting experiments; one between athkt1 and Col-0, and the other was between hkt1-c1 and Tsu-1. We observed that the grafted plants with athkt1 as the shoot, and Col-0 as the root, were as tolerant as self-grafted Col-0 plants; whereas the grafted plants with Col-0 as the shoot, and athkt1 as the root, were similar to self-grafted athkt1 plants (Fig 4A and 4B). This result indicates that salt tolerance of Col-0 predominantly attributes to root AtHKT1. However, the grafting experiment between Tsu-1 and hkt1-c1 showed the opposite results. The grafted plants with Tsu-1 scion and hkt1-c1 stock produced a similar salt tolerant phenotype as Tsu-1 self-grafted plants, whereas hkt1-c1 self-grafted plants and grafted plants with hkt1-c1 scion and Tsu-1 stock were sterile upon salt stress (Fig 4C and 4D). These data showed that the shoot expression of AtHKT1 drives the saline adaptation of Tsu-1. Moreover, they suggest that although Tsu-1 AtHKT1 is hypo-functional in roots, it is hyper-functional in shoots. The grafting results inspired us to analyze the expression of AtHKT1 in different aerial tissues. Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) showed that differences in AtHKT1 expression between Tsu-1 and Col-0 depend on the tissue and the treatment. AtHKT1 was expressed at similar levels in the rosette leaves of 4-week-old of Tsu-1 and Col-0 when exposed to 100 mM NaCl for 24 h, but its expression was higher in the rosette leaves of Col-0 than of Tsu-1 under normal conditions (S3A Fig). In flowers, Tsu-1 expressed AtHKT1 at a similar level as Col-0 under normal conditions but it expressed less than Col-0 when treated with salt (S3B Fig). Interestingly, Tsu-1 expressed significantly higher AtHKT1 in the stems, cauline leaves and siliques compared to Col-0 (Fig 5A). When treated with salt, Tsu-1 expressed 6 times more AtHKT1 than did Col-0 in the stems, which is consistent with the hypothesis that Tsu-1 allele of AtHKT1 is hyper-functional in shoot. To confirm the qRT-PCR results and further examine the spatial expression patterns of AtHKT1 in detail, we generated transgenic Col-0 plants carrying a β-glucuronidase (GUS) reporter driven by the long promoter of AtHKT1 from either Tsu-1 or Col-0. We examined 5 independent transgenic lines expressing pAtHKT1Tsu-1: : GUS and 7 independent lines expressing pAtHKT1Col-0: : GUS. The transgenic lines containing the same construct showed consistent GUS signals. At the seedling stage, all 5 lines expressing pAtHKT1Tsu-1: : GUS exhibited very weak GUS activity in the roots, whereas 6 of the 7 lines expressing pAtHKT1Col-0: : GUS strongly expressed GUS in the stele of the roots (Fig 5B). These observations were consistent with previous findings [14] and with our qRT-PCR results (Fig 5C), indicating that the constructs were correctly expressed in Col-0 plants. Contrary to the root results, the GUS signal in stems and cauline leaves of 4-week-old of plants was much stronger in Tsu-1 than in Col-0 (Fig 5D). This result further confirmed the qRT-PCR results. We further examined the detailed expression site (s) of AtHKT1 in stems by hand sectioning of the pAtHKT1: GUS transgenic lines. The GUS signal driven by the Tsu-1 AtHKT1 promoter was mainly observed in the xylem of the stem (Fig 5E), indicating that Tsu-1 AtHKT1 predominantly functions in removing sodium from the flux to the flowers. By contrast, the GUS signal driven by Col-0 AtHKT1 promoter was very weak in the xylem and was observable in the phloem (Fig 5E). These data demonstrated that variation in the AtHKT1 promoter affects not only the expression level but also the spatial specificity of AtHKT1 expression. In addition to large insertion/deletion (indel) polymorphisms in the promoter region (S4A Fig), multiple single nucleotide polymorphisms (SNPs) were also observed in the coding sequence with several affecting the amino acid sequence (S4B Fig). To test whether the amino acid changes in AtHKT1 in Tsu-1 affect its transport activity, we expressed AtHKT1 protein from Tsu-1 and Col-0 in Xenopus laevis oocytes for electrophysiological analysis. As expected, we detected sodium transporter activity of AtHKT1 (Fig 5F), but there was no significant difference between the proteins from Col-0 and Tsu-1, indicating that the amino acid changes do not affect AtHKT1 activity. This observation is consistent with previous findings that these changed amino acids are not in conserved regions of AtHKT1 [23,24]. The increased expression of AtHKT1 in the xylem of Tsu-1 stems suggested that Tsu-1 is more effective than Col-0 at limiting sodium flow to flower. To confirm this hypothesis, we measured the sodium content in the flowers. Under normal conditions, athkt1 accumulated 38% higher sodium in flowers than Col-0, and hkt1-c1 accumulated 46% higher than Tsu-1, but there were no significant differences between the wild-types and between the mutants (Fig 6A). Nevertheless, the sodium content in Tsu-1 flowers was only 62% of that in Col-0 flowers when exposed to salt treatment (Fig 6B). By contrast, upon salt stress, the loss of AtHKT1 function in Col-0 resulted in a 199% increase in floral sodium, whereas the loss of AtHKT1 function in Tsu-1 resulted in a 275% increase in floral sodium (Fig 6B). These data indicated that AtHKT1 in Tsu-1 stems is more effective in limitation of sodium flux to flowers than AtHKT1 in Col-0 roots. In addition, we analyzed the relationship between floral sodium content and seed number under salt stress and found these two variables to be negatively correlated (Fig 6B Embedded), demonstrating that the sodium content in flowers under salt stress is a determinant of saline adaptation. An examination of stem sodium content provided additional evidence supporting the idea that Tsu-1 AtHKT1 is involved in sodium removal in stems. There was no significant difference in stem sodium content among Col-0, Tsu-1, athkt1 and hkt1-c1 under normal conditions (Fig 6C). However, under salt stress, Tsu-1 accumulated 59% more sodium in stems than did hkt1-c1, whereas Col-0 accumulated 57% less stem sodium than athkt1 (Fig 6D); these results show that AtHKT1 in Tsu-1 promotes sodium accumulation in stems and that AtHKT1 in Col-0 has the opposite effect. This observation further confirmed that Tsu-1 AtHKT1 is hyper-functional in restricting sodium flow from stems to flowers and that Col-0 AtHKT1 functions to limit sodium flow from roots to shoots, and explains why the Tsu-1 allele of AtHKT1 endows the coastal accessions with higher adaptation ability to salinity environment than the inland accessions. Previous studies have illustrated that the leaf sodium content cross different A. thaliana accessions under normal conditions is not associated with its salinity tolerance [15]. However, the leaf sodium content under salt stress is unknown. Therefore, we also measured the leaf sodium content under normal conditions and after salt treatment. Consistent with previous studies, Tsu-1 and the two null mutants accumulated much more leaf sodium than Col-0 under normal conditions (Fig 6E). The result was similar when the plants were treated with salt, as the leaf sodium content in Tsu-1 and the two null mutants was much higher than that in Col-0 (Fig 6F). As seed number is an essential trait for population expansion [20], we used this trait for assessing saline adaptation of plants. Previous studies have suggested that AtHKT1 might be fine-tuned for balancing the ionomic and osmotic stresses caused by high salt, as weak alleles of AtHKT1 might improve salt tolerance, whereas null alleles of AtHKT1 suppress salt tolerance [13]. Our observation of the linear relationship between salt tolerance and AtHKT1 expression levels argued against this hypothesis. However, our QTL analysis suggested that AtHKT1, but not additional locus, control the salt adaptability of Tsu-1. Knockout of AtHKT1 in Tsu-1 by CRISPR-Cas9 technology and complementation athkt1-1 with Tsu-1 and Col-0 AtHKT1 finally established that AtHKT1 is not only a requirement for salt tolerance of Tsu-1 but is also a predominant, if not sole locus that drives the natural variation in salinity adaptation between Col-0 and Tsu-1. Though AtHKT1 has been found to drive natural variation in leaf sodium content, this is the first direct evidence to show that AtHKT1 plays a positive role in natural selection of saline adaptive A. thaliana accessions. These genetic evidences revealed that the Tsu-1 allele of AtHKT1 is hyper-functional but not hypo-functional. Our grafting experiments revealed that the saline adaptability of Tsu-1 is driven by shoot AtHKT1 expression, which is in contrast with the premise that salt tolerance of Col-0 is predominantly driven by root AtHKT1 expression. This data indicates that AtHKT1 functions differently in different accessions. GUS reporter system and qRT-PCR confirmed that Tsu-1 AtHKT1 is hyper-functional relative to Col-0 AtHKT1 due to its high expression in stems. Interestingly, histochemical staining experiment revealed that Tsu-1 AtHKT1 is predominantly expressed in xylem parenchyma cells of stems, suggesting that it functions in restricting sodium in the stems. Measurements of stem and floral sodium concentrations of different genotypes confirmed this hypothesis. Therefore, the Tsu-1 AtHKT1 functions through a new mechanism, which is to restrict sodium in the leaves and stems. This mechanism is in contrast with that of Col-0 AtHKT1 that restricts sodium in the roots. Flower is the most important organ of the plant. However, the germ cells in flowers are very sensitive to sodium, as such cells are weak in sodium compartmentalization and detoxification due to low vacuole capacity. By contrast, leaf cells have very large vacuoles; thus, even if they accumulate a high level of sodium, it is not highly toxic if it is sequestered in the vacuole [19]. Instead, the high accumulation of sodium in leaves would facilitate resistance to osmotic stress. This difference results in flowers being much more sensitive to salt compared to leaves. Consequently, reducing the flower sodium content but not the leaf or stem sodium content is much more important for A. thaliana upon salt stress. The Tsu-1 AtHKT1 and the Col-0 AtHKT1 represent two mechanisms for restricting sodium flow to flowers; Tsu-1 AtHKT1 unloads sodium to the leaves and stems, and Col-0 unloads sodium in the roots. According to the floral sodium concentrations in Tsu-1 and Col-0, we can conclude that the Tus-1 mechanism is more effective than Col-0 mechanism in limiting sodium flow to flowers. In addition, a previous study has shown that the Col-0 mechanism could inhibit root growth upon salt stress due to its improving sodium accumulation in the roots [17]. These findings together explain why the mechanism of Tsu-1 AtHKT1 endows A. thaliana with a higher adaptation ability to saline conditions. However, we also noticed that the Col-0 mechanism was selected by most inland accessions though it represents a low efficient mechanism in salt tolerance. So far, it is hard to conclude what is the driving force behind the selection of such a mechanism, but we suspect that it might be associated with the edaphic sodium concentrations. Inland soil contains subtoxic levels of sodium, and accumulating sodium in the roots thus might be the most economical way for A. thaliana to minimize the side-effect of sodium on flower development. By contrast, the Tsu-1 mechanism involves transporting of sodium to shoot cells and sequestrating it into the vacuoles that consume much more energy. Therefore, the selection of Col-0 mechanism or Tsu-1 mechanism might reflect the sodium levels in their habitat. According to a previous study, one of the tandem repeats in the distal promoter of AtHKT1 is an enhancer that promotes the expression of AtHKT1 in roots, and the absence of this distal repeat in Tsu-1 was confirmed to be responsible for the down regulation of AtHKT1 in roots [13]. However, we noted in the results of a previous study that the expression of AtHKT1 is apparently upregulated in shoots and markedly downregulated in roots in the absence of this hypothesized distal enhancer [14]. In our study, the expression of Tsu-1 AtHKT1 was significantly higher than that of Col-0 AtHKT1 in the shoots of young seedlings. This observation suggested that the second tandem repeat might not only serve as an enhancer in roots but may also function as a repressor in shoots. In this case, the high expression of AtHKT1 in Tsu-1 stems is probably caused by the deletion of the second tandem repeat. AtHKT1 has also been reported to be regulated by DNA methylation in the promoter [14]. However, we found no difference of the small RNA target site in the promoters of AtHKT1 between Tsu-1 and Col-0 (S4 Fig). Of course, we cannot exclude the possibility that other polymorphisms among the many in the promoter region control the variation in AtHKT1 expression patterns. Further research is necessary to clarify this question. Shoot sodium content is often used as a parameter for assessing salt tolerance of plants, but the relationship between shoot Na+ and salt tolerance raises arguments. In rice, low shoot Na+ is an indicator of salt tolerance [27]. A typical case is the identification of SKC1/OsHKT1; 5 that reduces leaf salt content and enhances salt tolerance [9]. However, it was reported that the growth and yield of wheat upon salt stress is not correlated with sodium concentrations in leaf blade or the whole shoot [28]. Meanwhile, the salt tolerant tomato varieties accumulate higher sodium in stems and leaves than the sensitive varieties [29,30]. These findings suggest that the relationship between shoot sodium content and salt tolerance might be genetic background dependent, and therefore not stable index for assessing plant salt tolerance. This study further supports the above conclusion. In this study, we observed that the sodium levels in flowers, but not in leaves, are positively associated with salinity adaptability. This suggests that floral sodium content might be more suitable for assessing salt tolerance of plants. Of course, this does not mean the leaf sodium does not matter for salt tolerance of plants. Actually, the increased leaf sodium causes severe leaf necrosis in athkt1 null mutants when exposed to high salt conditions [16,18,31]. But upon mild salt stress, the flowering phenotype of athkt1 null mutants is more obvious than the leaf phenotype, as its leaf still has photosynthesis ability (S5A Fig), but it is almost sterile in seed production. Therefore, leaf sodium may play different roles upon different levels of salt conditions. It is worthwhile pointing out that A. thaliana is a species of glycophytes and even salt tolerant ecotypes/accessions may not successfully colonize in high salt area. Even Tsu-1 barely produces seeds when exposed to 150 mM NaCl. Therefore, mild salt stress more accurately reflects the real habitat of A. thaliana. On the other hand, the leaf necrosis in Tsu-1 caused by high salt treatment is less severe than Col-0 (S5B Fig), though Tsu-1 accumulates as much sodium as hkt1 null mutants in the leaves. This could be caused by differential expression pattern of AtHKT1 in leaves, or controlled by some other QTLs/genes. To our knowledge, the relationship between flower sodium content and saline adaptability has never been reported. Identifying the positive relationship between the two factors provides a new angle to assess plant salt adaptability. Of course, we are not sure if this conclusion could be extended to all plant species in current stage, and further investigation is necessary to answer this question. The seed production is a most important trait of crops in agriculture. Our finding that different expression patterns and levels of AtHKT1 result in different yield upon salt stress could inspire engineering salt tolerant plants by manipulating expression pattern of HKT1. Moreover, this finding could also be translated to crops to improve their yields in saline area. Seeds of A. thaliana were surface sterilized and sown on 1/2 Murashige and Skoog (MS, Sigma-Aldrich, St. Louis, USA) medium (pH 5. 7) containing 1% (w/v) sucrose, followed by stratification at 4 °C in dark for 5 days. The seedlings were transferred into pots with artificial soil in a phytotron (16 h light/8 h dark, 22 °C day/20 °C night, 80% humidity, 80 μmol·m-2·s-1) after being grown on the plates for 7 days. Two-week-old soil-grown plants that were irrigated with 100 mM NaCl twice for two weeks and 1 additional week later plants were used for phenotyping. Four-week-old plants irrigated with 100 mM NaCl for 24 h were used for gene expression analysis and GUS staining. Each line used for salt treatment contained ≥ 12 plants, and every treatment included 3 biological replicates. The athkt1 mutant was isolated from an F2 population derived from a backcross between hkt1-1 (CS6531, obtained from the Arabidopsis Biological Resource Center) and Col-0. Seeds of A. thaliana (Tsu-1, Col-0, AtHKT1 RNAi lines and athkt1) were stratified for 5 days at 4 °C in water. The stratified seeds were sowed in punctured holes of 1. 5-mL Eppendorf tube caps stuffed with 0. 1% agar, and the caps were then placed on 1. 5-mL Eppendorf tube containing Hoagland solution for 14 days in a chamber at 22 °C, 70% relative humidity, and light intensity of 80 μmol·m-2· s-1 on 8 h light/16 h dark. The seedlings together with caps were then transferred to a big box containing Hoagland solution for growing 2 additional weeks. The culture solution was refreshed every 5 days. The roots of one-month-old seedlings were collected for qRT analysis. An F2 population derived from Tsu-1 and athkt1 was used to map the causal gene of the salt tolerance of Tsu-1. The number of seeds was calculated based on an estimation of the total grain weight and 100-grain weight. A set of 36 CAPS (Cleaved Amplified Polymorphism Sequences) and SSR (Simple Sequence Repeat) markers covering the 5 chromosomes of A. thaliana were used to genotype all F2 individuals. QTL analysis was performed using the R/qtl package (http: //www. rqtl. org). To narrow down the candidate region, 537 F2 individuals and 10 newly developed CAPS markers were used to fine map the target region. All CAPS and SSR markers are listed in S2 Table. For the complementation experiment, a 9. 3-kb AtHKT1 fragment including a 5. 3-kb promoter region from Col-0 was amplified by overlap PCR using primers HKT-pro-gene L, PHMS-HKT1-MID-R, PHMS-HKT1-MID-L and HKT-pro-gene R (S3 Table). The complementation fragment for Tsu-1 AtHKT1 was amplified using the same strategy but with primers Tsu-HKT-pro-gene L, PHMS-HKT1-MID-R, PHMS-HKT1-MID-L and HKT-pro-gene R. The fragments were cloned into a binary vector, pHMS, which was modified from the pHB vector [32] by infusion recombination cloning using a Hieff Clone One-step PCR Cloning Kit (Yisheng Co. Ltm, Shanghai, China). To construct the promoter: GUS vector, the AtHKT1 promoters were amplified from genomic DNA using primers HKT-GUSL TONG and HKT-GUSR TONG for Col-0 and primers HKT-GUSL TONG 2 and HKT-GUSR TONG for Tsu-1. The fragments were recombined into the binary vector pCAMBIA1303 using the Hieff Clone One-step PCR Cloning Kit. GUS histochemical staining was performed as described previously [9]. To construct the Col-0 AtHKT1 RNAi vector, two 184-bp fragments from the AtHKT1 full-length cDNA sequence were amplified using the primers AtHKT1 F (PstI restriction site) and R (XbaI restriction site) for sense and AtHKT1F (SacI restriction site) and R (NotI restriction site) for antisense. Both amplified fragments were sequenced and inserted into an intermediate vector, pBluescript II SK (+/-), containing a 120-nucleotide intron from the A. thaliana RTM1 gene at the XbaI and NotI sites. Then, the hairpin fragment was sub-cloned into the binary vector pCAMBIA2301, which contained a cauliflower mosaic virus (CaMV) 35S promoter and a NOS terminator. The kanamycin resistance gene (NPT II) was used as a selective marker gene. The construction of a CRISPR-Cas9 vector targeting AtHKT1 was performed as described previously [22]. Two oligos, AtHKT1 sgRNA4-F and sgRNA4-R, were synthesized and annealed to form a target site of 20bp in length with a requisite proto-spacer-adjacent motif, PAM (NGG) sequence at the 3’ end and G at the 5’ end of the sequence. The target site, containing two BbsI digestion sites, was cloned into the intermediate vector AtU6-26SK. The chimeric RNA expression cassette between KpnI and SalI was cloned into the KpnI and EcoRI sites of the pCambia1300 vector (Cambia, Canberra, Australia) together with the SalI and EcoRI fragment of the Cas9 expression cassette from the 35S-Cas9-SK vector, which was a gift from the Zhu lab [22]. The expression constructs were transformed into Agrobacterium tumefaciens strain GV3101, and were then introduced into the backgrounds as indicated using the floral dip method (26). Transgenic lines were screened on 1/2 MS medium solidified with agar containing 50 mg/L hygromycin and 1% sucrose. We inserted the cDNA of AtHKT1 from the Col-0 and Tsu-1 alleles into the SmaI site of the expression vector pGEMHE. The capped cRNA was transcribed using the RiboMAX Large Scale RNA Production System-T7 kit (P1300, Promega, Wisconsin, USA). The cRNA quality was verified using agarose gel electrophoresis. The concentration was determined at 260 and 280 nm and adjusted to a final concentration of 0. 8 μg μl-1. We injected freshly isolated X. laevis oocytes with 23 nL of cRNA and used the oocytes for voltage-clamp experiments 3–4 days later. Electrophysiological experiments were conducted using a two-electrode voltage clamp amplifier as described previously [9]. We bathed the oocytes in a solution containing 6 mM MgCl2,1. 8 mM CaCl2,10 mM MES-Tris (pH 5. 5), 185 mM D-mannitol and 10 mM Na-glutamate salts. The voltage protocols used are described in the figure legends. The process of A. thaliana elemental analysis via inductively coupled plasma mass spectrometry (ICP-MS) has been described previously [33]. Briefly, different tissues, including flowers, whole stems and rosette leaves, from five-week-old plants treated twice with 100 mM NaCl or from untreated plants were cut with a scalpel while holding the plants with plastic tweezers. The collected tissues were rinsed with four times with 18-MΩ water in a 1000-mL breaker to wash off the impurities. Then, the rinsed samples were placed into a glass tube, and the samples were shifted to the bottom of the tube, making sure that no samples were left on the tube wall. The tubes were transferred to an oven at 65°C for 12 h. After cooling, twelve samples were weighed out on an analytical balance. All the samples, including the blank controls, were digested with 1 mL of concentrated nitric acid containing an indium (In) internal standard at 115°Cfor 4 h, and then the digested samples were diluted to 10 mL with 18 MΩ water. Elemental analysis for Li, B, Na, Mg, P, S, K, Ca, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Rb, Sr, Mo, and Cd was performed using an ICP-MS (NexION 350D; PerkinElmer) coupled to an Apex desolvation system and an SC-4 DX auto sampler (Elemental Scientific Inc. , Omaha, NE, US). All samples were normalized to calculate weight, as determined using a heuristic algorithm using the best-measured elements, the weights of the eight weighed samples, and the solution concentrations, as previously described. Seedlings were grafted as previously described [13]. Graft unions were examined under a stereoscope before transfer to potting soil to identify any adventitious root formation at or above the graft unions. Healthy grafted plants were transferred to pots and grown in a controlled environment for 2 weeks. Then, the plants were irrigated with a 100 mM NaCl solution twice, and after 2 weeks, the plants were photographed. After harvesting, the graft unions were examined again, and grafted plants with adventitious roots or without a clear graft union were excluded from subsequent analysis. Total RNA was extracted from plants using TRNzol A+ RNA Purification reagent (TIANGEN, DP421, Beijing, China). Two micrograms of total RNA were used to synthesize first-strand cDNA with TransScript One-Step gDNA Removal and cDNA Synthesis SuperMix (Transgen, AT311-02, Beijing, China). qRT-PCR was performed using SYBR Green PCR Master Mix (TRT-101, TOYOBO, Osaka, Japan) with the first-strand cDNA as a template on a Real-Time PCR System (Bio-Rad CFX thermocycler, California, USA). Primers for qRT-PCR were designed using Primer Express Software Version 3. 0 (Applied Biosystems, USA). The primers HKT1 exon2-3-1L and HKT1 exon2-3-1R were designed to span an exon-exon junction and were used to detect the gene expression level. The primers UBCF and UBCR were designed for UBIQUITIN-CONJUGATING ENZYME21 (At5g25760), which was used as the control gene. The primer sequences are shown in S3 Table. Expression data analysis was performed as described previously [34].

Identifying the genetic variation driving plant adaptation to salinity is critical for understanding natural selection and evolutionary mechanisms. In this study, we have revealed that the gene AtHKT1 drives natural variation in the adaptation of A. thaliana to salinity. Our evidences directly show that the AtHKT1 allele in the coastal accession Tsu-1 endows A. thaliana with enhanced adaptability to salinity. Our further experiments have demonstrated that the enhanced adaptability of the coastal accession is attributed to high AtHKT1 expression in stems, leading to low sodium levels in flowers. Our work not only elucidates the role of AtHKT1 in local adaptation to salinity but also improves our understanding of salt tolerance mechanisms of plants.

Abstract Introduction Results Discussion Materials & methods

plant anatomy ecology and environmental sciences brassica plant physiology plant science model organisms experimental organism systems plant pathology evolutionary adaptation plant ecology plants flowering plants arabidopsis thaliana research and analysis methods chemical properties physical chemistry salinity chemistry leaves seeds flowers plant defenses eukaryota plant and algal models plant resistance to abiotic stress ecology evolutionary processes biology and life sciences physical sciences evolutionary biology plant-environment interactions organisms

2017

AtHKT1 drives adaptation of Arabidopsis thaliana to salinity by reducing floral sodium content

Social insects make elaborate use of simple mechanisms to achieve seemingly complex behavior and may thus provide a unique resource to discover the basic cognitive elements required for culture, i. e. , group-specific behaviors that spread from “innovators” to others in the group via social learning. We first explored whether bumblebees can learn a nonnatural object manipulation task by using string pulling to access a reward that was presented out of reach. Only a small minority “innovated” and solved the task spontaneously, but most bees were able to learn to pull a string when trained in a stepwise manner. In addition, naïve bees learnt the task by observing a trained demonstrator from a distance. Learning the behavior relied on a combination of simple associative mechanisms and trial-and-error learning and did not require “insight”: naïve bees failed a “coiled-string experiment, ” in which they did not receive instant visual feedback of the target moving closer when tugging on the string. In cultural diffusion experiments, the skill spread rapidly from a single knowledgeable individual to the majority of a colony’s foragers. We observed that there were several sequential sets (“generations”) of learners, so that previously naïve observers could first acquire the technique by interacting with skilled individuals and, subsequently, themselves become demonstrators for the next “generation” of learners, so that the longevity of the skill in the population could outlast the lives of informed foragers. This suggests that, so long as animals have a basic toolkit of associative and motor learning processes, the key ingredients for the cultural spread of unusual skills are already in place and do not require sophisticated cognition. Social learning is widespread in animals [1,2] and can enable novel behavior routines, sometimes introduced by a single “innovator, ” to spread among individuals in a group [1–5]. Examples are potato washing and termite fishing in primates, pinecone stripping in rodents and milk bottle opening in birds [6–10]. Such phenomena in animals have received considerable attention because researchers hoped to discover the key evolutionary ingredients of the cultural processes that define us as humans [11–13]. Social learning can lead to group-specific behavior patterns that are shared by a large number of animals in an area [13–15]. Two key components of culture-like phenomena in animals include spreading of a new behavior via social learning as well as persistence of the behavior in groups for extended periods of time [5,12,14,16,17]. Individuals that picked up the information from the initial demonstrator (s) can themselves become demonstrators to uninformed individuals [18], a process by which such group-specific phenomena can, at least in principle, persist across many generations [5,19]. In some cases, culture-like phenomena such as beach hunting in killer whales (Orcinus orca) [17] and lexigram communication in bonobos (Pan paniscus) [20] require relatively sophisticated learning mechanisms, for example imitation and/or teaching [21–23]. In insects, seemingly complex processes of social information acquisition, for example the gradual consensus building that occurs when honeybee swarms decide on new nesting locations [24], can sometimes instead be mediated by relatively simple learning mechanisms [25,26], suggesting that cultural processes may not necessarily require sophisticated learning abilities [27–29]. The spread of novel foraging techniques by means of a formal transmission chain experiment [5], in which an experimentally induced innovation is seeded into a group and the subsequent spread is monitored in a social network analysis, has never been explored in an insect. We suggest that doing so provides a unique opportunity for the exploration of the basic cognitive elements required for culture [18,30]. A variety of impressive cognitive skills in social bees has been described, such as object categorization, simple spatial concepts, and numerosity, as well as social learning skills by which bees can acquire information about valuable food sources by observing conspecifics [25,26,31–33]. Scholars in comparative cognition have advocated testing animal intelligence by exploring the flexibility and innovative skills in solving tasks that are relatively remote from the animal’s natural behavior [34,35]. Here, we explore whether bumblebees (Bombus terrestris) have the capacity to learn a string pulling task [36]. We also test whether naïve observers can acquire this technique through observation of trained demonstrators, as occurs in naturally widespread foraging techniques [37,38]. Finally, we explore whether and how swiftly such an experimentally introduced innovation can spread through a bumblebee colony. To do this, we used an “open group diffusion” paradigm (forager pairings were not determined by the experimenter) [39] to determine whether diffusion (or transmission) chains that begin with a trained bee follow a sequence in which observers successively become models for a subsequent observer in the chain [5]. To test bees’ capacity to learn the technique of string pulling, we first challenged untrained individuals with a stepwise training procedure (Materials and Methods; S1–S4 Videos). We presented individual bees with three blue artificial flowers with a string attached to each flower and placed under a small transparent Plexiglas table (Materials and Methods). After learning to associate the reward with artificial flowers in a flight arena (Step 0, Fig 1A), but prior to string pulling training, none of the bees from the eight colonies in which individuals were tested singly (n = 291) could solve the string pulling task on their first 5-min attempt (Test 1, Fig 2B). Naïve to the string task but attracted to the artificial flowers, these bees tried to reach the reward from the top of the table through the Plexiglas. In comparison, we were able to train 23 of 40 individuals (Colony 1) through a stepwise training procedure to successfully pull a string to obtain reward (Fig 1B horizontal black bar in column 4, S1–S4 Videos). The stepwise training consisted of four steps of incremental difficulty within which flowers with strings were placed at progressively more distant positions under the transparent table (Steps 1–4, Fig 1A and 1B). On average, successful training for an individual bee took 309 ± 18 min. Gaining access to the reward in the final step required grasping the string with the forelegs and/or mandibles and pulling it closer (S4 Video). The mean time required (latency) to obtain sucrose decreased significantly as a function of experience within each of the four successive training phases (Friedman test, Step 1: χ24 = 59. 1, p = <0. 001; Step 2: χ24 = 53. 1, p = <0. 001; Step 3: χ24 = 52. 1, p = <0. 001; Step 4: χ210 = 92. 3, p < 0. 001; Fig 1C and 1D). Eight, three, one, and five individuals gave up at Steps 1,2, 3, and 4, respectively, either because they ceased foraging activity or had irregular foraging activity (n = 11), or because they failed to obtain the reward (n = 6). Three of these successfully trained bees were later used as demonstrators in the social learning experiment. The success of bees learning such a behavior raises the question about the mechanisms by which the demonstrators learned to pull the string. One possibility is that demonstrators are stimulated to repeat the specific sequence of actions (moving the string with their legs) that induces the conditioned stimulus (i. e. , the blue flower positioned under the table) to move a little closer. If so, we would expect bees not to move the string with their legs and fail at the task if the colored target stimulus is not present. To test this prediction, we challenged bees (Colony 2) to access the reward when a string was attached to only a colorless inverted Eppendorf cap containing sucrose solution (Materials and Methods) immediately after their initial stepwise training and then again after extensive experience with blue flowers and strings. Without a colored stimulus, only 2 of 15 bees tested obtained the reward after their initial training. We thus hypothesized that relatively inexperienced bees rely on visual feedback of the colored target moving closer while the string is being pulled. To explore this further, we examined the video material for the unsuccessful bees to see if they would attempt to pull the strings and then abort this action when visual feedback was not forthcoming. However, none of the unsuccessful bees demonstrated even an aborted pulling action on the colorless flower’s string. This suggests that most relatively inexperienced bees require the presence of the blue flower to even begin attempting to string pull. (However, there is also evidence for the importance of visual feedback during pulling from an experiment with coiled strings; see section The Mechanisms of Observational Learning in String Pulling.) Conversely, after 48 h of extensive training (20 instances of string pulling), 11 of the 15 foragers solved the task without feedback from the moving blue flower (S5 Video). Latency to obtaining the reward (147 ± 23. 44 s) was much higher than for normal blue flower training (22. 1 ± 1. 5 s; t test: t25 = 6. 25, p < 0. 0001). The subjects’ success differs significantly from their performance when they were relatively inexperienced (McNemar Test, χ21 = 7. 111, p = 0. 008), thus indicating that the majority of highly experienced individuals may no longer require visual feedback to perform the necessary sequence of motor actions. In fact, experienced bees may not need the blue flower at all and perhaps have associated the string with the reward. We gave 50 individuals (Colony 1) the opportunity to solve the string pulling task spontaneously after having learnt that blue flowers are rewarding when they were openly accessible during pretraining (for a 5-min observation period). None of these individuals solved the task. When given a second 5-min opportunity, two of 25 untrained bees succeeded in obtaining the reward (S6 Video). However, they were more than ten times slower at obtaining the reward than experienced string pullers (22. 1 ± 1. 5 s, mean ± standard error [s. e. ], Mann–Whitney U test, U23 < 0. 001, p = 0. 024), requiring a relatively long latency of 245 ± 3. 53 s. These two bees were exceptionally explorative, trying a wide variety of methods, and solved the task in several attempts by moving the string accidently while trying to reach the flower under the table (see S6 Video and legend for more information). This shows clearly that string pulling can be learned individually by some bumblebees, but this may be an exceptionally rare ability. Across experiments (see below), 291 naïve individuals were tested once, and a total 110 were tested twice, but no further “innovators” were found. In one experiment (the transmission chain experiment below, in which control colonies were not seeded with a skilled demonstrator), bees were given extensive opportunities. After 5 d of foraging, with a maximum number of 18 foraging bouts per individual, no single bee learned to pull the string. Of the 165 bees tested in this experiment in total, nine individuals were tested more than 10 times, and 26 more than 5 times, but all were invariably unsuccessful. Thus, solving a string pulling task spontaneously is a relatively rare occurrence in bumblebees and might either reflect an unusually explorative “personality” in these individuals or simple “luck” in the process of random exploration. We explored whether uninformed bees (Colony 1) could learn this novel foraging technique via observation. After pretraining on blue flowers and Test 1 (Materials and Methods), an uninformed observer bee was placed in a transparent chamber (Fig 2A) where it could observe a demonstrator solve the string pulling task ten times. These observers (n = 25) were subsequently tested on the string pulling task alone (Test 2, Fig 2B). In this experiment, observers never interacted directly with demonstrators in the flight arena and had access only to visual social information (S7 Video). Sixty percent of the individuals (15 of 25) that had the opportunity to observe a skilled demonstrator managed to pull the string and obtained the reward on the first trial after having observed the demonstration (Test 2, Fig 2C, S6 Video). These bees, however, were initially almost as slow as the two individuals that solved the tasks without demonstration (181 ± 19 s; Fig 2D). We speculate that the observers picked up the correct location to access the reward from observing skilled demonstrators but did not learn from them the actual technique of string pulling (further explored in the section beneath about the mechanisms of social learning). We also wished to disentangle the effects of demonstrator copying and object movement copying in how string pulling was learnt by observation. To this end, we used an experimental “ghost control” ([40], S8 Video). We trained 15 nonsocial observers (Colony 3) in exactly the same manner as above with the modification that the flowers were moved without a visible actor: an experimenter pulled the flowers with thin nylon threads attached to the strings while the observers were locked inside the observation chamber (Materials and Methods). Once the string had been pulled, an untrained forager was released into the arena to feed from the now accessible flower. Without direct demonstration of string pulling by a bumblebee forager, none of the observers managed to solve the string pulling task. Nonsocial observers mostly tried to obtain the reward from the top of the table, indicating that the bees need to observe string pulling actions demonstrated by conspecifics to learn the technique. However, because no video material is available to show that observer bees directed their gaze towards the moving flower, it is also possible that in the absence of a conspecific demonstrator, observers simply failed to attend to the movement of the flower. Finally, because smaller bees might be able to reach further under the table than larger bees, we examined whether body size influenced success in solving the task (Colony 1). Thorax width (as a proxy for body size) was not different between demonstrators (n = 40), observers (n = 25), and untrained bees (n = 25) (ANOVA, F69 = 0. 728, p = 0. 486). Thorax width affected neither demonstrators’ (Student’s t test, t26 = 0. 659, p = 0. 516) nor observers’ success rate (Mann–Whitney U test, U23 = 79, p = 0. 846). Similarly, the latency to obtain the reward was not affected by thorax width of demonstrators (Pearson correlation, r23 = -0. 086, p = 0. 696) or observers (Pearson correlation, r15 = 0. 375, p = 0. 169). What mechanisms were the observers using to copy the behavior? To answer this question, we explored several associative mechanisms: local enhancement [30,41,42], whereby observers are attracted to the location of their conspecific; stimulus enhancement [30,43], an attraction to the item handled by the demonstrator; and perceptual feedback [44,45], a form of trial-and-error learning in which action causing movement of the rewarding object towards the animal produces positive feedback for continuing that action. We found that all three associative mechanisms were involved in the learning of the string pulling process. To examine the local and stimulus enhancement possibilities, we analyzed the video footage to determine the time bees spent in four different regions of the arena (see Fig 3A, Materials and Methods). In Test 2, unsuccessful observers (n = 10, Colony 1) spent more time in the region where the demonstrator was observed (Friedman test, χ23 = 14. 160, p = 0. 003, Fig 3B), and untrained bees (n = 23, Colony 1) spent more time on top of the table closest to the flower (Friedman test, χ23 = 35. 162, p < 0. 001, Fig 3B) than in Test 1, indicating that local enhancement played a part in learning. None of the bees managed to obtain the reward when the string protruded in an area incongruent with that seen during demonstration. However, the string itself also played a role. If the string protruded from a different side of the table compared to the location during the observation period, observer bees (Test 2, n = 14, Colony 4; S9 Video) spent more time exploring the region with the string than the region where the demonstrator had been observed (Mann–Whitney U test, U22 = 105, p = 0. 038, Fig 3C), indicating that observers had noticed the string during the observation period and were thus attracted to it. In theory, however, these longer dwelling times in the string region might be explained by bees randomly exploring the edges of the table and simply stopping at a region that contains any protruding object. To explore this possibility, we also evaluated bees’ first approach flights after being released from the observation chamber before they had a chance of interacting with the string. If the string was in the same location as during observation, 92% of observers flew straight to the side of the string. When the location of the string was incongruent with demonstrator location, only 28. 5% of observers first visited the region where the demonstrator had been observed (where chance expectation is 25%). The choice frequencies for the four sides of the table are significantly different depending on whether the string was in the correct location (Chi-square of fit, χ2 4 = 206. 857, p < 0. 0001), indicating that bees were able to see the string from the observation chamber and responded differently when it was presented in an unexpected location. However, there was no appreciable attraction to the string when its location was at variance with that seen from the observation chamber (28. 5%). Taken together, these results indicate a strong role for local enhancement (bees were attracted to the location where they had observed a demonstrator) and a subordinate role for stimulus enhancement (bees were attracted to the string when its location was concordant with that during prior observation) [25,46]. Finally, trial-and-error learning was also evident in the learning process. Because individuals might only learn where to obtain the reward and then learn the string pulling by trial-and-error, observer bees (n = 27, Colony 5) were tested with a coiled-string paradigm where trial-and-error learning of actions causing the rewarding object moving closer is ineffective. After a standard demonstration of string pulling (Materials and Methods), a 14 cm string was attached to the flower and coiled under the table so that initial tugs on the string would provide no visual feedback of the flower moving closer to the bee. Such coiled-string tests have in the past been used to test whether animals can solve a string pulling puzzle by means-end comprehension, without the perceptual feedback of the reward coming closer [44,45]. Long-tailed macaques (Macaca fascicularis) [47] and wolves (Canis lupus) [48] have indeed been shown to solve the task even if the string is coiled. However, none of these observer bees were able to solve this task (n = 27, Fig 2B, S10 Video), indicating that observers did not glean information about the string pulling technique itself by observing a demonstrator but instead were merely guided to the demonstrator’s previous location (by local enhancement) and the position of the string (stimulus enhancement). The actual act of string pulling relied on individual trial-and-error learning, which in turn necessitates the sensory feedback of tugging on the string, resulting in the target moving closer. We also tested eight experienced individuals (with an experience of more than 20 instances of string pulling) with the coiled-string test; three of these bees succeeded in pulling the coiled string to obtain the reward (S11 Video), indicating that highly experienced individuals do not necessarily require the feedback from seeing the flower move closer while they pull the string. In summary, these results suggest that observational learning of the string pulling task does not involve the “understanding” of the task (“insight”) but the combined use of several simple associative mechanisms and trial-and-error learning. Can the combination of multiple simple social learning mechanisms mediate the establishment of a culture-like phenomenon (e. g. group-specific behaviors, such as foraging techniques, that are transmitted via social learning and retained in the group over long periods)? We tracked the diffusion of an experimentally introduced string pulling behavior among foragers of test colonies (Colonies 6,7, 8) to explore the speed of diffusion and also the retention of the technique in the group beyond the demonstration provided by the first knowledgeable individual. To seed the technique, we trained a single demonstrator per colony to pull the string. Subsequently, we allowed pairs of bees to engage with the string pulling task and tracked the diffusion of the technique among the foraging population (Materials and Methods, Fig 4). Pairs of bees were tested in the order in which they arrived in the corridor connecting the hive to the arena; pairs could be any combination of bees regardless of whether they were naïve, the seeded demonstrator, or a successful learner (S12 Video). As a control, foragers of three separate colonies were tested in the same manner without a seeded demonstrator (Colonies 9,10,11). After only 150 paired foraging bouts, a large proportion of each of the test colonies’ forager population (Colony 6: n = 25/47, Colony 7: n = 17/29, Colony 8: n = 12/28) learnt to string pull, whereas none of the control colony foragers (Colony 9,10,11: n = 51,58,57) learnt to pull the string (Fig 5, Materials and Methods, S13–S18 Videos). We conducted additional foraging bouts in two of the tested colonies and found that the technique continued to spread among the foragers for as long as we allowed the spread to progress (Colony 6: 34/47, Colony 8: 18/28, Fig 5, S13 and S15 Videos). We quantified the behavioral changes in learner bees over the time of the diffusion experiments. We first screened 81 of the total 419 available videos (~20%) of the paired bouts between demonstrators and learners and inventoried the repertoire of behavioral interactions. We listed 11 types of interactions (Table 1), the frequency of which changed with increasing experience of the learners (Fig 6). Behaviors went through a series of steps with increasing competence, which typically followed the following sequence. During an observer bee’s first few bouts, she would spend most of her time flying around the arena, occasionally landing on top of the table (NI, No Interaction) and spend little or no time near the table, strings, or the other bee. She would gradually start to land beside a bee who had already pulled a string for reward, thereby gaining reward without pulling a string (Sc, scrounging). The observer thus learns to associate the other bee with reward and typically begins following her around the table, keeping in close contact as they both walk (Fo, following). After one or more occurrences of scrounging, the observer bee would begin to reach under the table, sometimes extending her proboscis towards the flower, seemingly in an attempt to gain access to the flower without manipulating the string. While moving around the edge of the table and trying to reach under it, the observer bee might accidentally move a string, but make no subsequent effort to continue moving it (AMS, Accidentally Moving String). Often the observer bee would then position herself next to the bee already pulling a string. She would be in direct contact with the string pulling bee throughout the pull, usually not touching the string (A, Attending), although in some instances ineffectively manipulating the string (STA, String Touching while Attending), and ultimately gaining reward through the other bee’s efforts. Eventually, while in direct contact with a more knowledgeable bee, the observer bee would pull the string, but not enough to move the flower close enough to the edge of the table, extract it, and obtain the reward (PA, Pulling Action with demonstrator). In this phase, she would still rely on the efforts by the more experienced bee to obtain the reward (RP, Rewarded Pull). After more experience, the observer bee would attempt to pull the string on her own without interacting with the other bee, for example, while the demonstrator was flying around the arena. On the first few attempts to string pull on their own, the observer bees did not move the flower enough to be able to obtain the reward (PAa, Pulling Action alone). Finally, after few unrewarded attempts, and typically when paired with a less knowledgeable bee, the observer bee would learn to pull the string on her own to the point of extracting the flower from underneath the table and gaining reward (RPa, Rewarded Pulling alone) and become a trained observer. These changes in behavior are reflected in the relative frequencies of behavior classes as a function of experience (Fig 6). Whilst nonsocial interactions such as NI and Sc represented more than 55% of the interactions at the onset of the diffusion experiment, they decreased rapidly to 0% over time (Fig 6). In comparison, the percentage of pulling actions displayed by the learners continuously increased with experience from 15% of the interactions at the onset to 60% after 11 bouts. Overall, no major change was observed for the other behavior classes. These results show that learners progressively changed their foraging behaviors from scroungers to competent string pullers. In test colonies, on average 2 ± 0. 06 string pulls were performed per foraging bout and 20 ± 3. 9 pulls were displayed per individual over the whole diffusion experiment. Bees needed to be shown 5 ± 0. 45 instances of string pulling by an experienced demonstrator before being able to pull the string themselves without demonstration and subsequently demonstrate the technique. Notably, 15 of 104 foragers (Colony 6,7, 8: n = 10,3, 2, respectively) picked up the technique very rapidly after only one or two observations. There was a significant variation between tested colonies in the average number of string pulls displayed per bee (Colony 6,7, 8: n = 13 ± 4. 7,15. 4 ± 9. 2,34. 5 ± 7. 6, respectively; Kruskal–Wallis test, H2 = 8. 790, p = 0. 012) and the number of observations necessary for a bee to learn the technique (Colony 6,7, 8: n = 4. 1 ± 0. 4,7. 6 ± 1. 1,5. 9 ± 0. 9, respectively; Kruskal–Wallis test, H2 = 17. 179, p ≤ 0. 001). In addition, some bees did not manage to acquire the technique despite having been shown the same number of string pulling by other bees (5. 6 ± 0. 7; Mann–Whitney test, U93 = 1075. 5, p = 0. 261). These results suggest colony and individual variation in social learning ability. To determine whether experience of the second bee influenced the observer bee’s choice of string to pull, we analyzed the pulling behavior of 25 randomly selected observer bees over the complete sequence of their foraging career during the diffusion experiment (282 paired foraging bouts). We found that observer bees more often pulled the same string as the other bee when paired with a more experienced observer bee or the seeded demonstrator (42 RP instances) than when paired with a less experienced bee (9 RP instances). In contrast, observer bees more often pulled a string alone when paired with a less experienced bee (72 RPa instances) than when paired with a more experienced observer bee or a seeded demonstrator (27 RPa instances). To test whether bees might cooperate during string pulling, we needed to compare whether experienced bees performed more efficiently when paired with another experienced individual than when foraging alone. Because the diffusion experiment contained only trials with dyads of foragers, the only way to make a direct comparison was to use trials in which an experienced demonstrator was paired with a fully naïve individual that had not shown any pulling action (PA, PAa, RP, or RPa) and thus did not interact or interfere with the skilled forager, who pulled the string singly. Such pairings were compared with instances where both bees were experienced (had already displayed a pulling action). We hypothesized that if cooperation was occurring, strings would be pulled faster and reward obtained quicker in such dyads. However, when paired with an experienced bee, demonstrators (n = 16 randomly chosen individuals) took 2. 5 times longer to pull the string and obtain the reward (39. 9 ± 9 s) than when the same individual demonstrators were paired with an experienced observer who did not interact or interfere with them (15. 6 ± 2. 1 s; Wilcoxon test, Z30 = 3. 409, p < 0. 001). These results suggest that bees do not cooperate to pull the string but in fact hinder each other’s efforts to some degree. Of particular interest for culture-like phenomena is the question of whether a socially learnt behavior routine persists in the population for longer than the original knowledgeable individual can serve as a demonstrator, so that former observers can themselves become demonstrators. If this is the case, then group-specific behavior routines can at least potentially be retained over biological generations. Our network analysis indeed indicates that the technique spread across sequential sets of learners, whereby some bees that learnt the technique never interacted with the seeded demonstrator. In fact, despite the death of the seeded demonstrator in one of the test colonies (Colony 6) after 58 paired foraging bouts, the technique continued to spread among foragers. Moreover we found that there were up to four sequential learning “generations” (as opposed to true biological generations) in two of the three colonies (Fig 5). Learners had string pulling demonstrated to them by up to eight different demonstrators (2. 1 ± 0. 13), and each demonstrator displayed the technique to 5. 3 ± 0. 93 learners. Overall, seeded demonstrators displayed eight times more string pulling (119. 7 ± 26. 5) than the other foragers (14. 6 ± 3) (Mann–Whitney, U68 = 4, p = 0. 004) and demonstrated the technique to five times more foragers (19 ± 2. 8) than the other foragers (4. 2 ± 0. 7) (Mann–Whitney, U36 = 2, p = 0. 006). This preponderance of the pretrained demonstrators could be a result of higher motivation simply because they obtained reward with every bout, whereas untrained bees often (in the beginning of the experiment) were unrewarded (i. e. , unsuccessful until they were paired with a demonstrator or until they learned to pull the string themselves). To test whether string pulling was diffused socially, we performed network-based diffusion analysis (NBDA). We used the time-based approach described by Hoppitt et al [49]. The Aikake Information Criterion (AIC) was used to determine if string pulling was diffused socially by comparing a social and a nonsocial model for each of the diffusion experiments. We found that for all three experiments, social transmission was more likely than asocial transmission (Table 2). We also analyzed the structure of the social networks using exponential-family random graph modeling [50] and found that for all diffusion experiments as well as the control experiments without a demonstrator bee, the structure of the networks was significantly different from random (see Table 3). This indicates that certain bees were more likely to forage together than other bees. Although this could be interpreted as certain individuals preferentially foraging together, given the open-diffusion paradigm and experimental design (in which bees could not freely distribute themselves in space but were forced through the “bottleneck” of the nest entrance tunnel to the foraging arena), this likely reflects temporal factors such differences in when bees began to forage each day, daily changes in foraging activity across bees, and how long each bee takes to return to foraging from the hive. Here, we show that an invertebrate can be trained to solve a string pulling task, and that a minority can even solve such a task without stepwise training or observation of skilled demonstrators. String pulling is a popular problem-solving paradigm to investigate cognitive abilities in vertebrates [45], in part because scientists in comparative cognition have been interested in exploring the limits of animal intelligence and behavioral flexibility by facing subjects with tasks that are outside their natural repertoires [35,36]. Although there are natural analogues to many standard laboratory tests in animal cognition, string pulling is indeed relatively remote from most animal’s daily behavioral routines. There is no question that many animals regularly pull objects (including bees—e. g. , to remove debris or corpses from their nests), but, specifically, the act of object pulling with the purpose of obtaining a food reward, and the learning of such techniques, is not commonly observed in many animals’ daily lives. As one aspect of exploring animal intelligence, string pulling tasks have been used to test the understanding of means–end relationship: the capacity to mentally model the string as a means to reach an end (the reward) and to understand the connection between the string and the reward. However, most animals appear instead to use perceptual feedback to solve string tasks [45]. Our results indicate that bees may not be different from birds, dogs, or apes in this respect. Bumblebees relied on the perceptual feedback provided by their actions, resulting in target movement to learn string pulling, and failed, for example, in an experiment in which the string was coiled, and therefore tugging on it did not result in immediate feedback. However, through experience, bees eventually learnt to associate the string with the reward and solved the task without the need for feedback. Nonetheless, this would not allow bees to solve tasks requiring means–end understanding such as the discrimination of connected and disconnected strings. More than a century of research in social learning in animals has revealed a plethora of evidence that animals, from primates and cetaceans to birds and fish, can acquire novel skills by observing the actions of others [1,3, 5,6, 30,51]. Growing evidence also shows that insects can glean critical information about their environment by observing others [52–54]. Here, we show that uninformed bumblebees can learn a novel and highly unnatural foraging technique, string pulling, by observation. To this end, our bees used a combination of simple forms of learning. Consistent with a previous study [55], during observation, bees were able to pick up the location of a new access to the reward (local enhancement of flower position). In addition, we showed that the observers were attracted to the string (stimulus enhancement). A recent study showed that learning about rewarding flowers from conspecifics resulted in stimulus enhancement, whilst learning from nonsocial or model demonstrators resulted in local enhancement [46]. In contrast, our results suggest that observers can use both forms of information independently of the cue type. In addition, the results of the coiled-string experiment also indicate that trial-and-error learning was involved in learning the technique. That is, when observing the demonstrations, the observers did not learn the specific sequence of actions used by the demonstrator with the string but simply knew to go to the correct location where the string was accessible. They had to learn the technique of how to move the target (blue flower) closer by individual exploration. These results suggest that the combination of relatively simple forms of social learning and trial-and-error learning can mediate the social learning of novel skills [19,29,56]. In this sense, our study adds to the growing evidence that simple principles of “asocial” associative learning can also account for many aspects of social learning [29,53]. For instance, observational learning about flower colors in bumblebees can emerge through the simple Pavlovian ability to integrate two learned associations (second-order conditioning) [28]. This mechanism has more commonly been explored in nonsocial learning and is also common to social and solitary species [28,57]. Moreover, social learning ability and asocial learning ability covary across and within species [57]. Overall, this suggests that social and asocial learning are mediated by the same “generic” mechanism [29,57]. The use of generic mechanisms in learning generates the possibility to combine different forms of learning, allowing bees to use local as well as stimulus enhancement and trial-and-error learning to learn string pulling by observation. Even if social and asocial learning rely on common associative learning mechanisms, the sensory filters that allow animals to recognize conspecifics can guide attention of observers to valuable resources [58]. This may explain why observer bees were not able to solve the task in the “ghost experiment. ” Without the presence of a visible demonstrator, the motivation or the attention paid by the observer to the flower movement and the string may not have been sufficient to learn the critical information required to solve the string task [57], or, indeed, bees may not have paid attention to the action of the moving flower. Consistent with this, a recent report on learning in bees suggested that the specific attention directed to mobile salient cues provided by conspecifics could explain the dissociation of social and asocial learning [53]. The spread of novel foraging techniques has often been viewed as evidence that animals have the basic cognitive tools needed for cultural transmission of skills [34,41,59]. From an evolutionary perspective, culture should allow for the passing along of advantageous information through generations of learners. Here, we show that a novel, experimentally seeded foraging technique can spread through social learning by observation in bee populations. Moreover, we report that the novel routine persisted in the population for longer than the original knowledgeable individual served as a demonstrator, so that successive learning “generations” became demonstrators. Though these sequential sets of learners are not true biological generations, these results indicate that, just as in birds [60] and mammals [61], an experimentally introduced innovative behavior can spread via cultural transmission in social insect groups and potentially be retained over long periods. Together with recent work documenting social learning in fruit flies [62], our results suggest that insects possess the essential cognitive elements for cultural transmission. It may be asked what the natural relevance of our findings is, or whether there is likely to be a natural analogue of the cultural transmission of a unique foraging routine as we have described in bumblebees. This may be unlikely to be the case, but our results indicate that this is a question of opportunity rather than a question of whether or not bumblebees have the cognitive toolkit to exhibit culture-like processes. We found that when the appropriate social and ecological conditions are present, culture can be mediated by the use of a combination of simple forms of learning [28,63]. Thus, cultural transmission does not require the high cognitive sophistication specific to humans, nor is it a distinctive feature of humans. It may well be that the absence of such cultural transmission phenomena in bees and other animals in the wild simply reflects the absence of natural opportunities. For example, the spread of the milk bottle opening “culture” in some British songbirds in the 20th century only arose because humans created a “niche” for the behavioral spread of this technique to become useful (by depositing accessible milk bottles on doorsteps of households hours before they were collected) [6]. If a majority of United Kingdom households presented artificial flowers that require unusual manipulation techniques to pollinators, especially at times of dearth, it is equally conceivable that these manipulation techniques would spread socially through pollinator populations. If nature presented such challenges, and if bee foraging activities were not discontinued during winter months, it is clear from our work that bees have the learning capacities to affect long-term, group-specific behavior patterns. More sophisticated forms of social learning and cognitive mechanisms specific to human culture may well have evolved from simpler forms of learning and cognition as described here. Human culture exhibits unparalleled complexity and diversity, and is unambiguously cumulative in character [21,64]. The combination of high-fidelity transmission (e. g. , via imitation, teaching, language) of beneficial modifications of cultural knowledge with the ability to identify “who knows” these modifications with greater accuracy and precision by metacognitive representation promotes cumulative culture in humans [63]. Despite the obvious differences between humans and other animals, understanding social learning and culture in animals holds a key to understanding the evolutionary roots of the peculiarities of social learning and culture in humans. It is clear from our study and others on cultural diffusion in animals that once experimenters create the conditions under which such diffusion is beneficial (often via allowing access to desirable nutrition via man-made devices that must be operated in specific ways), they can be instantly observed in many animals. Early tool-using hominids are likely to have created the conditions for themselves that favored the further evolutionary fine-tuning of social learning processes that results in high-fidelity transmission and cumulative culture [21,64]. Our findings add to the accumulating evidence suggesting that the capacity of culture may be within most animals with a relatively basic toolkit of learning processes as described here, in turn shedding light on the evolutionary precursors of the more sophisticated forms of culture in humans. Bombus terrestris foragers from 11 colonies obtained from a continuous rearing program (Biobest, Belgium N. V.) were used for the experiments. Bumblebee nests were kept in 40 × 28 × 11 cm bipartite wooden nest boxes. Colonies were provided with 7 g commercial pollen (Koppert B. V. , The Netherlands) every 2 d. Through a Plexiglas corridor (25 cm length, 3. 5 × 3. 5 cm in cross-section), bees were allowed access to a flight arena (100 × 75 × 30 [height] cm) where they were trained and tested. Three plastic sliding doors located along the corridor allowed controlled access to the arena. Before training and tests, all the bees were pretrained to associate blue artificial flowers (3 cm diameter blue discs with an inverted Eppendorf cap at the center) with the reward (30% sucrose solution, w/w). Pretraining consisted of bees foraging freely for 1 h on a patch of six blue artificial flowers (ad libitum reward, Step 0, Fig 1A) randomly located in the arena. This phase allowed for the experimenters to identify regular foragers that could be used in individual training. Training and tests were conducted between 9 a. m. and 7 p. m. under standardized light (12: 12, high-frequency fluorescent lighting [ (TMS 24F) lamp with HF-B 236 TLD (4. 3 Khz) ballasts, Phillips, Netherlands, fitted with Activa daylight fluorescent tubes, Osram]) and temperature (25 ± 2°C) conditions at the Bee Behavioural and Sensory Ecology Laboratory (Queen Mary University of London). Five colonies (1–5) were used, two for the string pulling acquisition experiment (Colonies 1 and 2), two for the social learning experiment (Colonies 1 and 3), and three to explore the mechanisms of social learning in string pulling (Colonies 1,4, and 5). In these experiments, bees were allocated randomly to the demonstrator, observer, or untrained group, and individuals were never used in different groups. We trained bees individually to string pull on one day and then used the trained individuals to demonstrate the technique to observers on subsequent days. Six different colonies were used for the cultural diffusion experiment (Colonies 6–11). In this experiment, we chose to train the bee that seemed to forage with regularity to seed string pulling in tested colonies. The other foragers were by default observers and became demonstrators once they learnt the technique. At the end of a training or testing day, bees were again allowed to freely enter the arena to forage from six blue, openly accessible artificial flowers (ad libitum reward) for 1 h. After testing was complete, tested bees were freeze-killed. To examine whether size influenced success, measurement of the bee thorax width were taken with an electronic digital caliper (NewOctave Global, Astoria, United States, precision of ±0. 02 mm). We tested bees in pairs in an arena set up with four artificial flowers with strings (Fig 4). Colonies 6–8 were each seeded with a single demonstrator, whereas colonies 9–11 only included untrained foragers. This is an “open diffusion” design [5], insofar as forager pairings were left open and not constrained by the experimenter. Such open diffusion experiments more closely simulate natural foraging conditions than alternatives such as highly constrained linear transmission chain designs [39]. We briefly explored a fully “open” design that allowed unlimited foragers into the flight arena, but this resulted in a “frenzy” of multiple foragers piling on top of each other near the artificial flowers, and this did not allow us to monitor which individuals learnt from which demonstrators. Therefore, the only constraint upon the openness of the diffusion was that we limited the maximum number of individuals that entered the arena to two, on a first-come, first-serve basis: the first two individuals that entered the tunnel leading to nest, irrespective of these foragers’ identities and prior information, were allowed into the experimental arena. Upon release, the pair of bees was given 5 min to solve the task. We videotaped each foraging bout and recorded whether individuals pulled the strings and drank from the flowers. We tested 150 paired foraging bouts per colony (Colonies 6–11) and tracked the diffusion of string pulling behavior among the foragers. In two of the tested colonies (Colonies 6 and 8), we conducted 95 and 39 additional foraging bouts to assess whether the technique would continue to spread. We mapped the diffusion of the technique on a social network created using a customized version of the R package ggnetwork (version 3. 2. 2, Fig 5). We conducted a second-by-second video analysis of 81 bouts (four randomly selected bees per test colony) to inventory the behavioral interactions between learners and demonstrators and make the ethogram. Table 4 summarizes all treatments, sample sizes, and success rates.

Social insects make use of simple mechanisms to achieve many seemingly complex behaviors and thus may be able to provide a unique resource for uncovering the basic cognitive elements required for culture. Here, we first show that bumblebees can be trained to pull a string to access a reward, but most could not learn on their own. Naïve bees learned how to pull strings by observing trained demonstrators from a distance. Learning the behavior through observation relied on bees paying attention to both the string and the position of the trained demonstrator bee while pulling the string. We then tested whether bees could pass this information to others during a semi-natural situation involving several colonies. We found that once one bee knew how to string pull, over time, most of the foraging bees learned from the initially trained bee or from bees who had learned from the trained bee, even after the initial demonstrator was no longer available. These results suggest that learning a nonnatural task in bumblebees can spread culturally through populations.

Abstract Introduction Results Discussion Materials and Methods

learning invertebrates plant anatomy medicine and health sciences legs sociology social sciences limbs (anatomy) neuroscience learning and memory animals plant science cognitive psychology animal behavior zoology bees foraging musculoskeletal system hymenoptera behavior insects culture arthropoda bumblebees flowers psychology anatomy biology and life sciences cognitive science organisms

2016

Associative Mechanisms Allow for Social Learning and Cultural Transmission of String Pulling in an Insect

In interocular suppression, a suprathreshold monocular target can be rendered invisible by a salient competitor stimulus presented in the other eye. Despite decades of research on interocular suppression and related phenomena (e. g. , binocular rivalry, flash suppression, continuous flash suppression), the neural processing underlying interocular suppression is still unknown. We developed and tested a computational model of interocular suppression. The model included two processes that contributed to the strength of interocular suppression: divisive normalization and attentional modulation. According to the model, the salient competitor induced a stimulus-driven attentional modulation selective for the location and orientation of the competitor, thereby increasing the gain of neural responses to the competitor and reducing the gain of neural responses to the target. Additional suppression was induced by divisive normalization in the model, similar to other forms of visual masking. To test the model, we conducted psychophysics experiments in which both the size and the eye-of-origin of the competitor were manipulated. For small and medium competitors, behavioral performance was consonant with a change in the response gain of neurons that responded to the target. But large competitors induced a contrast-gain change, even when the competitor was split between the two eyes. The model correctly predicted these results and outperformed an alternative model in which the attentional modulation was eye specific. We conclude that both stimulus-driven attention (selective for location and feature) and divisive normalization contribute to interocular suppression. The perception of a brief target stimulus presented to one eye is suppressed by the simultaneous presentation of a dissimilar competitor stimulus to the other eye. This phenomenon, called interocular suppression, can be so strong that it renders the (otherwise easily visible) target invisible [1–3, reviewed in 4]. It has been hypothesized that normalization [5,6] contributes to the neural processing underlying interocular suppression [7–10], in which the responses of a neuron tuned to the target are divided by the responses of a population of other neurons (the normalization pool), including those that respond to the competitor. The competitor increases the responses of the normalization pool which suppresses responses to the target, similar to scaling the contrast of the target (i. e. , changing the contrast gain of neurons that respond to the target). There is also evidence that interocular suppression depends on attention, along with normalization. Ling and Blake [11] measured the detectability of a small monocular target in the presence of dichoptic competitors of different sizes. They found changes in behavioral performance implying that a large competitor induced a change in the contrast gain of neurons that responded to the target, whereas a small competitor induced a response-gain change (a change of the asymptotic response). The dependence on competitor size was analogous to that observed when manipulating spatial attention [12,13]. Using an adaptation method, Ling and Blake further inferred that the response-gain modulation of the small competitor was absent when attention was withdrawn from the stimuli. Thus Ling and Blake [11] concluded that attention plays a major role in modulating competition in interocular suppression. Whereas it has been proposed that attentional modulation plays a critical role in interocular suppression [11], it is unknown what type (s) of attention modulates the competition between the target and the competitor. Attention helps prioritize behaviorally relevant stimuli. In this study, we considered the role of spatial–, feature—and eye-based attention in interocular suppression. Attention can be driven by two different sources: goal driven—voluntary attention in response to task instructions that has a sustained effect (also known as endogenous attention), and stimulus-driven—involuntary attention in response to the abrupt onset of a salient stimulus (known as exogenous attention when elicited transiently by a brief stimulus). In this study, we used the terms goal driven and stimulus driven, instead of the terms endogenous and exogenous, because the stimulus was presented at fixation and its duration was not brief (1. 5 s). When being deployed, attention can boost information at a given location (spatial attention [13–18]) or modulate the sensitivity of the sensory channels selective for relevant features; e. g. , one of several orientations, motion directions, or colors (feature-based attention [19–25]). In addition, some studies have reported that, unbeknownst to the observer, attention can be eye specific [26,27]. We investigated the neural processing underlying interocular suppression using a combination of computational modeling and psychophysics. Our empirical results supported a model in which both attentional modulation and divisive normalization contributed to interocular suppression, and in which the effects of attention and normalization were clearly distinguishable from one another. Attention, in the best-fit model, was stimulus-driven, selective for both feature (orientation) and spatial location, but was not eye specific. Divisive normalization, on the other hand, can account for the change of contrast-gain modulation when the eye-of-origin of the stimuli was manipulated. The experiment was conducted with the written consent of each observer and the experimental protocols were approved by the University Committee on Activities involving Human Subjects at New York University. Four observers (1 female and 3 males) participated in the main psychophysics experiment. All observers had normal or corrected-to-normal vision. Stimuli consisted of band-pass filtered noise patches (1–6 cpd, 10° orientation bandwidth centered at 45° tilted clockwise from vertical), with a target presented to one eye (Fig 1, top row) and an orthogonal competitor presented to the other eye (Fig 1, middle rows; small, medium, and large competitors) or split between eyes (Fig 1, bottom row; split competitor). We chose the parameters to match those used in the previous study by Ling and Blake [11]. Target contrast varied from trial-to-trial in a randomly shuffled order, and the competitor contrast was fixed on every trial (23% RMS contrast). The different competitor configurations were interleaved in a randomly shuffled order within each block of trials. Each observer performed at least 50 trials for each of 45 conditions (a combination of nine target contrast levels and five types of competitors). The target and the small competitor were 1. 5° in diameter, and the medium and large competitors were 2. 5° and 8°, respectively. The split competitor (Fig 1, bottom row) had the same overall size as the large competitor except that it was segmented into two regions: the center (same size as the small competitor) was presented to the competitor eye while the surround was presented to the target eye. The two subregions of the split competitor combined as a uniform large stimulus when fused. The apertures of all the stimuli were smoothed with a Gaussian. The fixation point and a black circular fusion frame (9° x 9°) were presented to both eyes throughout the experiment to stabilize the alignment of the images presented to the two eyes. Stimuli were presented on a calibrated CRT monitor (75 Hz) positioned 57 cm from the observers. Observers viewed the screen through prism glasses, and a black cardboard septum ensured that the stimuli presented on the left half of the monitor were visible only to observer’s left eye and vice versa. Observers performed an orientation-discrimination task on the target (Fig 2). Each trial started with the target presented monocularly (to the target eye) for 2 s. The competitor was then added to the other eye, or to both eyes (for the split competitor). Typically, the competitor was dominant for a period of time following its onset [1]. One second after competitor onset, the target orientation changed either clockwise or counter clockwise (4°) for 500 ms and then both the target and the competitor disappeared from the screen. Observers reported the orientation change, clockwise or counterclockwise, by pressing one of two buttons. Psychometric functions—discriminability (d' ) versus target contrast—were measured for each observer and for each competitor configuration. This procedure, also known as onset flash suppression, allowed us to measure the visibility of the target when the target was suppressed and the competitor was dominant. Presenting the target and the competitor simultaneously (with simultaneous onset) and briefly, as in a masking experiment, would lead to a fused percept of the target and the competitor [28], unlike the strong competition of the two images observed in our experiment. These parameters and procedures followed those in Ling and Blake’s experiments [11] except for two differences. First, we used a shorter interval (1 s) between the onset of the competitor and the orientation change of the target than the 2 s interval they used. In a pilot study, we found that the target reappeared frequently during the longer 2 s interval (unlike Ling and Blake, perhaps due to individual differences in temporal dynamics of interocular competition), which made it difficult to measure the suppression induced by the competitor. Second, we added the split competitor configuration which was critical for distinguishing between two alternative models. In the main experiment, we presented the target to the observer’s right eye (the target eye) and the competitor, except the surround of the split competitor, to the observer’s left eye (the competitor eye), across all conditions. Two observers participated in additional experimental sessions in which three conditions (no competitor, large competitor and split competitor) were tested and the target eye and competitor eye were swapped: the target was presented to the left eye and the competitors (except the surround of the split competitor) were presented to the right eye. All the experimental procedures and parameters were the same as those used in the main experiment. Psychometric functions were fit with Naka-Rushton functions to evaluate how different competitors influenced the visibility of the target: d′ (c) =d′mcncn+c50n (1) where d' (c) was the behavioral performance as a function of target contrast, d' m was the asymptotic performance at the highest contrast, and c50 was the semi-saturation constant determining the contrast level at which d' reached half the asymptotic performance. A change in c50 represented a contrast-gain change, and a change in d' m indicated a response-gain change. The exponent n determined the slope of the function, and was constrained to have the same value for all competitor configurations. Allowing the slopes of the psychometric functions to vary across conditions yielded similar results that supported the same conclusions: the slopes were statistically indistinguishable across conditions, and the improvement of the goodness of fit was very subtle (R2 improved by only 1. 3%, on average across observers, and had a cost of adding five free parameters). The statistical significances of the changes in contrast gain (c50) and response gain (d' m) were evaluated by a bootstrapping procedure. For each observer, we randomly resampled individual psychophysical trials with replacement to generate a bootstrapped data-set. Psychometric functions in the bootstrapped data-set were refit by Naka-Rushton functions. We then computed the differences between the group-averaged (across four observers) c50 and d' m values obtained in the different competitor conditions. This procedure was repeated 2000 times to test whether the differences in c50 and d' m values deviated significantly from zero. The statistical test and the confidence intervals of the estimated parameters for individual observers were obtained by the same procedure. We developed models to simulate the responses of a population of neurons to the stimuli used in the behavioral experiments (Matlab code is available on our website: http: //www. cns. nyu. edu/heegerlab/). Each model had two neural populations: one preferred the stimuli presented in the left eye and the other preferred the stimuli in the right eye (Fig 3 and Table 1). For each of the neural populations, we simulated a 2-dimensional array of neurons, with orientation preferences uniformly sampling orientations in steps of 1° and receptive field centers ranging from -20° to 20°. The responses of the neurons (R) were determined by three components: the excitatory drive (E), the suppressive drive (S) and the attentional gain factors (A) [12]. Responses of the left-eye monocular neurons were computed by the following equation (the details of each component in the equation are listed in Table 1): RL (x, θ) =[Ax (x, θ) Av (x, θ) EL (x, θ) n]/[SL (x, θ) +wISR (x, θ) +σn] (2) where x and θ represented the receptive field centers and preferred orientations of the neurons in the population, n was the exponent that controlled the slope of the neural contrast-response functions, wI was the interocular normalization weight, and σ was a constant that determined the semi-saturation contrast. The value of wI determined the contribution of inputs from the other eye to the normalization pool. The value of σ determined the contrast at which the neural responses saturated. The current model and parameterization only generated neural responses in positive numbers. We did not implement rectification or thresholding in Eq 2 because we did not simulate the effect of spiking threshold or neural responses below the baseline (zero in our case). The excitatory drive was determined by each neuron’s receptive field center and preferred orientation. The spatial excitatory field of each simulated neuron was a Gaussian with 1. 5° standard deviation and the orientation tuning was a Gaussian with 48° FWHM (full width at half maximum), approximating the values reported for macaque primary visual cortex [29]. The suppressive drive was computed by pooling the excitatory drives of neurons with a range of receptive field centers and orientation preferences (See suppressive drive and suppression kernel in Table 1 and Fig 1). Suppression was broadly tuned (all orientations) for stimuli within a neuron’s receptive field whereas surround suppression was narrowly tuned to the neuron’s preferred orientation, mimicking electrophysiological findings [30,31]. To compute the neural responses, the excitatory drive of each neuron was multiplied by its attentional gain factor [12] and then divided by its suppressive drive [5,6, 12,32]. Each neuron in the population had its own attentional gain factor that depended on the attentional state (i. e. , the attended spatial position and the attended orientation) in particular experimental conditions. Two sources of attentional modulation were considered: Ax was the stimulus-driven attentional modulation induced by the competitor and Av was the goal-driven attentional modulation induced by the demands of the orientation-discrimination task. According to the model, both attention and normalization contributed to interocular suppression. First, the presence of the competitor induced stimulus-driven attentional modulation increasing the attentional gains of neurons tuned to the competitor and reducing the gains of neurons tuned to the target. Following Ling and Blake [11], this attentional modulation was responsible for the shift from a response-gain change to a contrast-gain change with increasing competitor size. Second, the competitor contributed directly to the normalization pool, weighted by the interocular normalization weight (wI in Eq 2). This interocular normalization caused contrast-gain changes for all competitors, regardless of their size. To fit the simulated neural responses to the psychophysics data, performance accuracy, d' , was assumed to be proportional to the response of the neuron that was most responsive to the target (eye-of origin, orientation preference, and RF center). That is, we assumed additive, independent, and identically distributed (IID) noise [42]. We used a free parameter σn, representing the magnitude of the noise, to relate behavioral performance (d' ) to the underlying neural responses. A change in response gain of the underlying neuronal responses thereby yielded a proportional scaling of the psychometric function and a change in contrast gain of the underlying neuronal responses yielded a proportional horizontal shift (on the log contrast axis) of the psychometric function. This interpretation of our behavioral results depended on the assumption of additive IID noise. However, given that the neurons homogeneously represent the sensory parameters and the tuning curve is invariant to contrast, an alternative model with Poisson noise and a maximum-likelihood decision rule would yield the same results; the performance (d' ) of an orientation decoder is proportional to the mean firing rate of the neuron within the neural population that is tuned for the mean stimulus orientation [43,44]. For both the FS and ES models, there were seven free parameters (See Table 2), and the MATLAB fmincon function was used to search for the best (least-squares) fit. All seven free parameters were fitted across all the conditions. The values of wx, wv and p were constrained so that the attentional gain factors (Ax and Av) were non-negative. The only parameter that changed with different conditions was σαx, which controlled the spatial extent of spatial attention (see Table 1). Both FS and ES models assumed that the spatial spread of stimulus-driven attention was determined by the competitor size so we set σαx to be the same as the width of the competitor (i. e. , σαx was not a free parameter in the model fit). Confidence intervals for each of the free parameters were estimated by a bootstrapping procedure. Specifically, we randomly resampled individual psychophysical trials with replacement to generate a resampled data-set, which was refit with each of the models, and this procedure of resampling and refitting was repeated 2000 times to generate bootstrap distributions of the best-fit parameter values. We used cross-validation for model comparison. The raw data (a list of trials from each individual observer) were permuted and partitioned into a training set and test set, and the group-averaged psychometric functions were then computed for both the training set and the test set. Each model was fit to the training set to determine the best-fit parameters. These parameter values were then compared with the test set, i. e. , by computing the coefficient of determination (R2) which represented the goodness-of-fit of each model to the test set. The coefficient of determination was computed for the FS model (RFS2) and for the ES model (RES2), and the difference between the two values (RFS2−RES2) was taken as an index for model comparison. This procedure was repeated 2000 times to obtain a distribution of RFS2−RES2 values. The FS model was considered to outperform ES model if this distribution was significantly (95% of the distribution) larger than zero, and the ES model was considered as the better model if the distribution was significantly smaller than zero. The model fit and the model comparison were performed on the group averaged data consisting of 11079 trials pooled across observers and conditions. In a complementary analysis, we also used maximum-likelihood estimation and Bayesian information criterion to compare the models (Supporting Information, S2 Text). To assess whether there was parameter redundancy, we numerically computed the Hessian matrix (i. e. , the second derivatives with respect to the model parameters) of the best-fit model, and used singular value decomposition to compute the rank of the matrix. A full-rank Hessian matrix indicated that the parameters in the model were not redundant [45]. To investigate the role of attention and normalization in interocular suppression, we developed computational models to simulate the responses of a population of neurons, and fit the models to published psychophysical measurements [11]. Ling and Blake [11] reported changes in behavioral performance implying that a large competitor induced a change in the contrast gain of neurons that responded to the target, whereas a small competitor induced a response-gain change. Two models were considered as candidates. In the feature-specific (FS) model, the onset of the competitor induced stimulus-driven attentional modulation that increased the gain of neurons that preferred the competitor orientation, and reduced the gain of neurons that responded preferentially to the target orientation. In the eye-specific (ES) model, stimulus-driven attentional modulation increased the gain of neurons that responded preferentially to inputs from the eye to which the competitor was presented, and reduced the gain of neurons that responded to the target eye. In both models, the stimulus-driven attentional modulation was complemented by goal-driven attentional modulation that increased the gain of neurons that preferred the target orientation, and reduced the gain of neurons that preferred the competitor orientation. This goal-driven attention component was task-specific, enabling the simulated observer to discriminate the target orientation more accurately. Divisive normalization also contributed to interocular suppression in both models. Both models were able to explain the suppression induced by the competitor in Ling and Blake’s data (Fig 4). The models were fit to the data to determine best-fit values for the model parameters including the magnitudes of the stimulus-driven and goal-driven attentional modulations, and the interocular normalization weight (Table 2). There was no evidence for a difference between the goodness of fit of the two models (p = 0. 45; see Materials and Methods, Full model statistics). According to both models, behavioral performance depended on competitor size because the onset of the competitor induced stimulus-driven attentional modulation with a spatial extent determined by competitor size (Fig 4C and 4D). The parameter p (Table 2) represented the tradeoff between the magnitude and spatial extent of the stimulus-driven attentional gain factors. A value of p close to 1 would have indicated that the sum of the attentional gain factors was constant when size varied, i. e. , a complete trade-off between size and magnitude. The best-fit values were 0. 17 and 0. 31 for the FS and ES models, respectively, indicating a partial trade-off between size and magnitude. Due to this tradeoff, in addition to the spatial extent of attention, the values of the (stimulus-driven) attentional gain factors also changed with the size of the competitor (see the values reported in the parenthesis under wx in Table 2). The model fit also showed a significant role of goal-driven attention (wv in Table 2) which increased the gain of neurons tuned to the target orientation and reduced the gain of neurons tuned to the competitor orientation (Fig 4E and 4F). The interocular normalization weight wI was around 0. 8 for both FS and ES models (Table 2), implying nearly equal divisive normalization from stimuli presented to either eye. However, the confidence interval for this parameter value covered a wide range from 0 to 1. There was no evidence of parameter redundancy in the model fits because the Hessian matrices were full rank (see Full model statistics in Materials and Methods). However, the interpretation of interocular normalization weights could be difficult because the eye-of-origin of the competitor was not manipulated in Ling and Blake’s experiment, and the magnitude of the interocular divisive normalization did not solely depend on this parameter but also on the goal-driven attentional modulation which could reduce the gain of the responses evoked by the competitor. The role of interocular divisive normalization and how it influenced the predicted psychometric functions became clearer when the eye-of-origin of the competitor was manipulated in our psychophysics experiment. To distinguish the FS model and the ES model, we replicated Ling and Blake’s (2012) psychophysics experiment and added a critical new condition with a split competitor. The split competitor (Fig 1, bottom row) had the same overall size as the large competitor except that it was segmented into two regions: the center (same size as the small competitor) was presented to the competitor eye whereas the surround was presented to the target eye. The two subregions of the split competitor were perceived as a single large grating when fused. The FS model predicted that the split competitor would cause a contrast-gain change like the large competitor; according to this model, the split competitor induced the same attentional modulation as that driven by the large competitor because the attentional gain factors were blind to the eye-of-origin of the stimuli. The ES model predicted that the split competitor would cause a response-gain change like the small and medium competitors; according to this model, the split competitor induced less attention in the center region of the target eye (because the center of the competitor was presented to the other eye) and more attention in the surround region of the target eye (because the surround of the competitor was presented to the target eye). In addition, we predicted that the split competitor condition would constrain the best-fit value of the interocular normalization weight wI because the large competitor and the split competitor had the same overall size but differed in presentation to the two eyes. In the experiment above, we always presented the target in the observer’s right eye. To further investigate the role of eye-of-origin and its relation with the interocular normalization weight parameter wI, two observers (S1 and S4), who showed very different contrast gain changes in the large- and split-competitor conditions, participated in additional experimental sessions in which the target was presented to the left eye and the competitors were presented to the right eye (except the surround of the split competitor). The no-competitor, large-competitor and split-competitor conditions were tested. The comparison between the modulations induced by the large competitor and the split competitor could reveal the role of eye-of-origin because these two conditions had the same overall (perceived) competitor size, and the only difference between the two competitors was how the competitor was distributed between the two eyes. When the target eye and competitor eye were swapped, both split competitor and large competitor still induced a contrast gain change (large competitor: p<0. 001, split competitor: p<0. 01, bootstrap test) without modulating the response gain (large competitor: p = 0. 69, split competitor: p = 0. 46, bootstrap test) (Fig 7, left column). S1, who originally showed stronger contrast gain modulation in the split-competitor condition, now had stronger modulation in the large-competitor condition (p<0. 05, bootstrap test; Figs 5 and 7). For S4, the large competitor generated stronger suppression than the split competitor in the original experiment, but the suppressive effects of the two conditions became indistinguishable when the target eye and competitor eye were swapped (p = 0. 10, bootstrap test). When fitting the FS model to individual data, wI again reflected these individual differences. For S1, wI changed from 0. 01 (Table 4) to 1. 98 (Table 5). For S4, wI reduced from 3. 81 to 0. 71. The increase of wI indicated stronger suppression induced by the stimulus presented in the competitor eye than in the target eye. The change of wI value in the model fit followed the results that after swapping the eye-of-origin, the large competitor (which was presented entirely in the competitor eye) became more suppressive for S1 and less suppressive for S4. The fitted value of p was larger in Table 5 than that in Table 4. However, the value of this parameter should be interpreted with care. In the model fitting process of the main experiment, we found that the value of p was mainly determined by the amount of response gain reduction observed in the small and medium competitors (for example, a larger trade off between the magnitude of attentional modulation and the spatial spread of attention would let the response gain of the medium competitor be closer to that of the large competitor, and vice versa). Here, we only fit the no competitor, large competitor and split competitor conditions. None of these conditions exhibited response gain modulation, so the data might not have constrained the value of p. To determine if this was the case, we assessed the Hessian matrix of the best-fit model, for each subject. While the Hessian matrices were full rank, the eigenvectors corresponding to the smallest eigenvalues had a large projection (0. 64 for S1 and -0. 85 for S4) on parameter p, implying that this parameter was not as well-constrained as the other parameters. We fixed the parameters related to stimulus-driven attention, wx and p, at the values reported in the main experiment and fit the model again (parenthesis in Table 5). We found that the FS model could still account for the data, and the R2 only decreased by 0. 59% and 1. 14% for S1 and S4 respectively. The change of wI due to swapping the eye-of-origin was in the same direction as the original model fit (increase for S1 and decrease for S4 compared to the main experiment). In the normalization model, the responses of a visual neuron are modulated by both the neuron’s preferred stimulus and the visual context formed by the presence of other stimuli [6]. Several models of visual attention have suggested that the effect of attention can be modeled within the framework of normalization [46–48]. In the present study, we followed the normalization model of attention proposed by Reynolds and Heeger [12]. This model correctly predicts that endogenous and exogenous spatial attention cause contrast-gain and response-gain changes, evident in the psychometric functions, when the spatial extent of attention is manipulated [13], and response-gain changes when the featural extent of attention is manipulated [37]. In the present experiment, the competitor elicited an abrupt visual onset and thus stimulus-driven attention [25,49–52]. According to the model, stimulus-driven attention influences behavioral performance even when it opposes task demands (in this case, monitoring the orientation of the target), and implies that the bottom-up inputs have a considerable impact on the deployment of feature-based attention. This is different from most previous studies, in which feature-based attention was presumed, or manipulated, as a top-down process controlled voluntarily by observers [24,34,39,53]. The stimulus-driven, feature-based attention seems to have properties different from goal-driven feature-based attention. For example, the spatial extent of the stimulus-driven feature-based attention was constrained by the size of the competitor stimulus. In contrast, previous studies have found that goal-driven feature-based attention spreads across space [38–40, also see 54,55]. The partial trade-off that we observed between the magnitude of stimulus-driven attention and its spatial extent is consistent with the idea of limited attentional resources [16,24,56]. One study also found a significant influence of bottom-up inputs on feature-based attention by showing a reduction of reaction time when the pre-cue and the target to be discriminated were in the same color compared to when they were different [25]. The authors reported that this featural effect can spread across space, but only when the cue was presented in one of the potential target locations and not when it was presented in a target-irrelevant location. To explain the results of the present experiment, we infer that stimulus-driven feature-based attention followed the size of the competitor even when it extended beyond the task-relevant target location. The difference between our and Lin et al. ’s [25] results might be due to the discrepancy of the experimental designs between the two studies. The response time measure used by Lin et al. could be influenced by the target’s visibility, the speed of information processing and/or criteria changes [57,58]. On the other hand, in the present study, we focused on the visibility of the target by using d' as the index. Feature-based attention with benefits for attended features and costs for unattended features reported in our model is consistent with previous behavioral [40] and neurophysiological [59] studies. Even though the increases and reductions in attention gain were linked together in our model, we are agnostic as to whether they share a unified mechanism. Multi-unit recordings in FEF and V4 in monkeys [60] and human EEG experiments [61] have shown that such increases and reductions in attention gain occur with different time courses. Our psychophysical results and model comparison showed that the attentional modulation in interocular suppression was better described by a feature-specific rather than an eye-specific modulation. Solely based on our results, we can not conclude that attention is unable to modulate eye-specific information. Previous studies proposing eye-based attention found that manipulating the eye-of-origin of the stimuli [27], or the eye-of-origin of the image to be attentively tracked [26], can influence observers’ performance in a visual task. However, in our experiment and computational model, we also identified a source of suppression, in addition to attention, whose strength depended on the eye-of-origin of the stimuli (see Interocular divisive normalization below). Many studies [2,7, 9,62] have also shown that the magnitude of suppressive interactions between multiple stimuli can be changed by simply manipulating eye-of-origin (presented simultaneously either in the same eye or different eyes). Future studies should establish whether the eye-based attention effects reported earlier are distinguishable from this eye-dependent suppression. The goal-driven component of attention in our model is similar to that in previous models of feature-based attention in orientation discrimination tasks [23,35–37]. Indeed, there is evidence that feature-based attention also modulates interocular suppression [63]. Because goal-driven attention was the same for all of the stimulus conditions, it could not account for the dependence of competitor size on interocular suppression. In preliminary fits to Ling and Blake’s data set, we found that without goal-driven attention the interocular normalization weight wI was forced to take an extremely small value. However, a wI value close to zero predicted that the split competitor should induce a much stronger contrast-gain change than any of the other competitors, contrary to what we observed (Figs 5,6 and 7). Including goal-driven, feature-based attention is, consequently, in accordance with the feature-based attention literature and allows a unified model that can fit the results of both experiments. Other studies have modeled the demands of the task by including a weighting function at the decoding or decision stage of processing [64–66]. We acknowledge that the goal-driven gain factors in our current model might be replaced by multiple stages of information processing, and this could be the reason why the goal-driven gain factors estimated by the model were so strong that they greatly reduced the response of the task-irrelevant orientations (Fig 4E). We found, for some individuals, that moving a portion of the competitor to the other eye (changing the stimulus from large competitor to split competitor) can influence the magnitude of contrast-gain change without influencing the response gain. This pattern held when the target eye and competitor eye were swapped so that the target was presented to the left eye and the competitors were presented to the right eye (Fig 7). These results provided evidence that a source of the suppression from the competitor is eye-specific and it causes a change in contrast gain, consistent with previous neuroimaging [9] and psychophysics [8] studies on cross-orientation dichoptic masking. In the current model, the interocular normalization weight wI reflected individual differences in the magnitude of the contrast-gain change induced by the large and the split competitors: increasing the value of wI caused the large competitor to become more suppressive compared to the split competitor. The individual variation in the magnitude of interocular divisive normalization was consistent with previous studies reporting individual variations in the strength and the temporal dependency of dichoptic masking [7,62]. There seem to be multiple factors influencing the magnitude of wI: Individual differences in eye dominance are well documented for normal observers [67,68]. In the present experiment, if an observer had a significant imbalance in eye dominance, moving a portion of the competitor from the weaker eye to the stronger eye should have resulted in greater divisive suppression of the target. In addition, after swapping the target eye and the competitor eye, the competitor (either the large or the split competitor) that generated stronger suppression should have also switched (to the split or to the large competitor, respectively). We observed this pattern for subject S1, but not for S4 (Fig 7). Thus, the magnitude of interocular normalization can not be solely explained by eye dominance. The model fits (S1 and S4 in Tables 4 and 5) indicated that the interocular suppression weight from left- to right- eye was not equal to the weight from right- to left- eye, and thus wI not only varied across observers but also varied according to which eye was the target eye (and which eye was the competitor eye). The parameter wI can actually be considered to be two separate parameters: wLR in Table 4 and wRL in Table 5, such that wLR represents the strength of interocular divisive normalization contributed by the left-eye competitor to the right-eye neural population, and vice versa for wRL. There has been some controversy as to whether interocular suppression occurs in early visual cortex when attention is controlled. Whereas one study reported an absence of interocular suppression in V1 [69], two studies reported significant interocular suppression even when attention was diverted away from the stimuli [9,10]. According to the current model, both divisive normalization and attentional modulation contribute to interocular suppression. We demonstrated that these two components contributing to interocular suppression can be distinguished. Attentional modulation depended on the feature and size of the competitor resulting in a response-gain change for large competitors and a contrast-gain change for small competitors. Interocular divisive normalization depended on the eye-of-origin of the competitor resulting in contrast-gain changes of different magnitude for the large and split competitors. Previous psychophysics studies [7,8] investigated the suppression induced by dichoptic masks with orientation orthogonal to the target and the same size as the target. It was found that the mask elevated detection threshold of the target and the data were fitted by a model with interocular divisive normalization [7]. When a large range of target contrasts were tested, a contrast gain modulation was reported in a contrast detection task [8]. Such a pure contrast gain modulation is different from the suppression effect reported here. The discrepancy might be due to the fact that the interocular suppression was measured in distinct perceptual states. Our study used onset-flash suppression to ensure that the target was suppressed by the dominant competitors. This is different from the procedures used by Baker et al. [7,8], in which the target and the mask were presented simultaneously and briefly (from 25 ms to 400 ms). These parameters are known to generate a perceptual state of fusion (of the target and the mask) before the initiation of strong interocular competition [28]. Single-cell electrophysiological recordings have found mixed results regarding the suppression induced by dichoptic masks. Some reported predominately response gain modulation [70], and some reported a mixture of contrast gain and response gain modulation varying across neurons [71]. These experiments were conducted on anesthetized and paralyzed cats and one should be careful when linking these results with neuroimaging and psychophysics on humans. A series of studies have reported strong dependency between the animals’ states and the response of neurons in primary visual cortex including visually evoked response [72], magnitude and correlation of noise [73] and suppressive connections [74]. Electrophysiological and neuroimaging studies on monkeys and humans have shown divergent results when probing the neural correlate of interocular suppression in V1, either with the same or different types of neural signal and measurement [reviewed in 4]. Attentional state is a critical factor when comparing the results across experiments because interocular suppression is dependent on attentional state [11,75,76]. Two human neuroimaging studies [9,10] reported dichoptic masking effects in V1 and showed that the suppression could be accounted for by divisive normalization. These experiments used a central attention task to control observers’ attention at fixation. This is generally consistent with the prediction of the current model that if attention is withdrawn from the stimuli, the competitor will only generate a contrast gain change which can be accounted by interocular divisive normalization. Single-cell recording [77] and psychophysical [7] studies have shown that the origins of the suppression contributed by monocular and dichoptic masks can be distinguished by manipulating the spatiotemporal properties of the mask or by testing the adaptability of the mask. In the current model, the stimuli presented either in the same or different eyes can suppress the neurons through normalization. We did not aim to address the particular neural pathways that support the normalization in the same eye or across eyes; normalization might occur at different stages of neural processing and it might be implemented by different neural mechanisms [6]. In dichoptic masking studies, interocular suppression can be reduced by adding matched images, with either non-zero [78] or zero [79] disparity, in the two eyes. In contrast, in our experiment, presenting the split competitor to both eyes did not increase the correspondence between the images in the two eyes. Likewise, adding binocular fusion markers to manipulate the correspondence between the images in the two eyes did not change the size-dependent interocular suppression [11]. In binocular rivalry, it has been reported that a large stimulus presented to one eye has shorter dominance durations than that of a small stimulus presented simultaneously to the other eye [80], similar to the size effect reported here. We are agnostic about whether the surface-boundary account proposed by Ooi and He [80] to explain this binocular rivalry result could be used to model the contrast-gain and response-gain modulation in our experiments and the absence of response-gain modulation by the small competitor under withdrawn attention reported by Ling and Blake [11]. In addition to the onset-flash suppression used in this study, various related methods have been used to probe the interocular interactions regulating competing information from the two eyes, e. g. , binocular rivalry [81], generalized flash suppression [2], and continuous flash suppression [3]. Our model does not attempt to account for these various perceptual phenomena. For example, in continuous flash suppression, presenting a static target with a dynamic competitor gives rise to a depth of suppression much stronger than most of the other methods of interocular suppression. How such manipulations increase the dominance duration is beyond the scope of present study. Even so, it is likely that these distinct perceptual phenomena share common neural processes [8,82]. Further research is required to investigate whether the model proposed here, including divisive normalization, as well as spatial- and feature-selective attention, can be extended to explain this wide range of perceptual phenomena.

In interocular suppression, a visible target presented in one eye can be rendered invisible by a competing image (the competitor) presented in the other eye. This phenomenon is a striking demonstration of the discrepancy between physical inputs to the visual system and perception, and it also allows neuroscientists to study how perceptual systems regulate competing information. Interocular suppression has been explained by mutually suppressive interactions (modeled by divisive normalization) between neurons that respond differentially to the two eyes. Attention, which selects relevant information in natural viewing condition, has also been found to play a role in interocular suppression. But the specific role of attentional modulation is still an open question. In this study, we proposed a computational model of interocular suppression integrating both attentional modulation and divisive normalization. By modeling the hypothetical neural responses and fitting the model to psychophysical data, we showed that interocular suppression involves an attentional modulation selective for the orientation of the competitor, and covering the spatial extent of the competitor. We conclude that both attention and divisive normalization contribute to interocular suppression, and that their impacts are distinguishable.

Abstract Introduction Materials and Methods Results Discussion

2015

Deconstructing Interocular Suppression: Attention and Divisive Normalization

Over the past 30 years, benzimidazoles have increasingly been used to treat cystic echinococcosis (CE). The efficacy of benzimidazoles, however, remains unclear. We systematically searched MEDLINE, EMBASE, SIGLE, and CCTR to identify studies on benzimidazole treatment outcome. A large heterogeneity of methods in 23 reports precluded a meta-analysis of published results. Specialist centres were contacted to provide individual patient data. We conducted survival analyses for cyst response defined as inactive (CE4 or CE5 by the ultrasound-based World Health Organisation [WHO] classification scheme) or as disappeared. We collected data from 711 treated patients with 1,308 cysts from six centres (five countries). Analysis was restricted to 1,159 liver and peritoneal cysts. Overall, 1–2 y after initiation of benzimidazole treatment 50%–75% of active C1 cysts were classified as inactive/disappeared compared to 30%–55% of CE2 and CE3 cysts. Further in analyzing the rate of inactivation/disappearance with regard to cyst size, 50%–60% of cysts <6 cm responded to treatment after 1–2 y compared to 25%–50% of cysts >6 cm. However, 25% of cysts reverted to active status within 1. 5 to 2 y after having initially responded and multiple relapses were observed; after the second and third treatment 60% of cysts relapsed within 2 y. We estimated that 2 y after treatment initiation 40% of cysts are still active or become active again. The overall efficacy of benzimidazoles has been overstated in the past. There is an urgent need for a pragmatic randomised controlled trial that compares standardized benzimidazole therapy on responsive cyst stages with the other treatment modalities. Cystic echinococcosis (CE, hydatid disease) is a parasitic disease of worldwide prevalence. Hydatid cysts occur mainly in the liver (70%) and the lung (20%). Clinical symptoms and signs depend on their localisation, size, and number. Currently four treatment modalities are in use: (1) surgery, (2) PAIR (puncture, aspiration, injection of protoscolicidal agent, reaspiration), (3) chemotherapy with albendazole (ABZ) or mebendazole (MBZ), and (4) watch and wait for inactive, clinically silent cysts. The evidence supporting any of the four treatment modalities, from carefully designed clinical studies, is insufficient, and choosing treatment options for patients remains controversial [1]. The use of benzimidazoles in CE treatment started in the 1970s with MBZ. In the early 1980s ABZ was introduced and since then has largely replaced mebendazole. The main advantages of ABZ are claimed to be a lower dosage and better intestinal absorption. In treatment centres MBZ and ABZ are given at the World Health Organisation (WHO) recommended dosages of (MBZ, 40–50 mg/kg/day; ABZ, 10–15 mg/kg/day) [2]. Variability exists in the duration of treatment, which remains undefined. Duration of treatment is determined according to the ultrasound-based treatment response, resulting in repetitive treatment, which is part of our analysis. Chemotherapy for the treatment of CE was initially recommended for inoperable patients and patients with multiple organ disease [2], [3]. Over the past decade several studies, mainly case series, have been published suggesting that chemotherapy could be an alternative to surgery in patients with uncomplicated cysts, leading to an increased use of chemotherapy over the years [4]. After more than 30 y of benzimidazole use, the following crucial question remains unanswered: what is the efficacy of benzimidazoles stratified by type and size of cysts, daily dose, and duration of treatment? This project started with a systematic review of the published literature on the efficacy of treating CE with benzimidazoles. We had to conclude, however, that we could not obtain a clear picture of the long-term outcome of benzimidazole treatment because inclusion criteria, treatment, outcome measures, and follow-up of published studies varied widely with substantial overlap of cohorts [1], thus precluding a meta-analysis of published results. We therefore initiated EchinoMEDREV, a collaborative effort of CE specialists, to collect individual patient data from patients treated with benzimidazoles. The main objectives of this collaborative study were to describe cyst outcome after initiation of benzimidazole treatment, with outcome defined by cyst stage determined by ultrasound following the WHO classification of 2001 [3], and to explore differences in outcome by cyst stage and size at initiation of treatment by using a common analytical strategy for all data across treatment centres. A systematic search of MEDLINE, EMBASE, CCTR, and SIGLE was carried out from their inception until week 4 of 2004. The search was performed by a research librarian using the following search terms: echinococcosis, albendazole, mebendazole, hydatid disease, cystic echinococcosis. We also searched reference lists and asked researchers in the field for additional studies. No language restriction was used. Abstracts were screened for suitability by MS. The eligibility of studies was assessed independently by two investigators (TJ and MS). We included all types of study design with a minimum of 30 patients treated either with ABZ or MBZ. Studies in which drug treatment was an adjunct to surgery, PAIR, or a second drug were excluded. The studies identified in the literature search revealed that there were large differences in baseline assessment of cyst stages, interventions (dose and duration of chemotherapy), length of follow-up, and outcome measures between published trials. These differences precluded the possibility to perform a meta-analysis of published results. Therefore we decided to collect individual patient data from the identified centres and initiated the EchinoMEDREV project. Study centres that had conducted clinical studies on benzimidazole treatment of CE were contacted and asked to contribute published and unpublished individual patient data of benzimidazole-treated CE patients. Data extraction forms were developed, piloted, and revised. Data collection started in June 2005 and ended in December 2007. Data were extracted from patient charts at the individual treatment centres. Data collected were: demographic data (age, sex); treatment data (MBZ, ABZ, dosage, and duration of treatment, side effects, previous treatments); imaging data (initial ultrasound staging and staging at follow-up visits). The forms were sent to the coordinating centre at the University Hospital in Heidelberg where data were electronically entered into a database with EpiData, using data entry checks. Accuracy in data entry was checked by double entry verification. A final dataset was created after correcting detected data entry errors and exported to Stata for statistical analysis. Patients with single or multiple hydatid cysts were eligible. Cyst stage had to be recorded at the beginning and at least once after completion of the initial treatment episode. The minimum follow-up period was 1 y after completion of initial treatment. Cyst activity had to be assessed by ultrasonography and classified according to WHO (CL–CE5 or active [A]/transitional [T]/inactive [I]), Gharbi, Perdomo, or Caremani (Table 1). The analysis presented here includes only liver and peritoneal cysts (70%–75% of all CE cysts in humans), which were assessed by ultrasonography, and excluded lung cysts as they are not usually assessed by ultrasonography. The cyst was used as the unit of analysis for a description of achieved outcomes, and the presence of multiple cysts was not taken into account. Data were analysed by intention-to-continue-treatment, ignoring treatment changes (MBZ/ABZ), interruptions, and subsequent treatment episodes. We analysed several endpoints. First, initial treatment success for a cyst was defined as transformation from an initially active or transitional stage to an inactive stage or disappearance of the cyst (see Table 1 for classification based on ultrasonography). For this analysis time was measured from the start of first documented treatment to the date the stage was assessed as inactive or as disappeared or to the last documented assessment. Second, an analysis was made of the time for a cyst to become active again after the cysts had been staged as inactive; a necessary step, as some cysts that had reached an inactive stage had subsequently been staged as active again upon ultrasonography. For this analysis time was measured from the first (or second, or third) date at which a cyst was staged as inactive until the cyst was staged again as active. For these separate endpoints we performed time-to-event analyses using the Kaplan-Meier method and calculated the cumulative incidence of the events by subtracting the Kaplan-Meier survival estimate from one. Descriptive figures are presented stratified by centre where appropriate. Despite the fact that all previous studies on CE cyst development had treated cysts as an independent unit of analysis even if multiple cysts were present in the same patient, we addressed clustering of cysts within patients. The question of heterogeneity by centre was also examined with data from several treatment centres. When addressing clearly specified hypotheses—such as the association of cyst size and time to inactivity or cyst disappearance—Cox proportional hazards models were fitted and a robust variance estimator was used [5] to account for the clustering of cysts within patients. In addition indicator variables were included for each centre. Two questions were investigated using robust Cox proportional hazards model: (1) the association of cyst CE stage at baseline with time to first inactivity or disappearance, and (2) the association of cyst size (<6 cm versus >6 cm) with time to inactivity or cyst disappearance. For the second question the first year (day 0 to day 365) and the follow-up time after year one (day 366 onwards) were analysed separately, because descriptive cumulative incidence plots hinted at the possibility that the cyst size mattered only after year one. Wald test-based p-values were calculated to obtain a hypothesis test for a whole group of indicator variables to be included in the robust Cox proportional hazards model. p<0. 05 was considered statistically significant. For a general description of patients and cysts included in this analysis, counts and mean and standard deviation are provided where appropriate. After observing and describing a multiphase phenomenon of cyst response to treatment, we additionally performed a simulation of the fate of 5,000 hypothetical cysts starting at treatment initiation and moving across the different phases of becoming inactive and relapsing to active. We made the simplifying assumption that phase durations can be described by an exponential distribution. We took the median and/or the 25th percentile obtained from the cumulative incidence estimates for the transitions from active to inactive and for the transitions from inactive to active. We then derived the lambda parameter of the exponential distribution via standard formulae: median, lambda = median/ln (2); 25th percentile, lambda = p25/ln (1/0. 75). We further assumed that the duration of the next phase is independent of the duration of the previous phase. For each simulated cyst we assessed the current stage at year 1 to year 5 after treatment initiation. All analyses were performed using Stata Version 10 (StataCorp). Out of 353 citations identified, 23 papers met the inclusion criteria (Figure 1). Three publications were randomised controlled trials [6]–[8], all other studies included were prospective or retrospective case series [9]–[28]. 19 publications were in English, one was written in Romanian [19], one in Chinese [23], one in Russian [25], and one in Italian [15]. Tables 2 and 3 summarize the characteristics of publications by the specialist centres identified through the literature search. All 14 specialist centres identified were contacted: no reply was received from four centres [7], [8], [22]–[26]; one centre was unable to participate [6]; and nine centres agreed to participate. Of these nine centres six provided data [9]–[20], the others were unable to provide data [21], [27], [28]. In total we received data on 711 patients with 1,308 cysts. Table 4 summarizes the characteristics of patients with liver and peritoneal cysts. Table 5 shows cyst and follow-up details. Figure 2 shows length of follow-up per centre. The analysis presented here was restricted to patients with liver and peritoneal cysts, because of the reliability of ultrasound classification compared to other cyst locations. This restriction resulted in 1,159 cysts in 612 patients for analysis. Approximately 68% of data was obtained from one centre (Table 4). Figure 3 shows the treatment response of individual cyst stages. Overall, 1–2 y after initiation of benzimidazole treatment 50%–75% of cysts initially staged as active in the CE1 category were staged as inactive or had disappeared compared to 30%–55% of CE2 and CE3 cysts. In the robust Cox proportional hazards model, CE3 stage cysts responded poorer than CE1 cysts from year one onwards (p = 0. 043), but not up to year one (p = 0. 43), and a centre effect was noted from year one onwards (p = 0. 033, Wald test). We further analysed the rate of inactivation/disappearance with regard to cyst size (Figure 4). Overall, 50%–60% of cysts <6 cm at baseline responded to treatment after 1–2 y, compared to 25%–50% of cysts >6 cm. In the robust Cox proportional hazards model cysts <6 cm responded better than larger cysts (p = 0. 006) and a strong centre effect was noted (p<0. 0001). Figure 5 shows the cumulative incidence of reaching an inactive stage or a disappearance of cysts for the first time by centres. Data from Greece and Bulgaria show inactivation/disappearance rates of cysts of 75%, increasing to around 90% within 2 y in Bulgaria. In contrast data from Palermo show inactivation or disappearance of cysts in approximately 20% of cases after 2 y. Data from Rome, Romania, and Turkey are between Greece and Palermo. Overall, cysts that reached an inactive stage for the first time relapsed (returned into an A or T stage) in around 25% of cases 2 y after inactivation (Figure 6). Cysts that reached an inactive stage for a second or third time showed relapse at a higher proportion and at an earlier stage: 60% of cysts relapsed within 2 y after the second or third inactivation. Figure 7 shows the proportion of inactive/disappeared cysts over time stratified by the first, second, and third time the cysts started from A/T stages. The cumulative incidence curve after first A/I reflects what has been observed after treatment initiation. Cysts that were staged A/T for the second and third time were staged as inactive or as disappeared in about 75% and 85% 1 y later. These results were almost exclusively from the Rome centre. In the simulation of hypothetical cysts, we estimated that 1 and 2 y after treatment initiation, 60% and 40% of cysts are still active or become active again. In a collaborative effort, individual data from patients with CE were pooled from six centres in five countries and outcomes of liver and peritoneal cysts treated with benzimidazoles were described. We found a strong association between cyst activity and response to treatment, with a better response in highly active CE1 cysts, and an association in treatment response depending on the size of the cyst at the beginning of treatment, with cysts <6 cm in diameter responding better. Thus, our data suggest that small highly active cysts show the best initial treatment response. Overall 25% of cysts reverted to active status within 1. 5 to 2 y after having initially responded, and multiple relapses were observed. We estimated that 2 y after treatment initiation 40% of cysts are still active or become active again. Our results are biologically plausible because early in the disease host response resulting in an increasing thickness of the pericyst and consolidation of cyst content has not yet reached a degree that prevents the drug to reach its target [29]. Additionally, it is important to note that natural decay is a component of the observed rate of inactivation. Available data suggest that this decay may be as high as 13. 6% within 18 mo [29], and up to 20. 6% within 44 mo [30]. This finding clearly leads to an overestimation of response to benzimidazole treatment as calculated from longitudinal data, which increase with the length of observation. There are several limitations to this study. The published data collected from participating specialist centres are from case series. Results from case series are considered low level evidence in determining the efficacy of treatment options. In several analyses we found heterogeneity by centre. For example, the cumulative incidence curves for reaching an inactive cyst stage for the first time or the disappearance of a cyst after initial treatment showed large intercentre variability. Greek and Bulgarian data show a very rapid response, whereas data from Palermo show a very sluggish response to treatment. Time to first inactivation of cysts in the other centres looks quite similar. Rapid response to treatment in Greece and Bulgaria, however, remains unexplained. The particularly slow response to treatment shown in the dataset from Palermo could be due to the larger mean size of cysts at presentation, the difficulty of translating the Caremani classification into the WHO ultrasound classification (Table 1), and the well-known fact that inter-rater agreement between experts on classification of certain cyst stages is low, in particular for cysts containing daughter cysts—stage CE2 and stage CE3b according to WHO. Depending on the amount of consolidated matrix, cysts are either classified CE2 (daughter cysts with no matrix) or CE3b (daughter cysts with matrix). This distinction is important since the former is regarded active, the latter transitional [31]. Interobserver discrepancies occurred in the description of transitions from the inactive stages CE4 and CE5 to the active stages CE1 or CE2; some observers described these transitions and regard them as possible, whereas others do not. However, the number of misclassified cysts is not quantifiable in a retrospective dataset. It is very difficult, if not impossible, to consider all types and directions of biases when attempting to estimate the response of CE cysts to benzimidazoles from the available data. A very strong bias is certainly introduced by differing observation times with considerable impact on inactivation due to spontaneous involution. A final problem concerns data that we were unable to obtain (Table 2). Two centres that initially offered to deliver large datasets were eventually unable to do so. Two-thirds of our data have been provided by Rome, consequently the results are predominantly from one centre. Despite these limitations, to our knowledge, this study represents the largest CE dataset ever collected and analyzed in a uniform approach; further it is likely the only dataset obtained from the main international specialist groups. The recommendations on benzimidazole treatment of CE are currently based on the published results from these centres. Through the collection of individual patient data and the pooled analysis of these data we have managed to overcome some of the existing limitations present in the published literature. Does our study provide sufficient evidence to influence decisions for the treatment of CE? We think that our results are strong enough to cast doubts on overoptimistic expectations of the overall efficacy of benzimidazoles. When looking into substrata of the cyst population small CE1 cysts (diameter <6 cm) are a promising target for benzimidazole therapy, whereas stage CE2 and CE3 cysts respond poorly. The available evidence from this and other studies does not yet allow us, however, to formulate solid evidence-based drug treatment recommendations across all cyst stages, sizes, and locations. Our results highlight the urgent need to compare in a pragmatic randomised controlled trial the effect of standardized benzimidazole dose regimens on the individual active cyst stages (CE1, CE2, CE3a, and CE3b) substratified by cyst size. Such a trial would investigate as a primary outcome the proportion of cysts that become inactive (cyst stages CE4 and CE5) after treatment, and as a secondary outcome the yearly relapse rates up to 5 y after completion of treatment. The clarification of the efficacy of benzimidazoles in CE treatment is of paramount importance since benzimidazoles are the only drugs currently available to treat this neglected disease. Surgery as an alternative to benzimidazoles carries a significantly higher risk of adverse events, in particular intra- and postoperative morbidity and mortality and disseminated disease due to intraoperative spillage of viable hydatid material. Percutaneous fine needle techniques such as PAIR are only applicable to cyst stages CE1 and possibly CE3a, but not to CE2 and CE3b, which makes it necessary to explore large bore catheter techniques if albendazole turns out to be less effective in these cyst stages as suggested by our analysis.

Cystic echinococcosis (CE) is a parasitic infection of worldwide occurrence transmitted to humans by dogs. After infection cysts develop, mainly in the liver and lung. Ultrasound-based staging of cysts into active, transitional, and inactive has opened new venues for treatment and follow-up. Currently four treatment modalities are in use: (1) surgery, (2) percutaneous sterilization techniques, (3) chemotherapy with benzimidazoles, and (4) watch and wait for inactive cysts. However, evidence is insufficient for these four modalities, and determining individual treatment options for patients remains controversial. Medical treatment with benzimidazoles started in the 1970s. Important questions remain unanswered, however, such as efficacy stratified by cyst type and the duration of treatment. We therefore initiated EchinoMEDREV, a collaborative effort to collect individual patient data from patients treated with benzimidazoles and to analyze cyst outcome after initiation of benzimidazole therapy using a common analytical strategy across treatment centres. We found that the efficacy of benzimidazoles has been overstated in the past. Additionally, natural cyst decay has not been taken into account. Evidence from randomized controlled trials is urgently needed to determine the true efficacy of benzimidazoles. Our analysis will help to design benzimidazole trial arms on the basis of the currently available best evidence.

Abstract Introduction Methods Results Discussion

infectious diseases/helminth infections

2009

Treatment Response of Cystic Echinococcosis to Benzimidazoles: A Systematic Review

Cytomegalovirus (CMV) is a ubiquitous β-herpesvirus that establishes life-long latent infection in a high percentage of the population worldwide. CMV induces the strongest and most durable CD8+ T cell response known in human clinical medicine. Due to its unique properties, the virus represents a promising candidate vaccine vector for the induction of persistent cellular immunity. To take advantage of this, we constructed a recombinant murine CMV (MCMV) expressing an MHC-I restricted epitope from influenza A virus (IAV) H1N1 within the immediate early 2 (ie2) gene. Only mice that were immunized intranasally (i. n.) were capable of controlling IAV infection, despite the greater potency of the intraperitoneally (i. p.) vaccination in inducing a systemic IAV-specific CD8+ T cell response. The protective capacity of the i. n. immunization was associated with its ability to induce IAV-specific tissue-resident memory CD8+ T (CD8TRM) cells in the lungs. Our data demonstrate that the protective effect exerted by the i. n. immunization was critically mediated by antigen-specific CD8+ T cells. CD8TRM cells promoted the induction of IFNγ and chemokines that facilitate the recruitment of antigen-specific CD8+ T cells to the lungs. Overall, our results showed that locally applied MCMV vectors could induce mucosal immunity at sites of entry, providing superior immune protection against respiratory infections. Respiratory infections caused by influenza viruses usually are associated with mild-to-moderate disease symptoms but are linked with high morbidity and mortality in susceptible populations like the elderly, young children, patients with co-morbidities and immunocompromised patients. Influenza virus causes seasonal epidemics, with typically 3 to 5 million cases of severe illness worldwide, according to WHO reports [1], and influenza type A viruses (IAV) cause the more severe disease form. Vaccines against influenza are based on the induction of adaptive immunity that targets the projected yearly epidemics. While most vaccines are based on inactivated IAV formulations inducing anti-IAV IgG responses, live attenuated influenza vaccines (LAIV) are also used as formulations for i. n. administration. This is based on the assumption that the induction of local immunity may provide superior immune protection [2,3]. However, it remains unclear whether this protection depends on local IgA responses, on cytotoxic T cell responses, or on their combined antiviral activity. Of note, functional T cell responses were shown to substantially contribute to antiviral IAV immunity [4–6]. In particular, cytotoxic influenza-specific CD8+ T lymphocytes (CTLs) promote viral clearance indirectly by secretion of pro-inflammatory cytokines such as IFNγ [7] and directly by perforin/Fas-mediated killing of infected epithelial cells in the bronchoalveolar space [8]. However, it remained unclear if T cell responses alone may control IAV, or if Ig responses were the crucial contributor to LAIV-mediated immune protection. We considered that this question could be addressed by developing a vaccine formulation that optimizes T cell responses against IAV while excluding the humoral ones. CMV infection induces sustained functional T cell responses that are stronger in the long-term than the immune response to any other infectious pathogen [9]. Experiments in the mouse model have shown that defined CMV epitope-specific CD8+ T cells accumulate in tissues and blood and are maintained at stable high levels during mouse CMV (MCMV) latency [10]. This phenomenon was termed ‘‘Memory Inflation” [11]. While some MCMV derived peptides, as the ones derived from the IE3 (416RALEYKNL423) and M139 proteins (419TVYGFCLL426) induce inflationary responses, other peptides, such as the M45-derived (985HGIRNASFI993), induce conventional CD8+ T cell response [12]. Antigen-experienced CD8+ T cells are subdivided into CD62L- effector memory (CD8TEM) and CD62L+ central memory CD8+ T cells (CD8TCM). The antigen-specific CD8+ T cells during latent infection bear predominantly a CD8TEM phenotype and localize in secondary lymphoid or non-lymphoid organs [13]. They may provide immune protection against diverse viral targets [14–18], but also against bacterial [19] or tumor antigens [20,21]. The exceptionally long-lasting cellular immunity to CMV antigens has raised the interest in CMV as a potential new vaccine vector [14]. Many studies have demonstrated of optimal design of such CMV-based vaccines display efficient protection against virus infection such as rhesus macaque CMV (RhCMV) based Ebola [22] and SIV [23] vaccines and MCMV based Ebola virus vaccines [15]. Erkes, D. A. and Qiu, Z. et al. demonstrated that CMV based vaccine vectors provide protection in suppressing tumors [24,25]. Both CD8TEM and CD8TCM subsets recirculate between the blood, the lymphoid organs, and the peripheral tissues. A special subset of memory CD8+ T cells (CD8TRM) resides in non-lymphoid tissues such as lungs, the female reproductive tract (FRT), the skin, the brain or the small intestine [26–29]. These cells lose the capacity of recirculating, maintain themselves at the site of infection, and their phenotype and transcriptional profile differ from classical memory T cells [30]. The well-characterized CD8TRM cells express C-type lectin CD69 [26] and the integrin αEβ7, also known as CD103 [30]. They provide rapid and superior protection against pathogens at the site of infection [26,30,31]. A recent publication argued that a vaccine formulation adjuvanted by IL-1β enhances the immune control of IAV by improving mucosal T cell responses [32], but IL-1β improved both humoral and cellular responses in their study. Hence, the contribution of CD8TRM to IAV immune control remains unclear. CD8TRM are found in the salivary glands of MCMV-infected animals [33], but not in their lungs [34]. We showed that i. n. infection with MCMV induces inflationary CD8+ T cell responses, but also that memory inflation is more pronounced in relative and absolute terms upon i. p. infection [35]. The i. n. administration of an MCMV vaccine vector induced CD8TRM responses in the lungs [29] and only i. n. immunization restricted the replication of respiratory syncytial virus (RSV) upon challenge [29], indicating that CD8TRM elicited by i. n. administration of MCMV vectors might provide immune protection against respiratory virus infections, yet this evidence remains correlative. Upon antigen encounter, CD8TRM cells quickly reactivate at the mucosal site and secrete cytokines and chemokines or support the release of inflammatory mediators by other immune cells [28,36,37]. Lung airway CD8TRM cells provide protection against respiratory virus infection through IFNγ and help to recruit circulating memory CD8+ T cells to the site of infection in an IFNγ-dependent way [36]. Therefore, to understand if CD8TRM cell may provide immune control of respiratory infections may help to refine strategies for tissue-targeted vaccine design. In this study, we constructed an MCMV vector expressing the MHC-I restricted peptide 533IYSTVASSL541 (IVL533-541) [38] from IAV H1N1 hemagglutinin (HA) -MCMVIVL under the transcriptional control of the ie2 promotor. We investigated the potential of this recombinant virus to induce HA-specific CD8+ T cells that confer protection against a lethal IAV challenge. We showed that i. n. , but not i. p. immunization with MCMVIVL resulted in robust protection against an IAV challenge. Protection following i. n. MCMVIVL immunization was associated with high levels of antigen-specific CD8TRM cells in the lungs, and targeted depletion of lung-CD8TRM cells revealed that the control of the IAV in the lungs depended on these cells. We showed recently that MCMV vector expressing a single optimally positioned MHC-I restricted antigenic epitope provides a more efficient immune protection than vectors expressing the full-length protein [21]. Therefore, we constructed an MCMV influenza vaccine vector by inserting the coding sequence for the H-2Kd MHC-I restricted peptide IYSTVASSL from the hemagglutinin (HA) of the H1N1 (PR8) IAV [38] into the C-terminus of the MCMV ie2 gene. The resulting recombinant virus was called MCMVIVL (Fig 1A). To test if the recombinant virus retained its capacity to replicate in host cells, virus replication was assessed by multi-step growth kinetics assays in NIH-3T3 cells in vitro and by ex vivo quantification of virus titers in livers, lungs and spleens 5 days post-infection (dpi) and in salivary glands 21 dpi. MCMVIVL showed identical replication properties as the MCMVWT, both in vitro (Fig 1B) and in vivo (Fig 1C), indicating that the insertion of the IVL533-541 epitope does not impair virus replication and spread. We have shown that mucosal infection with MCMV by the i. n. route induces memory inflation, although to a lower extent than upon the i. p. infection route [35]. To define if this pattern would hold true for the artificially incorporated influenza epitope as well, we compared the magnitude of the CD8+ T cell responses to MCMVIVL and MCMVWT induced via the i. n. and i. p. route, respectively. S1 Fig shows the gating strategy of flow cytometry analysis. The kinetics of antigen-specific CD8+ T cell responses in peripheral blood was determined by IVL-tetramer staining. While we did not observe striking difference at early times post immunization, i. p. immunization induced an overall higher magnitude of inflationary CD8+ T cell response during latency (Fig 2A and 2B). This pattern was observed both in relative terms (Fig 2A) and in absolute cell counts (Fig 2B). We next analyzed the IVL-specific CD8+ T cell responses in spleens, lungs and mediastinal lymph nodes (mLNs) at times of latency (>3months post infection (p. i) ). Similarly, i. p. immunization induced higher levels of IVL-specific CD8+ T cells than the i. n. immunization in the spleen and lungs, both in relative (Fig 2C and 2D) and in absolute terms (Fig 2F and 2G). There were no significant differences in the mLNs (Fig 2E and 2H). In sum, mucosal (i. n.) immunization with MCMVIVL induces a systemic inflationary IVL-specific CD8+ T cell response, whereas the overall magnitude of the IAV-specific CD8+ T cell response is less pronounced compared to that induced in i. p. immunized mice. To test whether immunization with MCMVIVL protects against IAV infection, latently immunized BALB/c mice were i. n. challenged with IAV. IAV titers in the lungs were quantified 5 dpi. Viral loads in mice that were immunized with MCMVWT via either the i. p. or the i. n. route were comparable to those detected in mock-immunized mice (Fig 3A). In contrast, mice immunized with MCMVIVL via the i. n. route showed significantly lower IAV titers than in any other group. Interestingly, i. p. MCMVIVL immunization also resulted in reduced IAV loads, but to a lower extent than the i. n. immunization (Fig 3A). Similarly, animals immunized with MCMVWT suffered the most severe weight loss whilst i. n. immunization of MCMVIVL led to the least pronounced body weight loss. I. p. immunization with MCMVIVL displayed an intermediate level (Fig 3B). Numerous studies have reported that CD8+ T cells play an important role in protecting against influenza infection [39,40] and it was reasonable to assume that our vector provided immune protection by eliciting CD8+ T cell responses. To formally prove that efficient immune control of IAV observed in the MCMVIVL (i. n.) immunized group depends on CD8+ T cells, we depleted these cells by systemic treatment of mice with a depleting anti-CD8α antibody (depletion efficiency is shown in S2A Fig) one day prior to IAV challenge and quantified viral titers in lungs 6 dpi. While the virus titer was below the detection limit in mice that were i. n. immunized with MCMVIVL and received isotype control antibodies, CD8+ T cell depletion indeed resulted in a significant increase of the IAV titer to levels comparable to the groups that were i. p. immunized with MCMVIVL and to control mice immunized with MCMVWT by i. n. route (Fig 3C). CD8+ T cell depletion also slightly increased virus titers in both control groups—MCMVIVL (i. p.) and MCMVWT (i. n.), but not as pronounced as in the MCMVIVL i. n. immunized group (Fig 3C). Similar to Fig 3B, animals of all experimental groups showed a comparable body weight loss post-challenge, whereas i. n. MCMVIVL immunized mice showed a faster recovery than the i. p. immunized mice (Fig 3D). Of note, this difference disappeared in the groups lacking CD8+ T cells (Fig 3E). Together, these data demonstrate that IAV-specific CD8+ T cells induced by the mucosal (i. n.) administration of MCMVIVL confer protection against IAV in the lungs of vaccinated mice. We further compared the lung pathology upon IAV challenge by histology. A moderate perivascular inflammation was observed in the lungs of most mice (stars) and to a lesser degree surrounding the bronchioles (arrows) (Fig 3F). The CD8+ T cell depleted group showed more severe pathology than isotype-treated controls, but the difference was not very pronounced (Fig 3G). Taken together, these data imply that intranasal immunization with the MCMVIVL vector can limit IAV growth in the lungs by inducing IAV-specific CD8+ T cell responses, whereas the clinical outcome is only moderately improved. We assumed that intranasal MCMVIVL immunization may control the IAV replication by inducing CD8TRM cell responses in the lungs. In order to test this hypothesis, we identified the CD8TRM cell compartment by staining cells with the CD69 [26] and the CD103 [30] marker at times of MCMV latency (> 3 months p. i.). It is known that resident memory T cells reside in the mucosal tissue layer and are non-migratory [31]. In this study, intravenous (i. v.) injection of a fluorescent anti-CD45 antibody prior sampling allows for the discrimination of circulating leukocytes (fluorescence-positive) from emigrated or tissue-resident leukocytes (fluorescence-negative). The CD69+CD103+ cells from lungs were virtually absent from the CD45-labelled fraction (Fig 4A). Barely CD8TRM could be detected in the spleen and blood regardless of the route of immunization (Fig 4B). CD8TRM (CD69+CD103+) cells were solely induced in the lungs and the frequency was significantly higher when i. n. immunized with MCMVIVL than via the i. p. route (Fig 4B). Approximately forty percent of these lung CD8TRM cells were IVL-tetramer+ which is significantly higher than the i. p. immunization group (Fig 4C). Similar results were observed when gated reversibly (Fig 4D) and when analyze these tetramer+ CD8TRM cell numbers (Fig 4E). CD8TRM cells were also induced in the group that was i. n. immunized with MCMVWT, but not IVL-specific (S3A Fig). In addition, there was an overall higher percentage of the CD69+CD103-CD8+ T cells in the lungs of i. n. immunized mice (S3B Fig), although the absolute cell numbers did not show a significant difference (S3C Fig). Since CD69 and CD103 are imperfect markers of tissue residence, we validated if the CD69+CD103+ and CD69+CD103- populations in lungs are truly CD8TRM by staining them for Eomesodermin (Eomes) expression, because CD8TRM cells show low expression of Eomes. Low levels of Eomes were observed in CD69+CD103+ CD8 T cells, whereas CD69+CD103-CD8+ T cells showed high expression of Eomes which is consistent with primed CD8+ T cells (S3D Fig). Hence the data suggested the CD69+CD103- cell subset show distinct transcription profile from the CD8TRM cell. According to the expression of CD62L and KLRG1, IVL-specific CD8+ T cells are classified into three subsets: effector (TEFF: KLRG1+CD62L-); effector memory (TEM: KLRG1-CD62L-) and central memory (TCM: KLRG1-CD62L+) cells. The fractions of each subset in the mucosal (CD45-) and circulation (CD45+) in the blood, spleen and lung were analyzed (S4A & S4B Fig). While the fraction of CD8TCM cells remained relatively low in all compartments irrespectively of the route of administration, i. p. infection resulted in a response polarized towards TEFF cells, whereas i. n. immunization induced a larger fraction of TEM cells in all analyzed organs both in circulating and mucosal CD8+ T cell subsets (S4A & S4B Fig). Majority of the IVL-specific CD8TRM cells show an EM phenotype (S4C Fig). In summary, i. n. immunization with MCMVIVL induces an accumulation of IAV-specific CD8TRM response in the lungs. Moreover, antigen-specific CD8+ T cell responses induced via the mucosal route skew towards an effector memory phenotype. Resident memory T cells reside in the epithelial barrier of mucosal tissue [31] that is in close proximity to the airways. Hence, they reactivate rapidly to a virus challenge at the site of infection upon encountering cognate antigens [29,31]. To define the relevance of lung CD8TRM cells in protection against IAV challenge, we specifically depleted the airways CD8+ T cells by i. n. administration of αCD8 antibodies one day before challenge (Fig 5A). Upon depletion, tetramer+ CD8TRM cell number reduced significantly while the circulating CD8+ T cell number did not change a lot in the lungs (S2B Fig). The local depletion did not affect the CD8+ T cell counts in the blood (S2C Fig). An earlier day was chosen (4 days post-challenge) to assess whether CD8TRM cell could rapidly confer protection. The IAV titers in the lungs of i. n. MCMVIVL immunized mice were quantified. Targeted depletion of pulmonary CD8TRM cells was associated with a significantly higher viral burden during IAV infection (Fig 5B). These data indicate that CD8TRM cells induced by i. n. immunization with MCMVIVL promote the clearance of IAV. Influenza virus infection can induce a vigorous cytokine storm in airways and lungs, which promotes the recruitment of inflammatory cell. IFNγ as a pivotal antiviral cytokine is expressed early after influenza virus infection [41]. It has been demonstrated that CD8TRM cells activate rapidly when they re-encounter the cognate antigen and provide protection by secreting cytokines such as IFNγ and granzyme B [42,43]. Morabito et al. showed that intranasal immunization with an MCMV-based vaccine vector induced CD8TRM cells and IFNγ was secreted at the very early time upon challenge during RSV infection [29]. Therefore, we measured the production of IFNγ in the bronchoalveolar lavage fluid (BALF) early upon challenge. IFNγ levels were generally low on day 2 post-challenge and no difference could be observed between groups regardless of airway CD8+ T cells depletion (Fig 5C). On day 4, the IFNγ level was significantly increased compared to the level on day 2, but more IFNγ was induced in the control group than in the one lacking CD8TRM cell in the lungs (Fig 5C). IFNγ was also induced in the MCMVIVL i. p. immunization group and extremely low level of IFNγ could be detected in the MCMVWT control group (S5A Fig). These data suggest that primary cognate antigen immunization is needed for the rapid IFNγ secretion and that resident CD8+ T cells may not be the major IFNγ producer. In contrast to these effects, depletion of lung airway CD8+ T cells increased the concentration of IL-6 as compared to the group that was intranasally immunized with MCMVIVL and treated with isotype control antibodies (Fig 5D). Similarly, a higher concentration of IL-6 was also detected in the i. p. immunization group, whereas the MCMVWT control group displayed the highest IL-6 levels (S5B Fig). Very low levels of other cytokines could be detected in all groups, both on day 2 and 4 post-challenge and regardless of the depletion of the airway CD8+ T cell (S5C Fig), suggesting that the presence of pulmonary CD8TRM cells does not affect the Th1, Th2 and Th17 immune profile during early IAV infection. It has been demonstrated that TRM cells help to recruit immune cells to the infection site through the induction of chemokines such as CCL3 and CXCL9 in the female reproductive tract (FRT), and CCL4 in the lungs, either by direct chemokine expression or by their induction in nearby cells, such as epithelial cells [28,29]. To determine whether i. n. immunization with MCMVIVL induced inflammatory chemokines expression upon IAV challenge, a series of inflammatory chemokines were measured in the BALF on day 2 (Fig 5E) and day 4 (Fig 5F) upon IAV challenge. Airway depletion of CD8+ T cells reduced CCL3, CCL4, CCL5 levels on day 2. On day 4, CCL3 and CCL4 levels were significantly higher in the MCMVIVL i. n. group than in the MCMVIVL i. p. and in the MCMVWT i. n. immunization groups. Airway CD8+ T cell depletion reduced the level of CCL3 and CCL4 to values in the i. p. MCMVIVL immunization group (Fig 5F). CXCL9 levels were comparable between the MCMVIVL i. n. and i. p. immunization groups, but dramatically lower in the MCMVWT immunization group (Fig 5F), which was consistent with the low IFNγ level in the BALF, as IFNγ is known as an inducer of CXCL9, which then acts as a T cell-attracting chemokine. Together, these data indicate that CD8TRM cells induced by i. n. immunization with MCMVIVL promote the induction of the pro-inflammatory chemokines CCL3, CCL4, CCL5 and CXCL9, along with a reduction of IL-6 in the lungs. Since i. n. immunization induced stronger chemokine responses in comparison to the i. p. immunization route, we decided to define whether CD8TRM cells induced by MCMVIVL promoted the accumulation of CD8+ T cells in the lungs. We first analyzed the total IVL-specific and CD8+ T cell numbers (CD45+ plus CD45-) in the MCMVIVL i. n. immunization group. IVL-specific and CD8+ T cell numbers increased from day 2 to day 4 post-challenge, but only in mice that were not depleted for airway CD8+ T cells (Fig 6A and 6B). Both IVL-specific and total CD8+ T cell counts increased significantly in the BALF of i. n. immunized mice by day 4 post IAV challenge (Fig 6C and 6D), However, in the mice which CD8+ T cells were depleted prior to challenge, very few IVL-tetramer+ CD8+ T cells (Fig 6C, filled dots) and CD8+ T cells (Fig 6D, filled dots) could be detected in the BALF, both on day 2 and at day 4 post-challenge. These data indicate that CD8+ T cells accumulate in the lungs and migrate to the lung tissue and bronchoalveolar space upon IAV challenge. In addition, IVL-specific CD8+ T cell counts in the lung tissue and BAL were slightly higher in the MCMVIVL i. n. group than in the i. p. immunized group (S6A and S6B Fig). This differs from results prior to IAV challenge, where significantly larger amounts of IVL-specific CD8+ T cells were detected in the i. p. group (Fig 2G). CD8+ T cells in the lung tissue were further analyzed by in vivo labeling with anti-CD45 antibodies in the presence or absence of airway CD8+ T cells. The IVL-specific CD8+ T cell population failed to expand upon airway CD8+ T cell depletion, with significantly lower numbers in CD45- subset on day 4. However, IAV-specific CD8+ T cell counts showed an increasing trend both in the CD45+ and in the CD45- subsets on day 4 post-challenge (Fig 6E). Airway CD8+ T cell depletion prevented also the expansion of total CD8+ T cells counts on day 4 (Fig 6F). Interestingly, in contrast to IVL-specific CD8+ T cells, only the CD45- fraction of the total CD8 pool expanded on day 4 (Fig 6F), suggesting that bystander CD8+ T cells were also accumulated within the lungs. The depletion effects were more pronounced later upon antibody administration. Hence, direct depletion of incoming cells appears an unlikely scenario. Together with the increased chemokines in the BAL, these data indicate that the increase of IVL-specific CD8+ T cells upon challenge is not due to in situ proliferation or differentiation but most probably by recruiting CD8+ T cells from circulating system. We also checked whether CD8TRM cells expanded upon IAV challenge that may contribute to the accumulation of CD8+ T cells. We found that the number of CD8TRM cells in the lungs did not expand from day 2 to day 4; if anything, their frequency decreased (Fig 7A and 7B). Likewise, IVL-Tetramer+ CD8TRM cell counts were also slightly reduced from day 2 to day 4 post-challenge (Fig 7C), although IVL-Tetramer+ CD8+ T cell counts increased at the same time (Fig 6A). It seems that the effect of the i. n. depletion was local, since the frequencies (S6C and S6D Fig) and counts (S6E and S6F Fig) of IVL-specific CD8+ T cells in the blood and spleen did not significantly differ upon αCD8 or isotype-control antibody i. n. administration. In addition, CD4+ T cell numbers showed no difference upon IAV challenge when airway CD8+ T cells were depleted compared with the control group (S7 Fig), suggesting that CD8TRM cell do not promote CD4+ T cells trafficking into the lungs. Therefore, our data indicated that CD8TRM cells confer protection by recruiting circulating CD8+ T cells upon IAV challenge. Influenza-specific CD8+ T cells are known to contribute to virus elimination, as the clearance of influenza virus is delayed in T cell-deficient mice [5,44]. However, previous evidence did not clarify whether vaccines solely inducing influenza-specific CD8+ T cell responses improve immune protection. To avoid confounding humoral immune responses and focus on the potential of optimally primed CD8+ T cells in protecting against influenza, we generated a new MCMV based vaccine vector. CMV vaccine vectors expressing exogenous antigenic peptides fused to a CMV gene induce CD8+ T cell responses of unparalleled strength [14,15,21,29]. We show here that robust CD8+ T cell responses against a single MHC-I restricted epitope derived from the HA protein of IAV, promote the clearance of IAV from lungs, but only upon i. n. immunization. While some pathology was observed even in immunized mice, arguing that the protection was not complete, depletion assays confirmed that CD8+ T cells are crucial for the immune protection observed in our model. Morabito et al. showed that the volume of MCMV inoculum affects the magnitude of T cell responses [29]. Hence, a larger inoculum volume could have resulted in even stronger lung CD8+ T cell responses and protection. Remarkably, immunization with the same virus dose by the i. p. route induced even higher magnitude of CD8+ T cell responses, but conferred poor protection. This conundrum was resolved once we noticed that only i. n. immunization induces TRM responses in the lung. CD8TRM cells act as sentinels in the host and form the first line of defense, providing rapid and effective protection to fight against pathogens invasion [27,29,31,45]. Prior studies have revealed that direct delivery of vaccines to the target tissue is necessary for the generation of TRM cells [29,46] and that sustained lung CD8TRM responses in MCMV-infected mice are generated by immunoproteasome-independent antigenic stimulation [47], akin to the CD8 expansions in memory inflation [21], arguing that their induction may share similar or overlapping mechanisms. Some prior studies have claimed that skin-resident CD8TRM cells may confer protection in an antigen-unspecific manner [48], whereas others argued that only the antigen-specific CD8TRM cells respond to cognate antigens [49]. MCMVWT induced robust CD8TRM responses in our model, but these were not specific for IAV, and did not provide any protection against IAV in our study. Site-specific anti-CD8α antibody administration depleted CD8TRM, and increased IAV titers in immunized mice, indicating that CD8TRM cells facilitated IAV elimination. Thus, the protection against IAV challenge requires antigen-specific CD8TRM cells in our model. However, the weight loss was not different in the early days upon challenge and TRM depletion affected chemokine expression and T cell influx at later time points. Former studies have described that the inflationary MCMV-specific CD8+ T cells induced by intraperitoneal infection display an effector phenotype (KLRG1+CD62L-CD127-) [50,51]. Mucosal immunization of MCMVIVL affects not only the magnitude but also the quality of CD8+ T cells skewing cells towards an effector memory phenotype. Since KLRG1 is assumed a marker for terminally differentiated cells which usually are short-lived [52], the low expression of KLRG1 may contribute to a longer life-span of these cells. Additionally, recent work has demonstrated that KLRG1- CD8+ TEM cells traffic and migrate more rapidly to non-lymphoid tissues than the KLRG1+ CD8+ TEFF cells, which mostly remain in the circulation [53,54]. Accordingly, this may give us a clue that CD8TEM cells induced by mucosal immunization may rapidly migrate to the lungs and exert immune functionality there. Nonetheless, further studies need to be done to prove this hypothesis. It has been assumed that CD8TRM cells have poor proliferative capacity upon challenge. Previous work has demonstrated that airway CD8+ T cells fail to expand in vivo upon intratracheal transfer [36] and that CD8TRM cells induced by MCMV infection display a limited proliferative capacity in salivary glands [55]. However, this is in contrast to two recent studies demonstrating that CD8TRM cells in the skin [49] and FRT [56] maintain the capacity of in situ proliferation upon cognate antigen stimulation. Such stimulation differentiates circulating effector memory CD8+ T cells into CD8TRM cells without displacing the pre-existing CD8TRM population [49]. In our study, CD8+ T cells accumulated in the lungs upon IAV challenge, but the CD8TRM population did not expand and the number of antigen-specific CD8TRM cells even displayed a reduction trend. This appears unrelated to apoptosis, because CD8TRM cells showed less caspase3 expression than circulating CD8+ T cells upon challenge. It is possible that CD8TRM cells downregulated CD103 from the cell surface upon activation, and this intriguing question needs to be addressed in future studies. Therefore, our data argued that either lung CD8TRM in general or CD8TRM induced by MCMV i. n. immunization in particular, may behave differently from CD8TRM in other organs. This distinction, however, goes beyond the scope of our current work and remains to be addressed in future studies. Early upon IAV challenge, not only the IVL-specific CD8+ T cells but also the bystander CD8+ T cells in the lung tissue increased significantly, indicating that the accumulation was not due to in situ proliferation but probably due to recruitment from circulating system. In addition, IAV challenge expanded IVL-specific CD8+ T cell counts in the blood and spleen of i. n. immunized mice to levels observed in the i. p. immunized controls, although the levels were significantly lower in the i. n. immunization group before IAV challenge. Overall, these results indicated that i. n. immunization facilitated CD8+ T cell responses upon challenge, both locally in the lungs and systemically in the blood and spleen. It is unclear if this apparent alarming function of TRM cells is exclusive to the lung tissue. We have shown in this study that concentration of CCL3, CCL4 and CXCL9 in the BALF of the MCMVIVL i. n. immunization group is significantly higher than in MCMVIVL i. p. or MCMVWT i. n. immunized mice. Intravital CD45 labeling showed that CD8+ T cells accumulating in the lungs are sequestered from the bloodstream, but not CD8TRM, arguing that circulating antigen-specific cells were attracted into the lungs under the presence of mucosa-resident CD8+ T cells. This is in line with the work of Schenkel et al. showing a rapid local induction of chemokines CXCL9 and CCL3/4 in the FRT upon re-infection, and recruitment of memory CD8+ T cells from the periphery [28]. Depletion of mucosal CD8+ T cells depressed chemokine levels in the BALF to levels seen in the i. p. immunization group. This, together with the high levels of IFNγ in the MCMVIVL i. n. immunization group and extremely low IFNγ in MCMVWT i. n. immunization points to a putative model where antigen-specific re-stimulation induces IFNγ, which drives chemokine responses that recruit CD8+ T cells from the bloodstream to the lungs. We observed a surprising lower level of IL-6 upon challenge in mice that controlled influenza, because IL-6 is known as a cytokine that is involved in controlling virus infection [57–61]. It is not clear if IL-6 reduction was a consequence of lower virus titers or of negative regulation by TRM. McMaster et al. showed a reduced IL-6 production accompanied with lower virus titer in the appearance of lung airway CD8TRM cells [36] and this was similar to a report by Lee et al. in a clinical study in human patients [62]. In summary, our data demonstrate that CD8TRM cells promote the induction of chemokines, which help to drive the recruitment of IVL-specific CD8+ T cells and facilitates the elimination of IAV. Furthermore, the optimal induction of CD8TRM cells in the lungs by the MCMV vector can be only achieved after i. n. vaccination. Therefore, immunization with an MCMV vector at the local site provided CD8+ T cell-based protection against IAV infection. Our results, therefore, demonstrate that CD8+ T cell induction, and CD8TRM in particular, contribute to vaccination outcomes in influenza infection independently of humoral immune responses, and the selection of the adequate immunization route plays a critical role in terms for promoting superior protective efficacy. Mice were housed and handled in agreement with good animal practice as defined by EU directive EU 2010/63 and ETS 123 and the national animal welfare body ‘‘Die Gesellschaft für Versuchstierkunde /Society of Laboratory Animals (GV-SOLAS) ”. Animal experiments were performed in accordance with the German animal protection law and were approved by the responsible state office (Lower Saxony State Office of Consumer Protection and Food Safety) under permit number: 33. 9-42502-04-14/1709. BALB/c mice were purchased from Janvier (Le Genest Saint Isle, France) and housed in the animal facility of the HZI Braunschweig under SPF conditions according to FELASA recommendations [63]. Bone marrow stromal cell line M2-10B4 (CRL-1972) and NIH-3T3 fibroblasts (CRL-1658) were purchased from American Type Culture Collection (ATCC). The cells were maintained in DMEM supplemented with 10% FCS, 1% L-glutamine, and 1% penicillin/streptomycin. C57BL/6 murine embryonic fibroblasts (MEFs) were prepared in-house and maintained as described previously [64]. BAC-derived wild-type murine cytomegalovirus (MCMVWT clone: pSM3fr 3. 3) [65] was propagated on M2-10B4 lysates and purified on a sucrose cushion as described previously [66]. Virus titers were determined on MEFs by plaque assay as shown elsewhere [67]. Recombinant MCMV was generated by the ‘‘en passant mutagenesis”, essentially as described previously [68,69]. In brief, we generated a construct containing an antibiotic resistance cassette coupled with the insertion sequence and the restriction site Sce-I. This construct was flanked by sequences homologous to the target region of insertion within the MCMV BAC genome. Then, the fragment containing the insertion sequences was integrated into the MCMV BAC genome by homologous recombination. In a second step, Sce-I was induced to linearize the BAC followed by a second round of induced homologous recombination to re-circularize it and select for clones that discarded the antibiotic selection marker but retained the inserted sequence. The PR8M variant of Influenza A/PR/8/34 was obtained from the strain collection at the Institute of Molecular Virology, Muenster, Germany. Virus stocks from chorioallantoic fluid of embryonated chicken eggs were generated as previously described [70]. 533IYSTVASSL541 (IVL533-541) -tetramer was bought from MBL (cat. NO. TS-M520-1), anti-CD8α depletion antibody (clone: YTS 169. 4). Rat IgG2b isotype antibody (clone: LTF-2) was purchased from Bio X Cell. Antibodies for flow cytometry included anti-CD3-APC-eFluor780 (clone: 17A2; eBioscience), anti-CD4-Pacific Blue (clone: GK1. 5; BioLegend), anti–CD8α-PerCP/Cy5. 5 (clone: 53–6. 7; BioLegend), anti-CD11a-PE/Cy7 (clone: 2D7; BD Bioscience), anti–CD44-Alexa Fluor 700 (clone: IM7; BioLegend), anti-CD45-APC-eFluor780 (clone: 30-F11; Biolegend), anti-CD62L-eVolve 605 (clone: MEL-14; eBioscience), anti-CD127-PE & PE/Cy7 (clone: A7R34; BioLegend), anti-KLRG1-FITC & BV510 (clone: 2F1/KLRG1; BioLegend), anti-CD103-APC (clone: 2E7; BioLegend), anti-CD69-FITC (clone: H1. 2F3; BioLegend) and anti-IFNγ-APC (clone: XMG1. 2; BioLegend), anti-Eomes-PE & PE/Cy7 (clone: Dan11mag; eBioscience). NIH-3T3 cell monolayers were infected with MCMVWT and MCMVIVL at an MOI of 0. 1, incubated at 37°C for 1h, upon which the inoculum was removed, cells were washed with PBS, and supplied with fresh medium. Cells were incubated for 6 days; the supernatant was harvested every day and stored at -80°C until titration. 6 to 8 weeks old BALB/c female mice were infected with 2 x 105 PFU MCMVWT and MCMVIVL diluted in PBS. For i. p. infection, 100 μl virus dilution was injected. For i. n. infection, mice were first anesthetized with ketamine (10 mg/ml) and xylazine (1 mg/ml) in 0. 9% NaCl (100 μl/10 g body weight), then administered with 20 μl of virus suspension onto nostrils [35]. For IAV challenge, BALB/c mice that were latently (> 3 months) immunized with MCMV were i. n. inoculated with 220 focus forming units (FFU) or with 1100 FFU of PR8M influenza virus as described previously [35]. MCMV virus from organ homogenates was titrated on MEFs with centrifugal enhancement as described previously [17]. Mice were sacrificed by CO2 inhalation, whole lungs were excised and mechanically homogenized using a tissue homogenizer. Tissue homogenates were spun down and supernatants were stored at -70°C. Lung virus titers were determined by using the focus-forming assay (FFA), as described before [70] with minor modifications. Briefly, MDCK cells were cultured in MEM, supplemented with 10% FCS, 1% penicillin/streptomycin. Supernatants of lung tissue homogenates were serially diluted in DMEM, supplemented with 0. 1% BSA and N-acetylated trypsin (NAT; 2. 5 μg/ml) and added to the MDCK cell monolayers. After 1h, cells were overlaid with DMEM supplemented with 1% Avicel, 0. 1% BSA and NAT (2. 5 μg/ml). After 24h cells were fixed with 4% PFA and incubated with quenching solution (PBS, 0. 5% Triton X-100,20 mM Glycin). Cells were then treated with blocking buffer (PBS, 1% BSA, 0. 5% Tween20). Focus forming spots were identified using primary polyclonal goat anti-H1N1 IgG (Virostat), secondary polyclonal rabbit anti-goat IgG conjugated with horseradish peroxidase (KPL) and TrueBlue™ peroxidase substrate (KPL). Viral titers were calculated as FFU per ml of lung tissue homogenate. Blood, spleen and mLNs were prepared as described previously [35]. Lungs were perfused by injecting 5–10 ml PBS into the right heart ventricle. The lungs were cut into small pieces, resuspended in 1 ml RPMI1640 (0. 5% FCS), and digested with 1 ml of RPMI1640 with DNase I (Sigma-Aldrich Chemie) and Collagenase I (ROCKLAND™) in a shaker at 37°C for 30 min. Digested tissue was passed through cell strainers and single cell suspensions were washed with RPMI1640, centrifuged at 500x g for 10 min. Subsequently, the cells were resuspended in 7 ml of 40% Easycoll solution (Biochrom), overlayed onto 6 ml of 70% Easycoll solution in a 15 ml Falcon and centrifuged at 25 min at 1000x g at room temperature. The interface layer was transferred to a 5 ml tube, washed, and resuspended in RPMI1640 (10% FCS). T cells were stimulated with peptides (1 μg/ml) in 85 μl RPMI 1640 for 1h at 37°C, supplemented with brefeldin A (10 μg/ml in 15 μl RPMI 1640) and incubated for additional 5h at 37°C. Cells incubated without any peptide in the same condition were used as negative controls. Cytokine responses were detected by intracellular cytokine staining. Blood cells and lymphocytes from spleen, lung and mLNs were stained with IVL533-541-tetramer-PE and surface antibodies for 30 min, washed with FACS buffer and analyzed. For intracellular cytokine stainings, the cells were first stained with cell surface antibodies for 30 min, washed and fixed with 100 μl IC fixation buffer (eBioscience) for 5 min at 4°C. Following this, cells were permeabilized for 3 min with 100 μl permeabilization buffer (eBioscience) and incubated with anti-IFNγ antibody for 30 min. Afterwards, cells were washed with FACS buffer and acquired using an LSR-Fortessa flow cytometer (BD Bioscience). Mice were intravenously (i. v.) injected with 3 μg anti-CD45-APC/eFluor780 (clone: 30-F11; BioLegend). Mice were euthanatized 3–5 min after injection, and blood, spleen and lungs were collected. Following their isolation from the respective compartment, lymphocytes were stained and analyzed as described above. For systemic in vivo CD8+ T cell depletion, published protocols [71,72] were adopted as follows. BALB/c mice were i. p. injected with 200 μg anti-CD8α (αCD8: clone: YTS 169. 4) or isotype antibody (Rat IgG2b: clone: LTF-2; Bio X Cell) one day before IAV challenge. To deplete mucosal CD8+ T cells in the lungs, BALB/c mice were i. n. administered 10 μg αCD8 or IgG2b in 20 μl of PBS one day before IAV challenge [40]. Mice were sacrificed by CO2 inhalation, the chest cavity was opened and skin and muscle around the neck were gently removed to expose the trachea. A catheter was inserted and the lungs were carefully flushed with 1 mL PBS via the trachea. The BALF was transferred into a 1. 5 ml tube and stored on ice. The BALF was centrifuged at 500x g at 4°C for 10 min. The supernatant was aliquoted and stored at -80°C until further analysis. Mouse IFNγ enzyme-linked immunosorbent assay (ELISA) MAX™ kits (BioLegend) and the bead-based immunoassay LEGENDplex™ Mouse Inflammation Panel (13-plex, BioLegend) were used to quantify IFNγ and other cytokine levels in the BALF according to the manufacturer’s instructions. The bead-based immunoassay LEGENDplex Mouse Pro-inflammation Chemokine Panel (13-plex, BioLegend) was used to quantify multiple chemokine levels in the BALF. Lungs were harvested from BALB/c mice that were latently infected with MCMVWT and MCMVIVL and challenged with IAV during latency. Lungs were fixed in 4% formalin, paraffin embedded, sliced and hematoxylin and eosin (H&E) stained according to standard laboratory procedures. One-way ANOVA analysis was used to compare multiple groups at single time points. Two-way ANOVA analysis was used to compare different groups at multiple time points. Comparisons between two groups were performed using Mann-Whitney U test (two-tailed). Statistical analysis was performed using GraphPad Prism 7.

Vaccines against influenza typically induce immune responses based on antibodies, small molecules that recognize the virus particles outside of cells and neutralize them before they infect a cell. However, influenza rapidly evolves, escaping immune recognition, and the fastest evolution is seen in the part of the virus that is recognized by antibodies. Therefore, every year we are confronted with new flu strains that are not recognized by our antibodies against the strains from previous years. The other branch of the immune system is made of killer T cells, which recognize infected cells and target them for killing. Influenza does not rapidly evolve to escape T cell killing; thus, vaccines inducing T-cell responses to influenza might provide long-term protection. We introduced an antigen from influenza into the murine cytomegalovirus (MCMV) and used it as a vaccine vector inducing killer T-cell responses of unparalleled strength. Our vector controls influenza replication and provides relief to infected mice, but only if we administered it through the nose, to activate killer T cells that will persist in the lungs close to the airways. Therefore, our data show that the subset of lung-resident killer T cells is sufficient to protect against influenza.

Abstract Introduction Results Discussion Materials and methods

blood cells flow cytometry cell motility medicine and health sciences immune cells immune physiology pathology and laboratory medicine spleen pathogens immunology microbiology orthomyxoviruses viruses rna viruses cytotoxic t cells antibodies research and analysis methods immune system proteins influenza a virus white blood cells spectrum analysis techniques animal cells proteins medical microbiology t cells microbial pathogens immune response spectrophotometry chemotaxis biochemistry cytophotometry cell biology influenza viruses viral pathogens physiology chemokines biology and life sciences cellular types organisms

2019

Mucosal CD8+ T cell responses induced by an MCMV based vaccine vector confer protection against influenza challenge

With the rapidly increasing abundance and accessibility of genomic data, there is a growing interest in using population genetic approaches to characterize fine-scale dispersal of organisms, providing insight into biological processes across a broad range of fields including ecology, evolution and epidemiology. For sexually recombining haploid organisms such as the human malaria parasite P. falciparum, however, there have been no systematic assessments of the type of data and methods required to resolve fine scale connectivity. This analytical gap hinders the use of genomics for understanding local transmission patterns, a crucial goal for policy makers charged with eliminating this important human pathogen. Here we use data collected from four clinics with a catchment area spanning approximately 120 km of the Thai-Myanmar border to compare the ability of divergence (FST) and relatedness based on identity by descent (IBD) to resolve spatial connectivity between malaria parasites collected from proximal clinics. We found no relationship between inter-clinic distance and FST, likely due to sampling of highly related parasites within clinics, but a significant decline in IBD-based relatedness with increasing inter-clinic distance. This association was contingent upon the data set type and size. We estimated that approximately 147 single-infection whole genome sequenced parasite samples or 222 single-infection parasite samples genotyped at 93 single nucleotide polymorphisms (SNPs) were sufficient to recover a robust spatial trend estimate at this scale. In summary, surveillance efforts cannot rely on classical measures of genetic divergence to measure P. falciparum transmission on a local scale. Given adequate sampling, IBD-based relatedness provides a useful alternative, and robust trends can be obtained from parasite samples genotyped at approximately 100 SNPs. Molecular tools show great promise for helping us understand and contain the spatial spread of pathogens, and the application of population genetic approaches to monitoring and controlling infectious diseases is becoming routine. Routes and volumes of non-sexually recombining pathogens, such as the influenza and Ebola viruses, can be tracked using genomic surveillance [1], enabling time-calibrated phylogenies [2], which can be spatially projected [3], and used to jointly estimate transmission chains [4–7]. For sexually recombining pathogens such as the human malaria parasite Plasmodium falciparum, however, these methods are not readily applicable, especially on local spatial scales relevant for control and elimination strategies [8]. Furthermore, P. falciparum evolves more slowly than viral pathogens, and non-sampled asymptomatic infections, complex within-host dynamics, and extensive within-host diversity (multiple-genotype infections) obscure inference [5,7, 9]. As a result, despite increasing efforts to collect genomic data for epidemiological surveillance of malaria on local spatial scales, methods for making sense of them, and guidelines for study design, are lacking. On large or continental scales, or where recombination is limited, P. falciparum genetic data have been usefully employed to identify spatial relationships between parasite populations using standard approaches [10]. For example, microsatellite analyses have been used to infer the origins of drug resistant genotypes [11–13] or outbreaks [14], to monitor population dynamics [15,16] and to explore population structure in West Africa [17]; single nucleotide polymorphisms (SNPs) in non-recombining mitochondrion and apicoplast genomes have been used to infer the evolutionary trajectory of the parasite [18]; and whole genome data have been used to interrogate population structure across continents and within Southeast Asia, including Cambodia, the epicenter of drug resistant malaria [19–22]. Explicitly spatial methods applied to these data include tests and scans of spatial autocorrelation [23,24], which are typically suited to highly resolved geo-referenced data. Non-spatial methods include principal component and phylogenetic analyses, as well as many model-based Bayesian methods, including STRUCTURE [25], ChomoPainter and fineSTRUCTURE [26]. Measuring connectivity among proximal populations of P. falciparum is more challenging, however [27,28]. Classical measures include Wright’s fixation index (FST) [29,30], a measure of divergence between population pairs, which has been used to recover large-scale population structure in malaria [17,21], but has been shown to be less reliable at smaller spatial scales [20]. More recent studies have investigated relatedness using identity by descent (IBD) and identity by state (IBS), with some promise for smaller spatial scales. Henden and colleagues, for example, constructed networks of related parasites within and across countries using IBD inferred under a probabilistic model that accounts for recombination [31]. On a micro-geographic scale, Omedo and colleagues reported trends in relatedness using IBS [32], which approximates IBD [33], and has been used elsewhere to infer relatedness between malaria parasites [34]. These studies emphasize the need for tools on local scales that can account for transmission between local hotspots, particularly in areas considering or implementing elimination programs, and suggest that IBD-based measures are promising approaches. IBD is a fundamental concept in population genetics, relating ancestry to variability due to recombination [35]. FST can also be interpreted as a measure of IBD stemming from remote inbreeding [36], but unlike IBD, FST relies on allelic variation providing a traceable history of co-ancestry. Since recombination works on shorter times scales than mutation and genetic drift, estimates of IBD provide insight into more recent demographic events than FST [35], and IBD-based analyses have been used extensively in human genetics (e. g. to impute genotypes, to map disease loci, and to infer demographic histories [35,37]). Increasingly, it is thought that much of the useful signal in the malaria genome lies in the pattern of recombination, rather than variation at any one locus, and IBD is gaining popularity in malaria research and policy (e. g. to monitor disease transmission [38], relatedness within multiple-genotype infections [39], to aid surveillance of antimalarial resistance [40], and to detect signals of selection [31]). To explore the utility of IBD for estimating connectivity between very local parasite populations, we analyzed one of the largest joint data sets of both genotyping and sequencing data, collected between 2001–2014 from four Shoklo Malaria Research Unit (SMRU) clinics on the Thai-Myanmar border (Fig 1). The border is characterized by mobile migrant populations, villagers, and refugees from Myanmar, and is an area of low and declining malaria transmission [41–43]. This region is therefore representative of many near-elimination settings where remaining pockets of transmission are often found in border areas, and where human mobility is potentially difficult to measure for political or logistic reasons. Here, we focus specifically on measures that capture connectivity: FST between population pairs and relatedness between parasite sample pairs. We show that, unlike FST, IBD-based relatedness decreases significantly over inter-clinic distance. Importantly, where a tradeoff must be made between sequencing effort and sample sizes, we show that robust spatial trends can be recovered using 93-SNP barcodes, providing a cheap and simple approach to implementing these analyses in the field. First we explored spatial structure between parasites collected from different clinics on the Thai-Myanmar border using FST, a standard measure of divergence between populations, and one that has been applied frequently in the context of malaria. FST estimates were calculated using Hudson’s estimator [47–49], which is recommended for small and unequal sample sizes [49,50]. Estimates based on barcode and WGS data across all available years were statistically different from zero with p-values < 0. 001 (Tables A and B of S1 Text). Those based on barcode data were low (Fig 2), indicative of migration between populations. Those based on WGS data were an order of magnitude larger (Fig 3), but there was no evidence of spatial trends between clinics based on either barcode or WGS data (Table 1). We observed a positive correlation between FST estimates and within-clinic relatedness based on IBD (Fig G in S1 Text), and this appears to explain not only the difference in estimates using barcode versus WGS, but also the lack of spatial trend (S1 Text). We define relatedness using the expected fraction IBD, π^IBD, a probabilistic measure of the fraction of the genome that a pair of parasites inherited from a recent common ancestor [52]. For a given pair of clinics (e. g. Maela and Wang Pha) we obtained a single FST estimate versus many π^IBD (nMaela × nWang Pha, where n denotes the number of parasite samples per clinic). For comparison with FST estimates, we plotted proportions of highly related parasite sample pairs (those with π^IBD>0. 5) within and across clinics (Figs 4 and 5 and Fig O and P of S2 Text). However, to leverage the wealth of information across the many parasite sample pairs, spatial trends were estimated using individual π^IBD. Specifically, we regressed highly related parasite sample pair labels (equal to one if π^IBD>0. 5 and zero otherwise) onto spatial and temporal predictors within a logistic regression framework (see Materials and methods). Unlike FST, IBD-based relatedness decreased with inter-clinic distance (Figs 1,4 and 5), even after adjusting for heterogeneous temporal sampling within the regression model (Tables 2 and 3). Considering barcode data collected from 2001–2010, highly related parasite sample pairs were negatively associated with distance (km) both before (βunadjusted ΔDistance = -0. 026, p-value = 0. 002) and after (βadjusted ΔDistance = -0. 023, p-value = 0. 002) adjusting for temporal differences between parasite sample collection dates (Table 2). The spatial trend was of the same order as the temporal trend (βadjusted ΔWeeks = -0. 021, p-value = 0. 002). The impact of distance decreased with time, but the interaction was very small (βadjusted ΔWeeks × ΔDistance = 0. 0001, p-value = 0. 002). Importantly, the spatial and temporal trends were also negative upon exclusion of repeat barcodes within clinics (Table C of S2 Text). WGS data showed similarly negative spatial trends (Fig 5 and Table 3). Since contemporaneous WGS data from all four clinics was only available in 2014, we excluded prior years in the regression model, and found significant negative trends both before (βunadjusted ΔDistance = -0. 023, p-value = 0. 002) and after (βadjusted ΔDistance = -0. 026, p-value = 0. 002) adjustment for temporal differences between parasite sample collection dates. The trend based on data across all years was also significant, but only after adjustment for an overall increase in highly related parasite sample pairs in 2014 (βadjusted 2014 ΔDistance = -0. 020, p-value = 0. 035). IBD-based analyses recovered spatial and temporal trends where FST did not. Together with evidence of phenotypic differences in parasites across SMRU clinics [44], our results suggest IBD-based analyses are sensitive to local spatial genetic structure. To further validate these findings, we used ChromoPainter [26] to estimate average numbers of segments donated and received between parasites from different clinics, hereafter referred to as clinic-averaged co-ancestries (S3 Text). Like IBD-based analyses, ChromoPainter accounts for recombination [26], and it has been used to study malaria parasite populations in Cambodia [20]. Clinic-averaged WGS co-ancestry estimates showed a similar pattern as proportions of highly related parasite pairs, declining with inter-clinic distance (Fig B in S3 Text), and thereby supporting the spatial genetic structure observed in our IBD-based analyses. Clinic-averaged barcode co-ancestry estimates declined with distance only when considering both within-clinic and inter-clinic estimates (Fig D in S3 Text). Plots of pairwise estimates suggest that differences between inter-clinic averaged barcode co-ancestry estimates were unresolved because the range of estimates was narrow compared with π^IBD (Fig E in S3 Text), despite positive correlation with π^IBD (Fig F in S3 Text). This is expected, however, since ChromoPainter is not intended for sparse barcode data. To assess the sample sizes required to measure connectivity between proximal sites using IBD, we re-estimated trends using random subsets of the data across all years under temporally adjusted logistic regression models. Subsets ranged in size from 50 to 1171 barcode parasite samples, and from 50 to 176 WGS parasite samples. We also considered the impact of sequencing fewer SNPs, since many studies use a 24-SNP barcode (e. g. [38]). We use an ideal set of 24 SNPs with high minor allele frequency (Fig A in S4 Text), following the experimental design of a molecular barcode [53]. Our 24-SNP barcode results therefore represent a best-case scenario; “true” barcodes, which are constructed a priori, will almost surely deviate from this ideal due to spatiotemporal variations in minor allele frequencies. Fig 6 shows the relationship between sample size and significant negative spatial trends observed for different sequencing approaches. We estimated that approximately 147 WGS parasite samples, 222 93-SNP barcode parasite samples, and 344 24-SNP barcode parasite samples were sufficient to recover significant negative trends 95% of the time (Table 4). It is important to note, however, that spatial trend estimates based on only 24-SNPs converged to -0. 004, whereas equivalent estimates based on 93 or more SNPs converged to -0. 023 and -0. 020, respectively (Tables 2 and 3). The comparatively low spatial trend estimate based on only 24 SNPs was not unexpected. Previous studies have shown barcode size limits resolution of low genome wide identity [38], while simulated data show rapidly decreasing accuracy in π^IBD with fewer SNPs [52]. Moreover, due to the bounded nature of π^IBD, very wide error at low IBD is liable to result in a bias towards 24-SNP π^IBD that exceed genome-wide π^IBD and the 0. 5 threshold for highly related parasite sample pairs (Fig B in S4 Text), leading to poor resolution of spatiotemporal trends. We therefore do not recommend the use of 24 SNP barcodes for IBD-based analyses. The 93 SNP barcode provided a surprisingly robust estimate of geographic structuring, however. Like chromosome painting methods [26], IBD-based analyses capture information based on dependence between SNPs [35]. Although inter-SNP distances are large on the 93-SNP barcode, barcode SNPs are mostly dependent under hmmIBD because the recombination rate is low (Fig C in S4 Text). Despite the potential utility of genetic data for resolving fine-scale differences in connectivity among proximal populations in P. falciparum and other species, there are minimal guidelines about how to quantify gene flow between proximal locations. Here we show that IBD-based relatedness provides a more robust measure of local spatial structure than FST. Moreover, where a trade-off must be made between sample size and sequencing effort, 93 SNPs were sufficient to recover robust spatial trends using relatively few additional parasite samples compared with WGS. This is an important practical insight given the wide availability of historical barcode data, and the relative cost-effectiveness and ease of generating barcode data compared to whole genome sequences. We therefore propose that IBD-based relatedness is not only a useful metric of gene flow between proximal populations, but also that it can be efficiently estimated using 93-SNP barcodes, which are inexpensive and can be generated from parasite DNA extracted from dried blood spots on filter papers. FST estimates were strongly affected by clinics characterized by highly related parasites, and this association appeared to overwhelm spatial trends. With a view to monitoring malaria parasite populations, we consider this apparent sensitivity of FST potentially problematic for its routine use. This is especially true of regions of declining transmission, where fewer infections go together with the emergence of increasingly clonal hotspots. Although IBD-based analyses were not completely impervious to high within-clinic relatedness, they retain their ability to recover spatial trends. Furthermore, since IBD-based analyses allow explicit estimation of within-clinic relatedness, its impact on relatedness across populations can be assessed. We estimated that approximately 147 single-infection WGS parasite samples, or 222 single-infection 93-SNP barcode parasite samples, were required to recover robust spatial trends. In very low transmissions settings, such as those where the number of cases has dropped below the World Health Organization’s pre-elimination threshold of 1 infection per 1000 persons per year, the number of parasite samples required to estimate spatial trends would in many cases exceed the number of cases. Here, π^IBD could still be used to assess relatedness between individual cases and suspected source populations, which may be critical given the World Health Organization’s definition of elimination, which requires no local cases for 3 years, but allows for imported ones. Although our analyses suggest 93-SNP barcodes are sufficient to recover robust spatial trends at the population level, we do not recommend using 93-SNP barcodes for standalone analyses of individual parasite sample pairs due to large expected error in that application ([52] and Fig B in S4 Text). In high transmission settings overall relatedness will likely decrease due to increased recombination. To account for low population-level relatedness, one could genotype more SNPs and decrease the threshold for highly related sample pairs. Ideally one would also use a model capable of estimating IBD from complex samples of multiple-genotype infections, since these are liable to increase in abundance with transmission [54]. Henden et al. recently proposed an IBD model that can support parasite samples with one or two parasite strains [31], although it doesn’t currently output π^IBD directly. Models capable of supporting parasite samples with three or more strains are lacking. Although this combined set of barcode and WGS data is one of the largest of its kind, the sampling design was not intended for the question at hand. However, despite uneven sampling in time and space we find evidence of spatial genetic structure on the Thai-Myanmar border, which is consistent with earlier reports of phenotypic differences between parasites from different clinics [44]. Evidence of spatial structure is also supported by results from an independent method ChromoPainter [26]. Akin to IBD-based analyses, ChromoPainter leverages the wealth of haplotypic information in WGS data, but struggles to resolve variation in 93-SNP barcode data, for which it was never intended. Regardless of the method used, evidence of spatial structure calls for a better understanding of the drivers that sustain spatial trends. Epidemiological models parameterized by human mobility data have been used to estimate the spatial spread of pathogens in some cases [55–58], but data on human migration are difficult to obtain, particularly in sparsely populated areas and in regions near international borders, where there are political sensitivities around measuring migration. Analyses of spatial genetic structure are common beyond malaria (e. g. studies of pollen dispersal [59–62]). Measures used are largely variants of IBS and therefore sensitive to the marker system and reference population [62]. Unlike IBS-based methods, IBD-based methods explicitly account for the marker scheme by conditioning on allele frequencies. They could thus prove useful as IBS surrogates in spatial studies of other recombining organisms [63–66]. In summary, we propose that IBD-based relatedness will prove useful in the malaria field and in other infectious disease systems to compare data collected from local sites, from areas with more complex topologies, and where data are available, to compare human and parasite movement. IBD-based relatedness could also prove useful beyond epidemiological applications to complement spatial analyses of other sexually recombining organisms. The barcode data were generated as part of a longitudinal trial of artemisinin resistance and its genetic heritability [44], then later reanalyzed to identify correlates of declining malaria transmission [45]. Full details of sample collection and laboratory methods can be found in [44] and [45]. Briefly, 1173 filter paper blood spots were collected between 2001 and 2010 from hyper-parasitaemic patients (> 4% infected red blood cells) with uncomplicated P. falciparum malaria presenting at four SMRU clinics on the Thai-Myanmar border (Fig 1). DNA extracted using a two-step protocol was successfully genotyped at 93 SNPs using the Illumina GoldenGate platform. The 93 SNPs were distributed across the P. falciparum genome (Fig A in S4 Text), but not in regions likely under strong selection (supporting information of [45]). In total, 558 parasite samples were considered multiple-infection (containing more than one P. falciparum genotype), based on 6 or more heteroallelic genotyping outcomes [45], while 1173 were considered single-infection. Analyses in this study were based on single-infection parasite samples only (S2 Table). The WGS data were generated from 178 parasite samples collected between 2001–2014 from the same four clinics (S3 Table). Full details of sample collection and sequencing workflow can be found in [40]. Briefly, parasite samples collected prior to 2010 were derived from a single-infection subset of the aforementioned dried blood spots, selected such that no two showed identical 93-SNP genotypes, and sequenced following hybrid selection on an Illumina HiSeq 2500 platform. Parasite samples collected from 2010 onwards were collected as venous blood and directly sequenced on an Illumina HiSeq 2500 platform following leukocyte depletion. As described by Cerqueira and collegues [40], reads were aligned to the P. falciparum 3D7 v3 reference genome, genotypes called and sites filtered. Those removed included heterozygous sites, indels, sites with QUAL < 60, GQ < 30, polymorphic sites located in pericentromeric, subtelomeric and hypervariable regions, and sites occurring in genes belonging to large antigenic gene families. In addition to the sites listed above, we removed 121 sites with reference or alternative allele assignments indicating potential indels, monomorphic sites, sites lacking genotype calls in 20% or more of the parasite samples, and mitochondrial and apicoplast sites, leaving a total of 34911 polymorphic biallelic SNPs. Aside from IBD estimates generated using hmmIBD (v2. 0. 0) [52], and co-ancestry estimates generated by the ChromoPainter package within fineSTRUCTURE version 2 [26], all data analyses were performed in R [67]. P-values less than 0. 05 were considered significant and were calculated by permutation. They were exact if the number of possible permutations was less than 1000, otherwise they were Monte Carlo estimates [68]. Monte Carlo p-values can overestimate true p-values [69]; however, overestimation is small when the number of randomly sampled permutations, n, is large (at least 99 [68]). We use n = 100 when assessing the sensitivity of spatial trends to sample size (see below) and in sensitivity tests (Fig Q in S2 Text), otherwise n = 1000. All p-values were two-tailed, with the exception of those for FST estimates (Tables 1 and 2 of S1 Text), since FST is non-negative. Two-tailed p-values were calculated by summation over left and right-hand tails.

The spatiotemporal dispersal of organisms can inform efforts to conserve endangered species, to contain the spread of drug resistance, and to eliminate disease. As genomic data become increasingly more affordable and accessible via public depositories, the demand for methods capable of extracting fine-scale population structure from genomic data grows. However, to the best of our knowledge, there are no guidelines regarding the type of data and methods required to resolve local spatial trends over sexually recombining haploid organisms, such as the malaria parasite. The approach we present here compares relatedness based on identity by descent, which accounts for recombination while distinguishing genetic identity due to inheritance from genetic identity due to chance, to a classic population genetic measure of divergence, using data from sexually recombining malaria parasites. Using identity by descent, we uncover a significant decrease in highly related malaria parasites collected from proximal clinics on the Thai-Myanmar border, a region where human mobility is high. Our results demonstrate the power of analyses based on identity by descent to detect recent and local trends. Similar analyses could be used to inform the molecular epidemiology of other sexually recombining organisms.

Abstract Introduction Results Discussion Materials and methods

medicine and health sciences parasite groups plasmodium population genetics tropical diseases parasitic diseases parasitic protozoans parasitology ethnicities apicomplexa protozoans mathematics dna population biology discrete mathematics combinatorics malarial parasites thai people people and places biochemistry eukaryota permutation nucleic acids genetics biology and life sciences population groupings malaria dna recombination evolutionary biology physical sciences organisms

2017

Quantifying connectivity between local Plasmodium falciparum malaria parasite populations using identity by descent

Diverse clinical features have been reported in human African trypanosomiasis (HAT) foci caused by Trypanosoma brucei rhodesiense (T. b. rhodesiense) giving rise to the hypothesis that HAT manifests as a chronic disease in South-East African countries and increased in virulence towards the North. Such variation in disease severity suggests there are differences in host susceptibility to trypanosome infection and/or genetic variation in trypanosome virulence. Our molecular tools allow us to study the role of host and parasite genotypes, but obtaining matched extensive clinical data from a large cohort of HAT patients has previously proved problematic. We present a retrospective cohort study providing detailed clinical profiles of 275 HAT patients recruited in two northern foci (Uganda) and one southern focus (Malawi) in East Africa. Characteristic clinical signs and symptoms of T. b. rhodesiense infection were recorded and the degree of neurological dysfunction determined on admission. Clinical observations were mapped by patient estimated post-infection time. We have identified common presenting symptoms in T. b. rhodesiense infection; however, marked differences in disease progression and severity were identified between foci. HAT was characterised as a chronic haemo-lymphatic stage infection in Malawi, and as an acute disease with marked neurological impairment in Uganda. Within Uganda, a more rapid progression to meningo-encephaltic stage of infection was observed in one focus (Soroti) where HAT was characterised by early onset neurodysfunction; however, severe neuropathology was more frequently observed in patients in a second focus (Tororo). We have established focus-specific HAT clinical phenotypes showing dramatic variations in disease severity and rate of stage progression both between northern and southern East African foci and between Ugandan foci. Understanding the contribution of host and parasite factors in causing such clinical diversity in T. b. rhodesiense HAT has much relevance for both improvement of disease management and the identification of new drug therapy. Human African trypanosomiasis is caused by infection with African trypanosomes, which are transmitted by the haematophagous tsetse fly. Trypanosoma brucei gambiense (T. b. gambiense) HAT is present in West and Central Africa and T. b. rhodesiense HAT is found in East and Southern Africa. The disease is characterised by two stages, the early or haemo-lymphatic stage where trypanosomes proliferate at the bite site, travel to local lymph nodes and establish infection in the bloodstream, and the late or meningo-encephalitic stage in which trypanosomes invade the central nervous system (CNS) leading to coma and death if untreated [1], [2]. HAT has a severe social and economic impact across sub-Saharan Africa today with an estimated 60 million people at risk of African trypanosome infection and 70,000 people are infected [3]. T. b. rhodesiense infection is typically described as an acute disease with rapid progression to late stage infection, however, a wide range of disease pathologies have been reported for many years in East Africa from asymptomatic carriers and mild chronic infections with incubation times of several months in Zambia [4], [5], [6], [7] to accounts of acute infections causing severe disease pathology within 4–6 weeks [8] and 80% of deaths within 6 months [9] in the Busoga focus of Uganda/Kenya. Analysis of historic T. b. rhodesiense epidemics in East Africa led to the proposal that HAT spread from Zambia to Tanzania and Uganda increasing in acuteness and virulence towards the north [10]. However, this is unlikely as T. b. rhodesiense, not T. b. gambiense, is now thought to be responsible for the first recorded HAT epidemic in 1900 which occurred in the Busoga focus [11], [12]. Variation in HAT disease severity between foci suggests that there is genetic variation in trypanosome virulence and/or differences in host susceptibility to trypanosomiasis. Isoenzyme and minisatellite analysis of T. b. rhodesiense isolates from the Busoga focus and the Luangwa Valley focus in Zambia have confirmed distinct parasite genotypes between northern and southern East African HAT foci [13], [14], [15]. Correlation of parasite genotypes with different clinical profiles, however, is limited although analysis of a small number of T. b. rhodesiense isolates from within the Busoga focus (1989 to 1993) identified two main zymodemes (the ‘busoga’ and ‘zambezi’ zymodemes), which were associated with different clinical manifestations [16]. There is also evidence for differences in host susceptibility to T. b. rhodesiense infection. In Zambia different clinical profiles were associated with tribal group and previous HAT exposure [6], [4]. In addition, there are accounts of acute disease pathology in non-local individuals with HAT compared to the typically chronic disease described in Zambian patients [4], [17], [6], [7], [18], [19]. Recently we reported dramatic variation in disease severity between northern and southern East African HAT patients in Uganda and Malawi respectively [20]. HAT patients recruited in Soroti and Tororo foci in East (E) Uganda followed an acute disease course with progression to meningo-encephalitic infection in 87% of patients. In contrast, HAT cases from the Nkhotakota district Central (C) Malawi presented with a mild disease and trypanosome CNS invasion was only diagnosed in 7% of recruits even after several months' incubation. These disease phenotypes were associated with both different host inflammatory cytokine responses and different parasite serum resistance associated (SRA) gene polymorphisms. In addition, differences in disease pathology were also observed between HAT cases recruited in two geographically close foci (Tororo and Soroti) within Eastern Uganda [21]. A higher proportion of Tororo focus HAT cases had progressed to CNS infection, displayed more severe neurological dysfunction and higher plasma IFN-γ production than Soroti focus cases. Trypanosome isolates were found to be genetically distinct, thus, demonstrating that distinct parasite genotypes circulating in the field in spatially separated foci within Uganda are associated with differences in virulence. We also hypothesised that the magnitude of host systemic IFN-γ production determines the post-infection time at which trypanosomes invade the CNS in acute HAT, while the ‘mild’ and ‘severe’ phenotypes specific to northern and southern East African foci, which are geographically much further apart, involve differing abilities to down-regulate TNF-α mediated pathology by TGF-β. These studies suggest that both genetic variation in T. b. rhodesiense virulence and differences in host resistance to infection play key roles in HAT disease severity. To examine the influence of parasite and host genetics on HAT severity within and between foci it is essential to determine clinical profiles for HAT in southern and northern East African foci. Here we present an extensive clinical study of T. b. rhodesiense HAT in two northern and one southern East African foci (Figure 1). The two northern (Uganda) foci are 150 km apart and are referred to as Tororo and Soroti foci. The Tororo focus is part of the historic Busoga focus, while HAT cases were first detected in the Soroti focus in December 1998 [22], [23]. G. f. fuscipes is the vector for transmission in both ‘northern’ foci. The ‘southern’ focus was Nkhotakota district of C Malawi. Sleeping Sickness has been recorded here since 1911 and is it associated with transmission by G. m. morsitans and G. pallidipes in the Nkhotakota Wildlife Reserve and the Kasungu National Park [24], [25]. We have documented detailed clinical histories of each patient, recorded HAT signs and symptoms on admission, analysed the rate of stage progression, determined parasite burden and CSF WBC counts, and assessed the influence of co-infections on HAT clinical profiles. This study was conducted according to the principles expressed in the Declaration of Helsinki. All patients recruited received written and verbal information explaining the purpose of this study and gave informed consent. The ethical committees in Uganda (Ministry of Health), Malawi (College of Medicine) and the UK (Grampian Joint Ethics Committee) approved all protocols. Ethical consent forms were designed in English and also translated into local languages. Consent was given as a signature or a thumb print after verbal explanation. For those under 16 consent was given by their legal guardian, and for those whose clinical condition prohibited full understanding of the recruitment process, consent was gained from a spouse or other family member. Blood and CSF analysis was carried out on admission and on discharge as part of normal diagnostic and follow-up procedures. Data analyses were carried out using Statview and JMP (SAS Institute, Cary, North Carolina) statistical packages. The distribution of continuous variables was assessed for normality using the Shapiro-Wilks W test and Levene' s test for homogeneity of variance between groups, standard transformations were applied as appropriate [38]. Parametric statistics were used to analyse normally distributed continuous data including the unpaired Student' s t test for variation between groups, the paired t test for matched groups and analysis of variance (ANOVA) for comparing more than two groups in conjunction with the Tukey-Kramer post-hoc test. Non-parametric statistics were used when data was not normally distributed and could not be transformed, these included the Mann Whitney U test to compare two groups, the Wilcoxon signed rank test for matched groups and the Kruskal-Wallis test when more than two groups were being compared. Additionally, Spearman' s Rho was used to assess relationships between continuous variables and Fisher' s exact test was used to test for associations between discontinuous variables. To address missing data, the exact number of cases analysed for each parameter measured is given. To establish which clinical observations were indicative of early and late stage T. b. rhodesiense infection the most frequently observed signs and symptoms in each focus (present in more than one third of all HAT patients examined on admission) are summarised in Table 5. Mild anaemia was commonly diagnosed in both haemo-lymphatic stage and meningo-encephalitic stage T. b. rhodesiense infections in all three foci. In both Ugandan foci chancre was a frequent presenting sign of early stage HAT, however, Soroti early stage HAT was characterised by tremors, somnolence and walking only when aided pointing to early onset neuropathology, while Nkhotakota early stage HAT was characterised by hepatomegaly, splenomegaly and fever. Although somnolence was indicative of late stage HAT in both Ugandan foci only, severe neurological dysfunction was particularly prevalent in Tororo focus HAT. Clear differences in HAT disease profiles, therefore, were observed between distant HAT foci and those within close proximity. Parasitaemia (Table 7) did not significantly differ in early stage HAT cases between Soroti and Tororo Ugandan foci (no parasitaemia data was available for Nkhotakota focus cases). Parasitaemia was significantly higher, however, in early stage compared to late stage infections in both Ugandan foci (Mann Whitney U test, p<0. 05 & 0. 01 respectively). CSF parasite levels did not significantly differ between Ugandan foci (no data was available for Nkhotakota cases). CSF WBC counts were significantly elevated in Tororo late stage cases compared to Soroti patients (Mann Whitney U test, p<0. 0001, (Nkhotakota cases also presented with higher CSF WBC counts but this was not significant due to the low number of late stage cases). CSF WBC count was positively correlated to CSF parasite levels in late stage HAT cases in Tororo (Spearman' s Rho 0. 312, p<0. 05) and Soroti foci (Spearman' s Rho 0. 532, p<0. 0001), however, only high CSF parasite levels was found to be associated with a fatal outcome in both Ugandan foci (Mann Whitney U test, p<0. 01). HAT cases with malaria, schistosomiasis and filariasis were excluded from this study, however, those with hookworm co-infection were recruited. Hookworm co-infection was not observed in the Nkhotakota focus and only 1% of Soroti focus HAT cases presented with hookworm infection. In Tororo focus HAT cases hookworm was not highly prevalent (13%), however, it was more frequently observed than in other foci (Fishers Exact Test, p<0. 0001). Hookworm co-infection was not correlated with HAT stage of infection, infection duration or fatal outcome and did not account for the high prevalence of anaemia observed in HAT cases (Table 3). Overall, a comparison of clinical HAT profiles between foci established marked differences in disease severity between Uganda and Malawi demonstrating that T. b. rhodesiense HAT manifests as a chronic disease in Malawi with a low prevalence of neurological symptoms. Within Uganda, HAT is an acute disease with marked neurological involvement contrasting with the clinical profiles previously documented in SE Uganda. In addition, in Soroti a more aggressive disease course was observed with extremely early onset neurological involvement, although neurological deterioration associated with the final stages of HAT were more frequently observed in late stage patients in the Tororo focus. Establishing distinct clinical phenotypes in different HAT foci can improve disease management by focusing control and surveillance in regions where human disease progression and severity are most extreme. It is now critical to determine the mechanisms involved in causing this clinical diversity.

T. b. gambiense and T. b. rhodesiense cause human African trypanosomiasis (HAT). These parasite subspecies differ in their rate of progression to central nervous system (CNS) infection, and consequently to coma and death. Variation in disease progression and severity has also been documented between northern and southern East African T. b. rhodesiense HAT. However, it is unknown if this is caused by differences in patient susceptibility to infection, genetic variation in parasite virulence, or both, as despite the existence of good molecular tools, previous studies have involved limited numbers of HAT cases. In this paper we present extensive clinical data on T. b. rhodesiense cases and describe robust clinical profiles for three disease foci. Common presenting symptoms were identified. We also describe marked differences in disease progression and severity both between Ugandan and Malawi foci, and between two Ugandan foci (Tororo and Soroti) giving rise to three foci-specific clinical phenotypes ranging from a chronic haemo-lymphatic stage infection in Malawi, to rapid disease progression and neurological dysfunction in Soroti and severe neuropathology in Tororo cases. Most importantly, we have now established clinical focus-specific phenotypes that will be available to supplement host and parasite genetics studies to determine their contribution to HAT disease virulence.

Abstract Introduction Materials and Methods Results Discussion

infectious diseases/protozoal infections infectious diseases/neglected tropical diseases infectious diseases/epidemiology and control of infectious diseases

2010

Focus–Specific Clinical Profiles in Human African Trypanosomiasis Caused by Trypanosoma brucei rhodesiense

Bovine tuberculosis is a zoonotic disease with largely unknown impact in Africa, with risk factors such as HIV and direct contact with animals or consumption of Mycobacterium bovis infected animal products. In order to understand and quantify this risk and design intervention strategies, good epidemiological studies are needed. Such studies can include molecular typing of M. bovis isolates. The aim of this study was to apply these tools to provide novel information concerning the distribution of bovine tuberculosis in cattle in Mozambique and thereby provide relevant information to guide policy development and strategies to contain the disease in livestock, and reduce the risk associated with transmission to humans. A collection of 178 M. bovis isolates was obtained from cattle in Mozambique. Using spoligotyping and regions of difference analysis, we classified the isolates into clonal complexes, thus reporting the first characterisation of M. bovis strains in this region. Data from MIRU-VNTR typing was used to compare isolates from a number of African countries, revealing a deeply geographically structured diversity of M. bovis. Eastern Africa appears to show high diversity, suggesting deep evolution in that region. The diversity of M. bovis in Africa does not seem to be a function of recent importation of animals, but is probably maintained within each particular region by constant reinfection from reservoir animals. Understanding the transmission routes of M. bovis in Mozambique and elsewhere is essential in order to focus public health and veterinary resources to contain bovine tuberculosis. Bovine tuberculosis (BTB) is an infectious disease caused by Mycobacterium bovis that affects cattle, other domesticated animals and many free ranging or captive wildlife species. BTB is of global concern on at least three socio-economic levels: the negative impact on animal production; the potential spread to wildlife species; and the risk of zoonotic tuberculosis in humans [1]. BTB has a worldwide distribution with very low prevalence in most industrialized countries, although eradication has been claimed for a few countries only. Factors such as poor or no BTB veterinary control, consumption of uninspected raw meat and/or milk, difficult access to medical care, high prevalence of HIV/AIDS and malnutrition contribute to the increased risk for exposure and susceptibility of humans to M. bovis [2–4]. The maintenance of the pathogen in symptomatic and asymptomatic animals and in the environment creates conditions for dissemination not just to humans but also to a wide variety of wildlife species including endangered ones with the obvious negative consequences on conservation and tourism [5]. While BTB is known to be widespread in Africa, limited or outdated information exists concerning its precise distribution and incidence. In 1993, the World Health Organization (WHO), with the participation of the Food and Agriculture Organization (FAO), convened a meeting on BTB, where the worldwide public health significance of M. bovis in humans and animals was discussed. It was then concluded that data collected from most African countries, mainly from sub-Saharan Africa, were insufficient to reveal the true epidemiological picture of the disease, and it was recommended that collection of data on BTB should be prioritized [6]. Several small studies have been done, but the lack of detailed information in the vast majority of African countries is still of concern [1], particularly since the burden of BTB might be considerably underestimated in humans [7]. There are estimates suggesting that approximately 85% of cattle and 82% of the human population of Africa live in areas where BTB is either not controlled or only partially controlled [8]. This reflects the situation in Mozambique where efforts to improve BTB control are gradually being put into place. Implementation of a test and slaughter policy for cattle for the entire country would be the ideal strategy, but it is unrealistic at present, given resource limitations. Efforts to determine to what extent human tuberculosis cases in Mozambique are caused by M. bovis are ongoing [9]. By reviewing relevant scientific literature and public health reports, Müller and colleagues estimated the incidence of M. bovis, to be a median of 2. 8% (range 0 to 37. 7%) of total human tuberculosis cases in Africa [1], however more robust estimates are needed. According to WHO, over 70 thousand cases of tuberculosis in Africa in 2015 were caused by M. bovis [10], however, the absence of routine reports in the majority of countries imply considerable uncertainty for this estimate. Scattered information from the past and results from ongoing studies in specific regions show that the prevalence of BTB varies widely from region to region and within each region. For example in the Govuro District, in the Southeast of Mozambique, 39. 6% of cattle were skin test positive reactors for BTB [11]; while the disease was practically absent (0. 98%) in cattle in the Limpopo National Park in the South West of Mozambique [12]. Although it is known that the differences in the BTB prevalence are associated with several risk factors, recent results from Govuro District clearly show that control of animal movement appears to be of high importance [11]. In that study, a high prevalence of BTB was observed in almost all livestock areas where small scale farming was practiced, in sharp contrast to that observed in the commercial sector, where no skin test positive animals were detected. While in the commercial sector animals are normally tested for BTB and kept in quarantine before being introduced, trading of animals among small-scale farmers is done without previous information concerning the BTB status of the animals. It is also in these familiar settings that close contacts with animals still occur, particularly for young children who manage daily livestock herding. In addition, due to their stature, children who herd animals may be highly exposed to M. bovis airborne transmission from infected animals [11]. It is increasingly clear that genotyping of M. bovis strains is a useful tool to identify possible transmission routes of BTB, and that the genotype of M. bovis isolates is largely dependent on the geographical area. Genotyping tools have already been widely used in Africa [13–25] but not in Mozambique. We therefore obtained 178 M. bovis isolates from 8 Mozambican provinces and genotyped them using spoligotyping in the entire sample set and MIRU-VNTR and RD deletion testing in a subset. Additionally, we collected a comparative dataset from other African countries (based on spoligotyping patterns and MIRU-VNTRs) that allowed us to place the Mozambican strains in the African context. The analyses allowed us to address the following questions: is there a specific Mozambican M. bovis genetic pattern; do the M. bovis strains reveal cattle importation from neighbouring countries; is there evidence of movement of particular strains across districts within Mozambique; what is the position of Mozambique in the African scenario concerning M. bovis? We obtained M. bovis isolates from 2007 to 2013, from samples collected during a BTB prevalence study (n = 228) and from samples sent to the Central Veterinary Laboratory in Maputo by district veterinary officers and farmers that suspected BTB at post mortem, either in the process of standard meat inspection or necropsy (n = 220) (Table 1). The samples from the prevalence study were from both small-scale and commercial herds of 10 selected districts, from the Provinces in the South (Manhiça, Magude, Chibuto and Govuro), Centre (Buzi, Mutarara and Gondola) and North (Mechanelas, Mogovolas and Angoche) of Mozambique. There is an overrepresentation of samples from the South due to a pilot study using test and slaughter performed in one district (Govuro), on two commercial farms that volunteer to cull the majority of the positive animals and send samples to the Central Veterinary Laboratory that is located also in the South. A list of all herds from the respective districts was supplied by the provincial Veterinary Services department. A herd was defined as the group of animals from a commercial farm or a combination of animals owned by small-scale farmers sharing a dip tank or crush for regular veterinary assistance. Three localities per district were randomly selected. In commercial farms with fewer than one hundred animals the whole herd was tested. In commercial herds with more than one hundred animals, a random sample of at least one hundred animals older than 6 months was selected. High prevalence of BTB cases based on previous skin testing results together with good logistic conditions for the accomplishment of the work were the prime factors for district selection. Based on a strong response to M. bovis tuberculin purified protein derivative (PPD), animals were purchased and slaughtered. In two commercial farms (in Manhiça and Manica provinces), where skin test reactor cows could not be slaughtered, milk was retrieved and tested for the presence of M. bovis by microbiological culture. Institutional permission to conduct the study was obtained from the National Directorate of Veterinary Services in Maputo, Mozambique (Nota 162/ MINAG/DNSV/900/2013) and the AUC (Animal Use Committee) of Stellenbosch University (SU-ACUM13-00009). Sampling and culling was performed as part of the Veterinary Services regular activity for disease control, following the procedures determined by the Mozambican Animal Health Regulation. The slaughter was done in registered abattoirs according to stipulated procedures. All mycobacterial cultures were performed in the National Tuberculosis Reference Laboratory, Ministry of Health Mozambique. Sample data, including the name of the owner and the origin (province and district) of the animals were recorded when available. In Mozambique, for commercial beef production or small-scale farming, the animals are mostly from the local breed Landim and Brahman or crossbreeds between the two. Eight isolates were supplied by another study, four of which were from a milk production farm with Holstein Friesian and Jersey breeds. Maceration of tissue samples was performed in a stomacher apparatus (in duplicate sterile stomacher bags with sterile distilled water). Samples were next decontaminated by adding 4% sodium hydroxide to the same volume of the macerate for 20 min. Supernatant was discarded after centrifugation at 3000 rpm for 20 min. For milk samples, equal volumes of milk and 4% sodium hydroxide were mixed and after 20 min they were centrifuged at 3000 rpm for 20 min. Distilled water was added to the sediment and after agitation and re-centrifugation the sediment obtained was used for inoculation of duplicate tubes with Löwenstein–Jensen medium with glycerol or with pyruvate. Incubation was done at 37 °C for up to 12 weeks. Fifty samples were discarded owing to contamination (Table 1). Isolates were identified as acid-fast bacilli with Ziehl–Neelsen staining (positive samples in Table 1). M. tuberculosis complex (MTC) was confirmed by PCR [26]. Briefly, DNA was extracted using a standardized protocol [27] and PCR amplified with the primers TB1-F 5’-GAA CAA TCC GGA GTT GAC AA-3’ and TB1-R 5’-AGC ACG CTG TCA ATC ATG TA-3’. The PCR protocol started with an initial denaturation step of 95 °C for 10 min, followed by 35 cycles of 95 °C for 1 min, 61 °C for 30 s and 72 °C for 2 min ending with a final step of 72 °C for 10 min. The PCR products were analysed using 1. 5% agarose gels. All isolates that generated a product of around 370 bp were considered as belonging to the MTC. M. bovis was identified by spoligotyping [28] in 170 out of 185 MTC positive isolates tested (Table 1). Genotyping was performed using spoligotyping, region of difference (RD) analysis followed by MIRU-VNTR typing in a subset of the samples. Spoligotyping was done on 178 isolates as described by Kamerbeek and colleagues [28], using membranes and equipment provided with a kit (Isogen Life Science B. V. , Utrecht, The Netherlands, no longer available from this company). This method determines the presence or absence of 43 direct and variable repeat sequences within the direct repeat region, thereby generating spoligotype signatures which are hypothetically characteristic of defined strains [29]. RD analysis was based on PCR genomic analysis to determine the presence or absence of specific regions of difference (RD). It was done on 54 isolates by the assessment of the status of the RD Eu1, RD Af1 and RD Af2 regions. PCR products were visualized after electrophoresis on 1% agarose gels. RD Eu1 was amplified with primers Eu1_FW (5’-CCGATGAACTTGGCCCACAG-3’) and Eu1_Rv (5’–CGTGGTGGTGGGATGTCTTG-3’). A 1206 bp fragment was generated if the RD Eu1 was intact and a 400 bp fragment if the region was deleted [13,14,30]. For RD Af1 two primers targeting the flanking regions of RD Af1 (FW, 5’-ACTGGACCGGC AACGACCTGG-3’, and Rev, 5’-CGGGTGACCGTGAACTGCGAC-3’) and one primer hybridizing with the internal region of RDAf1 (Int Rev, 5’-CGGATCGCGGTGATCGTCGA-3’) were used [13,14] for a band of either 349 bp (intact) or 531 (deleted). For RD Af2 two primers targeting the flanking regions of RD Af2 (RD Af2_Fw, 5’-ACCGCCCTGTCCTATGTGAG -3’, RD Af2_Rev, 5’-TGACGGTTGCCTTTCTTGAC-3’) and one primer hybridizing with the internal region of RD Af2 (RD Af2_IntRev, 5’-CACTGTCTCCGCTCATCATG-3’) were used with bands of 458 bp (intact) and 707 bp (deleted) [14]. MIRU-VNTR typing was done on 59 strains using a standardized 24-locus MIRU-VNTR typing procedure [31]. The analysis was done using the MIRU-VNTR typing kit (Genoscreen, Lille, France). The PCR-products were run with 1200 LIZ size standard (GeneScan, Applied Biosystems) on ABI3500 sequencers. Sizing of the PCR-fragments and assignments of MIRU-VNTR alleles were done with the GeneMapper software version 4. 1 (Applied Biosystems) according to the manufacturers’ instructions, generating a numerical profile for each strain. The genotyping of three extra VNTR markers (MIRU3232, MIRU3336 and MIRU2163a) was done by Genoscreen, France. In order to evaluate relatedness of the spoligotype and VNTR patterns of the different samples we generated a dendogram using the UPGMA algorithm using the tool provided on the site http: //www. miru-vntrplus. org/. The spoligotype patterns of the respective isolates were entered into the Mbovis. org database. In this database each unique spoligotype pattern is named by' SB' followed by a four integer number e. g. SB0120 [32]. To further evaluate the relatedness of the 178 spoligotypes and possible clusters within the data we generated a minimum spanning tree using the tool provided on the site http: //www. MIRU-VNTRplus. org, and the individual allelic diversity was calculated for all 24 MIRU-VNTR loci using the same site. A maximum difference of 1 mutation within a Clonal Complex was considered for the definition of clusters. The index of discrimination [33,34] was calculated to determine the overall discriminatory power of the spoligotyping and MIRU-VNTR typing techniques using a tool provided on the site http: //insilico. ehu. es/mini_tools/discriminatory_power. The number of isolates assigned to each type (spoligotyping and MIRU-VNTR) was introduced in the formula provided. The genotyping of the three extra VNTR markers for increasing the discrimination power of the analysis had the drawback that these three markers could not be used in the MIRU-VNTRplus website. A UPGMA tree was calculated in MEGA [35] following a calculation of genetic distances between genotypes using the software Arlequin [36]. Phylogenetic reconstruction was performed on a larger dataset in order to contextualize the Mozambican diversity to the overall M. bovis diversity in Africa. In order to do so we collected data from 959 M. bovis strains (575 from Central and Western Africa, 104 from Eastern Africa, 6 from North Africa and 274 strains from Southern Africa) [13,14,25,37–44]. The VNTR markers were very divergent between studies but 5 of these were present in most studies and for more than 90% of the samples (viz. MIRU2165, MIRU2461, MIRU577, MIRU580 and MIRU3192 or ETR-A, -B, -C, -D and -E). The set have been suggested to provide enough resolution in the African context [45]. To the data from these, we added the 43 spoligotyping markers. Sample selection was based on a strategy to maximize this resolution (43 spoligotyping markers and 5 VNTRs) and thus studies/samples lacking these markers were excluded. On that premise, all available African M. bovis data in the literature following these parameters was included, although some geographic areas are underrepresented (North Africa) compared to others (Central/Western Africa). Reconstruction was done using a matrix of these 48 markers and by applying the reduced median algorithm [46] followed by the median joining algorithm [47], both present at the network software (freely available at http: //www. fluxus-engineering. com). The algorithm is able to reconstruct missing genotypes allowing some data to be included even if there are missing markers. However to keep this extrapolation to a minimum, only samples that missed a single VNTR marker were included. For comparison two M. tuberculosis samples and two M. caprae samples were extracted from the MIRU-VNTRplus database and included in the analysis. The 178 M. bovis isolates were obtained from samples that originated from 8 out of the 11 Mozambican provinces from the Southern (n = 113), Central (n = 47) and Northern (n = 18) parts of the country, covering 21 of the 128 districts of Mozambique. The isolates originated from 34 small-scale farms with 103 isolates (1 to 6 isolates per farm, an average of about 3 isolates per farm) and 10 commercial farms with 68 isolates (1 to 17 isolates per farm, an average of 7 isolates per farm) (Table 2). The type of farm was unknown for 7 isolates. Five of the commercial farms were located in the South, 4 in the centre and 1 in the North of Mozambique (Table 2). Among the 178 M. bovis isolates, 15 individual spoligotype patterns were identified (S1 Table). All isolates lacked spacers 3,9, 16, and 39 to 43, which is the signature profile of the M. bovis BCG vaccine strain, the ancestral spoligotype pattern of M. bovis (SB0120) as defined by the international M. bovis spoligotype database [www. mbovis. org]). Seven patterns were previously registered in the Mbovis. org database while the other 8 patterns were incorporated into the database with SB numbers (SB2304 to SB2311). The most common spoligotype was SB0961, accounting for more than half (n = 109,61. 2%) of the isolates. Seventeen isolates (9. 5%) had the “BCG-like” spoligotype profile, identical to the spoligotype of the vaccine M. bovis BCG strain. The most frequent spoligotype SB0961 together with 6 others (SB1272, SB2304, SB2307, SB2308, SB2309 and SB2310) were of the “BCG-like derived” type, lacking one or more spacers in relation to SB0120 and not Af1, Af2 or Eu1 typical spoligotype patterns. Of the total isolates, 39 had spoligotypes lacking spacer 11, which is the signature of the Eu1 clonal complex [30]. Of these, 29 had the spoligotype SB0140, which has spacer 6 and 8 to 12 absent. Isolates with some of the previously not described spoligotypes (3 isolates with pattern SB2305,4 isolates with SB2306 and 1 with pattern SB2311) were included in these patterns without spacer 11. None of the isolates had spacer 21 missing, a marker for isolates of the Eu2 clonal complex. None of the isolates had spoligotypes specific to the Af2 clonal complex, lacking spacers 3–7 [14]. The discriminatory index using spoligotyping was 0. 5909. RD analyses were performed to identify M. bovis clonal complexes. The 11 tested isolates with the typical Eu1 spoligotype SB0140 had the Eu1 specific deletion [30]. A possible Af1 spoligotype pattern (absence of spacer 30) did not have the Af1 deletion, a result supported by the phylogenetic analysis (below) that placed this sample as part of the “BCG-like derived” pattern. Additionally 42 isolates with spoligotypes SB0120, SB0140, SB0961, SB2305, SB2308 and SB2310 were tested for the Af1 deletion, 23 spoligotypes with SB0120, SB0140, SB0961, SB2305, SB2308 and SB2310 [13,14] were tested for the Af2 deletion and 14 for the Eu1 deletion (SB0961, SB1099, SB2305 and SB2306) and no deletions were detected. With calculation of the minimum spanning tree based on the spoligotyping results, 4 clonal complexes and 4 singletons (genotypes that are not clustered with other genotypes in the analysis) were identified (Fig 1A). Complexes 1 and 3 consisted of spoligotypes with “BCG-like derived” signatures, complex 1 also included the patterns with the “BCG-like” signature, while complexes 2 and 4 included spoligotypes with Eu1 signature. The M. bovis clonal complex distribution (based on spoligotyping results) according to Mozambican provinces is illustrated in Fig 1B. The most common spoligotype, SB0961, was detected in all of the provinces investigated. Similarly, isolates with the spoligotype SB0120 were present in most provinces, in both commercial and small-scale farms. The 29 isolates with the spoligotype SB0140, likely from the Eu1 clonal complex, were obtained from animals from the South of Mozambique only, [Maputo (n = 22), Inhambane (n = 2) and Gaza (n = 5) ]. Of these 29 isolates, 23 were from 3 commercial farms, 4 have unknown origin (1 from Inhambane, 1 from Maputo and 2 from Gaza provinces) and 2 from two different small-scale farms (from Inhambane province and from Maputo province). The 10 isolates with Eu1 signature other than SB0140 were also all from the south of Mozambique, 4 from commercial farms (SB0290, SB2124, and SB2311) and 6 from small-scale farms in Maputo and Inhambane, except one (SB2305), that was isolated from Sofala province (Centre of Mozambique). Only one of the small-scale farms generated 6 isolates (5 isolates with pattern SB0961 and 1 with SB0120); the other yielded only 1 or 2 isolates that in some cases had identical patterns. In the commercial farms there was a variability of patterns, ranging from 1 to 6 different patterns per farm. To further discriminate the M. bovis isolates defined by spoligotyping and in an attempt to define potential specific links between farms (potentially associated to transmission or common source of infection), 24-locus MIRU-VNTR typing was done on 59 isolates. These isolates were selected based on the better preserved material from which bacteria and/or DNA remained. Amongst the isolates tested, we found a fairly high degree of diversity, which resulted in a splitting of the 59 isolates into 23 different MIRU-VNTR types (Fig 2). The loci MIRU154, MIRU580, MIRU802, MIRU2059, MIRU2401, MIRU2531, MIRU3171, MIRU3192, MIRU3690, MIRU4156 and MIRU4348 showed no variation in all the isolates analysed. From the loci that showed variability, MIRU2165 (ETR-A), MIRU2687 (MIRU24), MIRU2996 (MIRU26), MIRU4052 (QUB26), MIRU3007 (MIRU27) and MIRU2347 (Mtub29) showed higher variability, with allelic diversities ranging from 0. 50 to 0. 43. The discriminatory index using MIRU-VNTR was 0. 8708. Thirty-six isolates with spoligotype SB0961 could be further differentiated into 15 new types, 17 isolates with SB0140 into 5 types and 3 isolates with SB1099 into 2 different MIRU-VNTR types. The Multiple-locus variable-number tandem repeat analysis (MLVA) identified major clusters: “16919–270” shared by 11 isolates of the spoligopattern SB0140 from 2 different commercial farms located in 2 different districts (Manhiça and Magude) and “16911–1358” shared by 15 isolates with spoligopattern SB0961,12 being from various small-scale farms and 1 from a commercial farm from 3 neighbouring districts (Machanga and Buzi from Sofala Province and Govuro from Inhambane province) and the remaining 2 from 2 different districts (Zavala and Angoche). Another cluster of interest was “16917–270” (SB0140) shared by 3 isolates all from different districts (Boane and Macia and Magude), 1 isolate was from a commercial farm, 1 from a small-scale farm and the third was of unknown origin. The introduction of 3 extra VNTR markers (MIRU3232, MIRU3336 and MIRU2163a) resulted in minor changes only in the overall dendogram (S1 Fig), however with an increment of the discriminatory index to 0. 9293. In order to integrate the Mozambican M. bovis data obtained here into the overall African diversity, we generated a general phylogenetic network based on spoligotyping data and 5 VNTRs collected from published manuscripts (S2 Table) [13,14,37–44]. Median networks, the approach presented here to establish the hypothetical links, attempts to establish evolutionary relationships between genotypes, such that nodes in the tree represent ancestral genotypes. The network displays a set of clades, geographically structured across the continent. The comparative genotypes of M. tuberculosis and M. caprae were clearly differentiated from the M. bovis strains (Fig 3). The analysis also allowed the inference of the positioning of Mozambican M. bovis diversity in the general African scenario. One important aspect that emerged from the analysis is that diversity seems deeply geographically structured with sharing of genotypes occurring mostly between neighbouring regions. Mozambican genotypes fall into 3 main monophyletic clades (Fig 3). The first one, labelled clade I in Fig 3, is found across Mozambique and corresponds to the “BCG-like” and the “BCG-like derived” clades mentioned above. This clade is related to a minor branch from Cameroon, but shows deeper phylogenetic relationship to southern African isolates. The second clade (clade II) is related to South African genotypes, including not only genotypes detected in cattle but also in wildlife, and it is only detected in Gaza and Maputo, the 2 southern sampling locations matching the geographical structuring. This clade corresponds to the Eu1-type complex mentioned above. A third minor clade (clade III) with previously unknown spoligotypes SB2305 and SB2306, was detected only in Inhambane and does not appear to be closely related to any other genotypes in the available dataset. This is the most extensive study on genetic diversity of M. bovis isolates from Mozambique. The genetic diversity of M. bovis was integrated in the general diversity of Africa, considering the publicly available strains, in what is the major strain comparative study of the continent to date. Generally, we detected a few clusters that are most likely localised. The dominant clusters are from the “BCG-like” and “BCG-like derived” spoligotype patterns, which are widely spread in Mozambique and neighbouring Southern African countries, but so far absent in the remaining Sub-Saharan Africa [48]. In this study, 39 isolates had the signature of the Eu1 clonal complex. This group of strains is common in the British Isles, the New World, as well as UK’s former colonies and trading partners [30]. The global distribution of these strains suggests a complex history involving recent international cattle trade [48] since Eu1 is rare or absent in the African countries thus far surveyed, except for South Africa [17]. In South Africa, Eu1 strains (SB0140) were identified in samples from cattle [17] and from wildlife [37]. All except one (SB2305) of the 39 isolates with the Eu1 spoligotype pattern were isolated in the South of Mozambique, and the majority were from commercial farms (n = 26), which may have imported infected cattle from South Africa where the Eu1 clonal complex is common [30]. Information collected from the reports of the veterinary services showed that cattle importation to the South of Mozambique in the years 1950 to 1957 were mainly from South Africa (467 animals) and Portugal (59 animals) while in the centre of the country cattle were imported mainly from Zimbabwe (917 animals) but also from South Africa (116 animals). In the North, cattle were imported mainly from Malawi. In the phylogenetic network, many Mozambican isolates represent sub-branches of the South African clades. The Mozambican strains could be placed into only 2 monophyletic clades (clades I and II in Fig 3), meaning at most two introductions of M. bovis, one for each clade. A third minor cluster was identified, with spoligotype patterns SB2305 and SB2306, initially thought to be from the Eu1 clonal complex but lacking the RD Eu1 deletion. These samples formed an independent cluster in the network with no close relationship, possibly suggesting a local deep Mozambican clade. It is interesting to point out that Eastern Africa (Ethiopia, Burundi and Tanzania) seem to display very deep diversity in terms of M. bovis strains in our phylogenetic analysis, suggesting a deeper evolution of M. bovis in that region. These deeper clades likely correspond to the Af2 clonal complex that have been recovered from cattle in East Africa [14], Uganda, Burundi, Tanzania and Ethiopia. However, no isolates with the spoligotype marker of the Af2 clonal complex (absence of spacers 3 to 7) were identified in our sample. Although Tanzania borders Mozambique, our data suggest that cattle movement between the two countries is limited. However sample size from the North of Mozambique where there is a shared border with Tanzania, was very small and one needs to consider that this signal might be present at low prevalence and was not seen in this study. The “BCG-like” strains are dominant in Mozambique, being present from North to South. These strains dominate in most of mainland Europe [48], but also in North Africa (e. g. Algeria [49]), indicating a probable importation of live cattle from Europe (mostly from France) to Algeria. In Zambia the “BCG-like” spoligotype is also dominant [50] with the predominant ancestral spoligotype in cattle [50,51] and wildlife [51]. Thus, M. bovis might have been introduced from Europe into the two neighbouring countries. The positioning of the Mozambican and Zambian branches in the network suggests two independent sources of introduction. The predominant spoligotyping pattern identified in our study, SB0961, was also isolated in Argentina [52], in the Czech Republic and Slovakia [53], but at very low frequencies. It is likely that the genotype was one of the first ones to be introduced in Mozambique, which became common over time. The phylogenetic network suggests that all members derive from a single ancestor. BTB control needs to focus not just on animals but also on humans and their dynamics. Cattle are obviously the most direct source of BTB transmission between themselves and transmission to humans. In Mozambique the overall prevalence of BTB in cattle is 13. 6% [11,12]. In terms of policies, given the current unrealistic likelihood of testing all animals in Mozambique, there is a need to reinforce the regulations that require a negative BTB test result before cattle importation. The same should be enforced for internal movements, since the frequency of shared genotypes from cattle originating from different parts of the country strongly suggests internal transmission of BTB. In that sense, looking at the geographic structuring of the African data in the phylogenetic reconstruction (Fig 3) it is likely that the major current issue in the continent is not spread of genotypes across borders but mainly the maintenance of existing local genotypes in the various countries or regions. In addition to meat consumption of infected animals, one possible source of transmission to humans is milk. We found that analysis of a single milk sample was sufficient to obtain a M. bovis isolate from as many as 9 out of 41 (22%) skin test positive cows. Thus, milk that is frequently consumed with no treatment is on one hand a serious potential source of transmission to calves and humans, and on the other hand, it represents an easily available source of isolates from infected herds to estimate prevalence of BTB. M. bovis infection in humans has been clearly shown in several African countries, as is shown in the network (S2 Fig). The very limited studies done in Mozambique thus far have not shown BTB in humans, but studies are very limited and have not been done on the populations known to be at higher risk. The relevance of the presence of M. bovis in wildlife is often overlooked. BTB has been reported in wildlife in Mozambique [54]. BTB in natural populations is very difficult to control, since it requires a different set of expertise (ecologists, zoologists, ethologists) demanding a deeper collaboration between very different entities. While BTB is a risk for natural populations, they could also represent a reservoir of M. bovis that can re-infect cattle populations [55]. Another issue is that wildlife borders are limited by ecological contexts and not political borders, meaning that although strict measures and surveillance on cattle importation might be in place, wildlife movement, particularly where border fences are dropped for transborder parks, cannot be controlled. The HIV epidemic in Africa, affecting over 25. 5 million individuals, places the population of this continent at higher risk of contracting BTB than elsewhere globally. The impact of BTB as a zoonotic disease in Mozambique remains unknown and no study addressed specifically the prevalence of BTB in risk groups, namely individuals with HIV. Given that BTB has been estimated to range from 0% to 37. 7% of reported human TB cases in Africa [1], we conclude that there is a considerable and underestimated BTB risk to humans in Africa. This is particularly concerning given that the emergence of multidrug-resistant strains of M. bovis has been reported [56,57]. It is therefore of vital importance to continue the efforts made in Mozambique in order to completely characterise and understand the extent of BTB. The information concerning M. bovis presented here represents a foundation stone in that process.

Bovine tuberculosis is a rather neglected zoonotic disease caused by Mycobacterium bovis that is of global concern owing to the persistence of the bacillus in reservoirs that can spread bovine tuberculosis between animals and humans. Africa remains understudied regarding this pathogen, and should be an area of concern given that in many regions the consumption of raw milk or meat from infected animals persists and the presence of HIV infection renders the population more susceptible. In order to control the disease, we need to understand M. bovis epidemiology, which includes the sources of infection. The important conclusion drawn from the work presented here is that there is a strong association between M. bovis genetic characteristics and geography. This implies that the diversity of M. bovis isolates in Mozambique does not seem to be caused by recent introductions to the territory, but is probably maintained within reservoirs in each particular region.

Abstract Introduction Methods Results Discussion

biogeography animal types ecology and environmental sciences milk medicine and health sciences body fluids ruminants population genetics geographical locations vertebrates diet animals mammals nutrition population biology zoology bacteria africa veterinary science geography veterinary diseases actinobacteria phylogeography people and places mozambique eukaryota wildlife anatomy earth sciences beverages genetics physiology biology and life sciences cattle evolutionary biology amniotes bovines organisms mycobacterium bovis

2018

Genetic diversity and potential routes of transmission of Mycobacterium bovis in Mozambique

Software produced for research, published and otherwise, suffers from a number of common problems that make it difficult or impossible to run outside the original institution or even off the primary developer’s computer. We present ten simple rules to make such software robust enough to be run by anyone, anywhere, and thereby delight your users and collaborators. Scientific software is typically developed and used by a single person, usually a graduate student or postdoc [1]. It may produce the intended results in their hands, but what happens when someone else wants to run it? Everyone with a few years of experience feels a bit nervous when told to use another person’s code to analyze their data: it will often be undocumented, work in unexpected ways (if it works at all), rely on nonexistent paths or resources, be tuned for a single dataset, or simply be an older version than was used in published papers. The potential new user is then faced with two unpalatable options: hack the existing code to make it work or start over. Being unable to replicate results is so common that one publication refers to it as “a rite of passage” [2]. The root cause of this problem is that most research software is essentially a prototype, and therefore is not robust. The lack of robustness in published, distributed software leads to duplicated efforts with little practical benefit, which slows the pace of research [3,4]. Bioinformatics software repositories [5,6] catalogue dozens to hundreds of tools that perform similar tasks: for example, in 2016, the Bioinformatics Links Directory included 84 different multiple sequence aligners, 141 tools to analyze transcript expression, and 182 pathway and interaction resources. Some of these tools are legitimate efforts to improve the state-of-the-art, but often, they are difficult to install and run [7,8] and are effectively abandoned after publication [9]. This problem is not unique to bioinformatics or even to computing [2]. Best practices in software engineering specifically aim to increase software robustness. However, most bioinformaticians learn what they know about software development on the job or otherwise informally [1,10]. Existing training programs and initiatives rarely have the time to cover software engineering in depth, especially since the field is so broad and developing so rapidly [4,10]. In addition, making software robust is not directly rewarded in science, and funding is difficult to come by [1]. Some proposed solutions to this problem include restructuring educational programs, hiring dedicated software engineers [4,11], partnering with private sector or grassroots organizations [1,5], or using specific technical tools like containerization or cloud computing [12,13]. Each of these requires time and, in some cases, institutional change. The good news is you don’t need to be a professionally trained programmer to write robust software. In fact, some of the best, most reliable pieces of software in many scientific communities are written by researchers [3,11] who have adopted strong software engineering approaches, have high standards of reproducibility, use good testing practices, and foster strong user bases through constantly evolving, clearly documented, useful, and useable software. In the bioinformatics community, Bioconductor and Galaxy follow this path [12,14]. Not all scientific software needs to be robust [15], but if you publish a paper about your software, it should, at minimum, satisfy these rules. So what is “robust” software? We implied above that it is software that works for people other than the original author and on machines other than its creator’s. More specifically, we mean that: Our rules are generic and can be applied to all languages, libraries, packages, documentation styles, and operating systems for both closed-source and open-source software. They are also necessary steps toward making computational research replicable and reproducible: after all, if your tools and libraries cannot be run by others, they cannot be used to verify your results or as a stepping stone for future work [16]. Version control is essential to sustainable software development [17,18]. In particular, developers will struggle to understand what they have actually built, what it actually does, and what they have actually released without some mechanical way to keep track of changes. They should therefore put everything into version control as soon as it is created, including programs, original field observations, and the source files for papers. Files that can be regenerated as needed, such as the binaries for compiled programs or intermediate files generated during data analysis, should not be versioned; instead, it is often more sensible to use an archiving system for them and store the metadata describing their contents in version control instead [19]. If you are new to version control, it is simplest to treat it as “a better Dropbox” (or, if you are of a certain age, a better FTP) and to use it simply to synchronize files between multiple developers and machines [20]. Once you are comfortable working that way, you should use a feature branch workflow: designate one parallel copy (or “branch”) of the repository as the master, and create a new branch from it each time you want to fix a bug or add a new feature. This allows work on independent changes to proceed in isolation; once the work has been completed and tested, it can be merged into the master branch for release. How to write high-quality documentation has been described elsewhere [21], and so here, we only cover two minimal types: the README and usage. The README is usually available even before the software is installed, exists to get a new user started, and points them towards more help. Usage is a terse, informative command-line help message that guides the user in the correct use of the software. Numerous guidelines exist on how to write a good README file [22,23]. At a minimum, your README should: The program should also print usage information when launching from the command line. Usage provides the first line of help for both new and experienced users. Terseness is important: usage that extends for multiple screens is difficult to read or refer to on the fly. Almost all command-line applications use a combination of POSIX [24] and GNU [23] standards for usage. More standard command-line behaviours are detailed in [8]. Your software’s usage should: Usage should be printed to standard output so that it can be combined with other bash utilities like grep, and it should finish with an appropiate exit code. Documentation beyond the README and usage is up to the developer’s discretion. We think it is very important for developers to document their work, but our experience is that people are unlikely do it during normal development. However, it is worth noting that software that is widely used and contributed to has and enforces the need for good documentation [14]. Being able to change parameters on the fly to determine if and how they change the results is important as your software gains more users since it facilitates exploratory analysis and parameter sweeping. Programs should therefore allow the most commonly changed parameters to be configured from the command line. Users will want to change some values more often than others. Since parameters are software-specific, the appropriate “tunable” ones cannot be detailed here, but a short list includes input and reference files and directories, output files and directories, filtering parameters, random number generation seeds, and alternatives such as compressing results, using a variant algorithm, or verbose output. Check that all input values are in a reasonable range at startup. Few things are as annoying as having a program announce after running for two hours that it isn’t going to save its results because the requested directory doesn’t exist. To make programs even easier to use, choose reasonable defaults when they exist and set no defaults at all when there aren’t any reasonable ones. You can set reasonable default values as long as any command line arguments override those values. Changeable values should never be hard-coded: if users have to edit your software in order to run it, you have done something wrong. Changeable but infrequently changed values should therefore be stored in configuration files. These can be in a standard location, e. g. , . packagerc in the user’s home directory, or provided on the command line as an additional argument. Configuration files are often created during installation to set up such things as server names, network drives, and other defaults for your lab or institution. Software evolves over time, with developers adding or removing features as need dictates. Making official releases stamps a particular set of features with a project-specific identifier so that version can be retrieved for later use. For example, if a paper is published, the software should be released at the same time so that the results can be reproduced. Most software has a version number composed of a decimal number that increments as new versions are released. There are many different ways to construct and interpret this number, but most importantly for us, a particular software version run with the same parameters should give identical results no matter when it’s run. Results include both correct output as well as any errors. Increment your version number every time you release your software to other people. Semantic versioning [25] is one of the most common types of versioning for open-source software. Version numbers take the form of “MAJOR. MINOR[. PATCH], ” e. g. , 0. 2. 6. Changes in the major version number herald significant changes in the software that are not backwards compatible, such as changing or removing features or altering the primary functions of the software. Increasing the minor version represents incremental improvements in the software, like adding new features. Following the minor version number can be an arbitrary number of project-specific identifiers, including patches, builds, and qualifiers. Common qualifiers include alpha, beta, and SNAPSHOT, for applications that are not yet stable or released, and -RC for release candidates prior to an official release. The version of your software should be easily available by supplying --version or -v on the command line. This command should print the software name and version number, and it should also be included in all of the program’s output, particularly debugging traces. If someone needs help, it’s important that they be able to tell whoever’s helping them which version of the software they’re using. While new releases may make a program better in general, they can simultaneously create work for someone who integrated the old version into their own workflow a year or two ago and won’t see any benefits from upgrading. A program’s authors should therefore ensure that old released versions continue to be available. A number of mechanisms exist for controlled release that range from adding an appropriate commit message or tag to version control [20] to official releases alongside code on Bitbucket or GitHub to depositing into a repository like apt, yum, homebrew, CPAN, etc. Choose the method that best suits the number and expertise of users you anticipate. In the spirit of code reuse and interoperability, developers often want to reuse software written by others. With a few lines, a call is made out to another library or program and the results are incorporated into the primary script. Using popular projects reduces the amount of code that needs to be maintained and leverages the work done by the other software. Unfortunately, reusing software (whether software libraries or separate executables) introduces dependencies, which can bring their own special pain. The interface between two software packages can be a source of considerable frustration: all too often, support requests descend into debugging errors produced by the other project due to incompatible libraries, versions, or operating systems [16]. Even introducing libraries in the same programming language can rely on software installed in the environment, and the problem becomes much more difficult when relying on executables or even on web services. Despite these problems, software developers in research should reuse existing software provided a few guidelines are adhered to. First, make sure that you really need the auxiliary program. If you are executing GNU sort instead of figuring out how to sort lists in Python, it may not be worth the pain of integration. Reuse software that offers some measurable improvement to your project. Second, if launching an executable, ensure the appropriate software and version is available. Either allow the user to configure the exact path to the package, distribute the program with the dependent software, or download it during installation using your package manager. If the executable requires internet access, check for that early in execution. Third, ensure that reused software is robust. Relying on erratic third party libraries or software is a recipe for tears. Prefer software that follows good software development practices, is open for support questions, and is available from a stable location or repository using your package manager. Exercise caution, especially when transitioning across languages or using separate executables, as they tend to be especially sensitive to operating systems, environments, and locales. To compile code, deploy applications, and automate other tasks, programmers routinely use build tools like Make, Rake, Maven, Ant, or MS Build. These tools can also be used to manage runtime environments, i. e. , to check that the right versions of required packages are installed and install or upgrade them if they are not. As mentioned in Rule 5, a package manager can mitigate some of the difficulties in software reuse. The same tools can and should be used to manage runtime environments on users’ machines as well. Accordingly, developers should document all dependencies in a machine-readable form. Package managers like apt and yum are available on most Unix-like systems, and application package managers exist for specific languages like Python (pip), Java (Maven/Gradle), and Ruby (RubyGems). These package managers can be used together with the build utility to ensure that dependencies are available at compile/run time. For example, it is common for Python projects to include a file called requirements. txt that lists the names of required libraries, along with version ranges: requests>=2. 0 pygithub>=1. 26, <=1. 27 python-social-auth>=0. 2. 19, <0. 3 This file can be read by the pip package manager, which can check that the required software is available and install it if it is not. Whatever is used, developers should always install dependencies using their dependency description, especially on their personal machines, so that they’re sure it works. Conversely, developers should avoid depending on scripts and tools which are not available as packages. In many cases, a program’s author may not realize that some tool was built locally and doesn’t exist elsewhere. At present, the only sure way to discover such unknown dependencies is to install on a system administered by someone else and see what breaks. As use of virtualization containers becomes more widespread, software installation can also be tested on a virtual machine or container system like Docker. Root (also known as “superuser” or “admin”) is a special account on a computer that has (among other things) the power to modify or delete system files and user accounts. Conversely, files and directories owned by root usually cannot be modifed by normal users. Installing or running a program with root privileges is often convenient, since doing so automatically bypasses all those pesky safety checks that might otherwise get in the user’s way. However, those checks are there for a reason: scientific software packages may not intentionally be malware, but one small bug or over-eager file-matching expression can certainly make them behave as if they were. Outside of very unusual circumstances, packages should not require root privileges to set up or use. Another reason for this rule is that users may want to try out a new package before installing it systemwide on a cluster. Requiring root privileges will frustrate such efforts and thereby reduce uptake of the package. Requiring, as Apache Tomcat does, that software be installed under its own user account—i. e. , that packagename be made a user and all of the package’s software be installed in that pseudo-user’s space—is similarly limiting, and makes side-by-side installation of multiple versions of the package more difficult. Developers should therefore allow packages to be installed in an arbitrary location, e. g. , under a user’s home directory in ~/packagename, or in directories with standard names like bin, lib, and man under a chosen directory. If the first option is chosen, the user may need to modify his or her search path to include the package’s executables and libraries, but this can (more or less) be automated and is much less risky than setting things up as root. Testing the ability to install software has traditionally been regarded as difficult, since it necessarily alters the machine on which the test is conducted. Lightweight virtualization containers like Docker make this much easier as well, or you can simply ask another person to try and build your software before releasing it. It’s easy to write software that reads input from a file called mydata. csv, but it’s also very limiting. If a colleague asks you to process his or her data, you must either overwrite your data file (which is risky) or edit your code to read otherdata. csv (which is also risky, because there’s every likelihood you’ll forget to change the filename back or will change three uses of the filename but not a fourth). Hard-coding file paths in a program also makes the software harder to run in other environments. If your package is installed on a cluster, for example, the user’s data will almost certainly not be in the same directory as the software, and the folder C: \users\yourname\ will probably not even exist. For these reasons, users should be able to set the names and locations of input and output files as command-line parameters. This rule applies to reference datasets as well as the user’s own data: if a user wants to try a new gene identification algorithm using a different set of genes as a training set, he or she should not have to edit the software to do so. A corollary to this rule is to not require users to navigate to a particular directory to do their work, since “where I have to be” is just another hard-coded path. In order to save typing, it is often convenient to allow users to specify an input or output directory, and then require that there be files with particular names in that directory. This practice is an example of “convention over configuration, ” a principle used by software frameworks such as WordPress and Ruby on Rails that often strikes a good balance between adaptability and consistency. Every package should come with a set of tests for users to run after installation. Its purpose is not only to check that the software is working correctly (although that is extremely helpful) but also to ensure that it works at all. This test script can also serve as a working example of how to run the software. In order to be useful, make the tests easy to find and run. Many build systems will also run unit tests if provided them at compile time. For users, or if the build system is not amenable to testing, provide a working script in the project’s root directory named runtests. sh or something equally obvious. This lets new users build their analysis from a working script. For example, with its distribution, the graph-based sequence aligner HISAT2 includes a full set of very small files, and a “Getting Started with HISAT2” section in its manual that leads you through the entire data lifecycle [26]. Equally important is to make the test script’s output easy to interpret. Screens full of correlation coefficients do not qualify: instead, the script’s output should be simple to understand for nonexperts, such as one line per test, with the test’s name and its pass/fail status, followed by a single summary line saying how many tests were run and how many passed or failed. If many or all tests fail because of missing dependencies, that fact should be displayed once, clearly, rather than once per test, so that users have a clear idea of what they need to fix and how much work it’s likely to take. Research has shown that the ease with which people can start making contributions is a strong predictor of whether they will or not [27]. By making it simpler for outsiders to contribute, a test suite of any kind also makes it more likely that they will, and software with collaborators stands a better chance of surviving in the busy field of scientific software. The usage message tells users what the program could do. It is equally important for the program to tell users what it actually did. Accordingly, when the program starts, it should echo all parameters and software versions to standard out or a log file alongside the results to increase the reproducibility of that step. Given a set of parameters and a dataset, a particular version of a program should produce the same results every time it is run to aid testing, debugging, and reproducibility. Even minor changes to code can cause minor changes in output because of floating-point issues, which means that getting exactly the same output for the same input and parameters probably won’t work during development, but it should still be a goal for people who have deployed a specific version. Many applications rely on randomized algorithms to improve performance or runtimes. As a consequence, results can change between runs, even when provided with the same data and parameters. By its nature, this randomness renders strict reproducibility (and, therefore, debugging) more difficult. If even the small test set (#9) produces different results for each run, new users may not be able to tell whether or not the software is working properly. When comparing results between versions or after changing parameters, even small differences can confuse or muddy the comparison. And especially when producing results for publications, grants, or diagnoses, any analysis should be absolutely reproducible. Given the size of biological data, it is unreasonable to suggest that random algorithms be removed. However, most programs use a pseudo-random number generator, which uses a starting seed and an equation to approximate random numbers. Setting the seed to a consistent value can remove randomness between runs. Allow the user to optionally provide the random seed as an input parameter, thus rendering the program deterministic for those cases where it matters. If the seed is set internally (e. g. , using clock time), echo it to the output for reuse later. If setting the seed is not possible, make sure the acceptable tolerance is known and detailed in documentation and in the tests. There has been extended discussion over the past few years of the sustainability of research software, but this question is meaningless in isolation: any piece of software can be sustained if its users are willing to put in enough effort. The real equation is the ratio between the skill and effort available and the ease with which software can be installed, understood, used, maintained, and extended. Following the ten rules we outline here reduces the denominator and thereby enables researchers to build on each other’s work more easily. That said, not every coding effort needs to be engineered to last. Code that is used once to answer a specific question related to a specific dataset doesn’t require comprehensive documentation or flexible configuration, and the only sensible way to test it may well be to run it on the dataset in question. Exploratory analysis is an iterative process that is developed quickly and revised often [4,11]. However, if a script is dusted off and run three or four times for slightly different purposes, is crucial to a publication or a lab, or is being passed on to someone else, it may be time to make your software more robust.

Many researchers have found out the hard way that there’s a world of difference between “works for me on my machine” and “works for other people on theirs. ” Many common challenges can be avoided by following a few simple rules; doing so not only improves reproducibility but can accelerate research.

Abstract Introduction Rule 1: Use version control Rule 2: Document your code and usage Rule 3: Make common operations easy to control Rule 4: Version your releases Rule 5: Reuse software (within reason) Rule 6: Rely on build tools and package managers for installation Rule 7: Do not require root or other special privileges to install or run Rule 8: Eliminate hard-coded paths Rule 9: Include a small test set that can be run to ensure the software is actually working Rule 10: Produce identical results when given identical inputs Conclusion

sequence analysis operating systems computer and information sciences sequence alignment bioinformatics engineering and technology research assessment computer software database and informatics methods software engineering reproducibility software tools software development editorial research and analysis methods

2017

Ten simple rules for making research software more robust

Rift Valley fever virus (RVFV) is a mosquito-borne virus in the family Bunyaviridiae that has spread throughout continental Africa to Madagascar and the Arabian Peninsula. The establishment of RVFV in North America would have serious consequences for human and animal health in addition to a significant economic impact on the livestock industry. Published and unpublished data on RVFV vector competence, vertebrate host competence, and mosquito feeding patterns from the United States were combined to quantitatively implicate mosquito vectors and vertebrate hosts that may be important to RVFV transmission in the United States. A viremia-vector competence relationship based on published mosquito transmission studies was used to calculate a vertebrate host competence index which was then combined with mosquito blood feeding patterns to approximate the vector and vertebrate amplification fraction, defined as the relative contribution of the mosquito or vertebrate host to pathogen transmission. Results implicate several Aedes spp. mosquitoes and vertebrates in the order Artiodactyla as important hosts for RVFV transmission in the U. S. Moreover, this study identifies critical gaps in knowledge which would be necessary to complete a comprehensive analysis identifying the different contributions of mosquitoes and vertebrates to potential RVFV transmission in the U. S. Future research should focus on (1) the dose-dependent relationship between viremic exposure and the subsequent infectiousness of key mosquito species, (2) evaluation of vertebrate host competence for RVFV among North American mammal species, with particular emphasis on the order Artiodactyla, and (3) identification of areas with a high risk for RVFV introduction so data on local vector and host populations can help generate geographically appropriate amplification fraction estimates. Rift Valley fever virus (RVFV) is an emerging infectious disease in Africa and the Middle East. If introduced to North America, RVFV is capable of serious health and socioeconomic consequences potentially incapacitating large numbers of humans, decimating susceptible farm animals, and instigating heavy restrictions on livestock trade [1], [2]. Although transmission of the virus can occur through aerosol inhalation or direct tissue-tissue contact by handling of infected organisms, an enzootic cycle between mosquito vectors and domestic or wild animals has been repeatedly proposed as a main mechanism of transmission [3]. Clinical signs vary by vertebrate species and age, but infected pregnant ruminants generally suffer spontaneous abortions and juvenile ruminants suffer high mortality while occasional spillover into human populations results in a self-limiting, febrile illness that may progress to encephalitis, retinitis, blindness, hemorrhagic fever or death [2]–[5]. In 1931, RVFV was first reported in Kenya. It spread to Egypt in 1977 and was detected on the Arabian Peninsula in 2000 [6], [7]. Since advancing beyond African borders in 2000, total human cases of RVFV include 768 confirmed fatalities, 4,248 confirmed infections and over 75,000 suggested unconfirmed cases [8]–[15]. The emergence of arthropod-borne viruses (arboviruses) through geographic expansion is facilitated when amplification hosts include wild or domestic animals, as demonstrated by West Nile virus (WNV), Japanese encephalitis, and epizootic hemorrhagic disease [2], [16]. Aedes and Culex spp. mosquitoes are proposed to be the main vectors of RVFV, where Aedes spp. act as the reservoir and maintenance vectors that emerge after flood events and feed heavily on livestock [17]. Culex spp. mosquitoes then become involved as amplifying hosts of RVFV leading to epizootics and the eventual spillover to human populations [5], [17]–[19]. However, the understanding of RVFV transmission biology in Africa and the Arabian Peninsula remains underdeveloped. Additionally, unresolved questions surround endemic persistence of the virus, such as transovarial transmission [17]. Should RVFV arrive, diagnosing the disease and controlling the spread of infected vertebrates will take time, and proactive management plans should be created to minimize the time to react and break transmission of the pathogen. Even though RVFV is identified as an emerging infectious disease threat and is classified as a “Category A select agent” by both the Centers for Disease Control and Prevention and the US Department of Agriculture, gaps in data are preventing a proper evaluation of the different roles vectors and vertebrate hosts potentially may play in RVFV transmission in the U. S. beyond qualitative conjecture [1], [20]. To prepare for an arbovirus introduction, it is essential to understand which vectors and vertebrate hosts may be responsible for viral amplification and transmission, as disease control methods vary depending on the target species [21], [22]. For example, mosquito species using small container habitats for larval development are often controlled using larvicides and source reduction of aquatic habitat, whereas mosquito species with synchronous emergence following flooding events are controlled by adulticides or granular larvicides applied prior to flooding [23], [24]. To assess the role of mosquitoes and hosts in the transmission of a virus, it is important to quantify the ability for a mosquito species to transmit a pathogen (vector competence), the infectiousness of vertebrate host species (host competence), and contact rates between mosquitoes and vertebrate hosts. In the WNV system, Kilpatrick et al. [25] combined data on vector competence, abundance, and mosquito feeding patterns to identify the species of mosquitoes responsible for bridge transmission of WNV to humans. Several studies have then implicated important avian hosts disproportionately responsible for WNV amplification based on mosquito host feeding patterns, mosquito vector competence data, and vertebrate host competence data [26], [27]. By applying models utilized in the WNV system, we can implicate potentially important vectors and vertebrate hosts in RVFV transmission should the virus arrive. A number of reviews discuss potential vertebrate hosts, disease vectors, and environments that may support RVFV transmission in the U. S. , through environmental receptivity models [28] and spatial overlap of important host populations [22]. However, to our knowledge, no study has quantitatively evaluated the theoretical importance of different mosquito species and vertebrate hosts to RVFV transmission and amplification in the U. S. [28]. This study utilized published and unpublished vector and host competence data and mosquito feeding patterns to model the theoretical roles of different mosquito and vertebrate species in the amplification and transmission of RVFV in the U. S. Although predictions from this analysis are strictly theoretical, and limited by available data, these results highlight critical gaps in knowledge necessary to properly evaluate the potential transmission activity of RVFV in the U. S. and provide hypotheses that can support proactive arbovirus surveillance and control programs. Mosquito vector competence studies evaluate the ability of mosquitoes to develop an infection and ultimately transmit the pathogen during feeding. Data generated from vector competence studies include viral dissemination and transmission rates. Viral dissemination rates are defined as the percentage of orally exposed mosquitoes with virus detected in their legs seven or more days after RVFV infection. Transmission rates are defined as the percentage of orally exposed mosquitoes (regardless of infection status) that transmitted virus by bite upon refeeding [21]. Selected studies evaluated mosquito species that occur in the U. S. and monitored dissemination and transmission rates after feeding on a RVFV infected animal at the incubation temperature of 26°C. RVFV vector competence studies were located using Web of Science, NCBI' s Pubmed, and the Armed Forces Pest Management Board Literature Retrieval Systems [21], [29]–[35]. Analyzing viral dissemination and transmission data drawn from multiple studies is problematic because these data are dependent on the viremic titer of exposure [33] and the compiled transmission data for this analysis reflects mosquitoes exposed to viremia that ranged from 104. 3 to 1010. 2 plaque-forming units/ml (PFU/ml). To address this issue, a regression analysis of log viremia versus experimental transmission data from 17 mosquito species (Figure S1, A and B) was utilized to estimate the dependence of dissemination and transmission rates on viremic dose. Slopes from these regressions were combined with experimental data from each mosquito species to interpolate what the dissemination and transmission rates would be at the exposure viremia of 107. 5 PFU/ml (equations shown in Table S1). Mosquito species that demonstrated low overall vector competence in experimental transmission studies due to midgut escape barriers or salivary gland barriers (i. e. Anopheles crucians (Wiedemann), Cx. nigripalpus (Theobald) and Ae. infirmatus (Dyar & Knob) ) or had a limited sample size (N<2 mosquitoes) were not used in the regression analyses [29]. The viremia-dissemination equation was equal to 0. 098* (Log10 viremia) −0. 268 and the viremia-transmission rate of a mosquito with a disseminated infection equation was equal to 0. 056* (Log10 viremia) −0. 0155 (Figure S1, A and B; Table S1). Both equations show a positive relationship for dissemination (N = 27; R2 = 0. 28; p = 0. 0049) and transmission (N = 27; R2 = 0. 13; p = 0. 07) as viremic dose increases. For each mosquito species we generated a linear equation and the y-intercept was adjusted for each mosquito species based on the difference between the experimentally observed rate and what the standardized equations described above (Figure S1, A and B) would predict at a specific viremic dose. This adjusted y-intercept and the standardized slopes from Figure S1, A and B (Dissemination m = 0. 098, Transmission m = 0. 056) were utilized to create two unique linear equations for each mosquito species: one to calculate dissemination rate and one to calculate transmission rate with respect to viremic dose for each vector species. By solving for y when x = log10 7. 5 PFU/ml we were able to estimate dissemination and transmission rates at an exposure viremia of 107. 5 PFU/ml for each mosquito species (Table 1, Table S1). When there were multiple data points for a mosquito species the averages of exposure viremia and the observed experimental transmission data were used to calculate the two linear equations for vector competence standardization. Additional data points were estimated that describe transmission rates for Ae. dorsalis (Meigen), Cx. erythrothorax (Dyar), Cx. tarsalis, and Cx. erraticus (Dyar-Knab) mosquitoes that developed a disseminated infection based on the estimated transmission rates of Turell et al. [32]. These data were standardized with the same methodology described above. Vector competence (Cv) was calculated by multiplying the fraction of mosquitoes that develop a disseminated infection after feeding on a viremic host by the transmission rate of mosquitoes with disseminated infection based on estimated values for an exposure viremia of 107. 5 PFU/ml [36]. When mosquitoes feed on an infected vertebrate a fraction of those mosquitoes will become infectious depending on the intensity of the vertebrate host' s viremia and the mosquito' s susceptibility to the virus [37]. Experimental infection studies that exposed vertebrate species to RVFV and monitored post-infection viremias were used to create a host competence index (Ci). The vertebrate reservoir competence index represents the relative number of infectious mosquitoes that may result from feeding on infected vertebrate hosts and is calculated as the product of susceptibility to infection, mean daily infectiousness to each species of mosquito, and duration of infectiousness [38]. Published studies were located using Web of Science, NCBI' s Pubmed, and the Armed Forces Pest Management Board Literature Retrieval Systems. Studies utilizing PFU/ml and Tissue Culture Infectious Dose 50% (TCID50) techniques to quantify viral titers after experimental infection with virulent strains of RVFV (ZH501, T1, T46, AN1830, Kabete, 80612A, AnD100286, AnD100287, Z8548, FRhL2) were the only inclusion criteria for host competence data as no universal conversion between Lethal Dose 50% (LD50) and Mouse Lethal Dose 50% (MLD50) was found. Conversion from TCID50 to PFU/ml was obtained by the equation: PFU/ml = TCID50/ml×0. 69 [39], [40]. To calculate the vertebrate host competence index for RVFV, an equation describing vector competence was calculated utilizing available mosquito transmission experiments performed at 26°C as a linear function of log (host viremia). This viremia-vector competence equation (Figure S1, C) describes the fraction of mosquitoes that would become infected after feeding on a single viremic host indicating the infectiousness of a vertebrate [37], [38]. Because of limited species-specific experimental transmission data, the viremia-vector competence equation is based on the combined experimental transmission data of 17 mosquito species (See Figure S1). Mosquito species that demonstrated low overall vector competence in experimental transmission studies due to midgut escape barriers or salivary gland barriers or had a limited sample size as described above were not used to calculate the viremia-vector competence relationship [29]. The viremia-vector competence equation (vector competence = 0. 062 (Log10 viremia) −0. 276; R2 = 0. 27; N = 27; P = <0. 001) was used to calculate the daily infectiousness of vertebrate hosts by inserting daily vertebrate host viremia titers into the equation. When the equation calculated a vertebrate host' s infectiousness to be negative the vertebrate host' s daily infectiousness was set to zero [37]. These daily values were summed over the host' s viremic period and used as the vertebrate species' competence index (Ci). When multiple experimental studies existed for a particular vertebrate species or taxonomic group a mean Ci was calculated [37], [38], [41]. To determine the theoretical importance of a mosquito to RVFV transmission it is important to consider contact rates between vectors and vertebrate hosts. The amplification fraction estimates the number of infectious mosquitoes resulting from feeding on a particular host and can be utilized as an index to compare the relative role of various vectors in transmission. In the WNV system, the relative number of infectious (transmitting) mosquito vectors resulting from feeding on a vertebrate host was estimated by Kent et al. [42] utilizing the following equation: Fi = Bi2 * Ci where Fi = the relative number of infectious mosquitoes resulting from feeding on each vertebrate species i, where Bi = the proportion of blood meals from species i and Ci = reservoir competence. This equation was modified from Kilpatrick et al. [43] which estimated the fraction of WNV-infectious mosquitoes, Fi, resulting from feeding on each avian species, i, as the product of the relative abundance, the vertebrate reservoir competence index, Ci, and the mosquito forage ratio. Kent et al. [42] found that the relative abundance of each avian species cancelled out when multiplied by the forage ratio, of which the denominator is relative abundance. Fi as defined by Kilpatrick et al. [43] was therefore reduced to the product of Ci and the proportion of blood meals from species i. Because the viremia-vector competence relationship used in this analysis is based on data from multiple mosquito species, Kent et al' s [42] Fi equation was modified to multiply by the mosquito' s vector competence value (Cv) to account for the differences observed in mosquito vector transmission competence across species. The modified equation is referred to as the vector amplification fraction (Fvi) and provides a theoretical means to compare the role of various vector species in the transmission of RVFV. In the Fvi equation, the number of infectious mosquitoes resulting from feeding on a vertebrate host, Fvi, is equal to vertebrate host competence (Ci), multiplied by the vector competence (Cv), multiplied by the fraction of the total blood meals from host i squared (Bi2) [27], [42]. Bi represents the number of blood meals taken from a vertebrate host species divided by the total blood meals taken. Bi is unique to each mosquito species and is used as an indicator of exposure to RVFV and as an indicator of potential RVFV-infectious bites received by a host species, or taxonomic group [44]. Mosquito host feeding data from 39 studies were combined to generate a robust estimate of mosquito feeding patterns at the taxonomic resolution of Class and Order compiled into Table S2. Vertebrate hosts fed on by mosquitoes lacking a competence index (Ci) were assigned the closest taxonomic mean [41]. Only mosquito species with over 40 recorded blood meals to calculate vertebrate host feeding proportions (Bi) were included in this analysis. When vector competence data were missing for a given mosquito species, vector competence values were substituted based on the taxonomic subgenus average (Aedes- Ochlerotatus: 0. 15; Culex- Melanoconion: 0. 04, Culex: 0. 11), genus average (Anopheles: <0. 01; Psorophora: 0. 18, Mansonia: 0. 07) or family average (Culicidae: 0. 15). To include Ae. aegypti in this analysis host-feeding patterns were estimated based on mosquito feeding patterns in Puerto Rico [45]. Fvi is unique to each mosquito vector-vertebrate host pair and assumes initial seroprevalence, susceptibility and competence values are equal among all adult and juvenile vertebrate hosts [27], [46]–[47]. In an attempt to control any effect of the exposure dose of RVFV on the outcome of mosquito transmission competency, the Fvi calculation only utilized mosquito competence values standardized to an exposure dose of 107. 5 PFU/ml as described above. To calculate a mosquito species' vector amplification fraction resulting from feeding on all vertebrate hosts, all Fvi values reflecting a vector-vertebrate pair were summed for each mosquito species (equations shown in Table S3). This overall risk for a mosquito species to contribute to RVFV transmission in the U. S. was calculated based on a weighted percentage relative to the total Fvi displayed by all mosquitoes. To explore the theoretical contribution of vertebrates to RVFV amplification and transmission in the U. S. , Fvi values unique to each vector-vertebrate pair described above were summed across each vertebrate host instead of by mosquito vector. The resulting index expresses the relative number of infectious mosquitoes generated by each vertebrate host. Since species-specific competence data was lacking for all vector-vertebrate host contacts, the role of vertebrate hosts was explored at the taxonomic resolution of class, order, and family. By summing Fvi values with respect to vertebrate host at different taxonomic levels we were able to quantify the theoretical amplification fraction displayed by each vertebrate host taxonomic group. This index was expressed as a weighted average by dividing the summed Fvi values for a vertebrate group by the total Fvi value calculated for the mammalian order (Table S3). Eight experimental studies were identified that fit the inclusion criteria for this analysis [21], [29]–[35]. Data for 26 mosquito species were adjusted utilizing the viremic dose-dependent relationship of dissemination and transmission rates based on 17 species of mosquitoes (Figure S1, A and B). Standardized dissemination and transmission values were multiplied together to calculate vector competence (Table 1 and S1). The most competent transmission vectors of RVFV when exposed to 107. 5 PFU/ml of viremia are estimated to be Coquillettidia perturbans (Walker) (0. 38), Ae. japonicus japonicus (Theobald) (0. 37), Cx. tarsalis (0. 33), and Ae. excrucians (0. 28). Some mosquito species were estimated to be incompetent for RVFV, such as An. crucians (<0. 01), Ae. infirmatus (<0. 01), and Cx. quinquefasciatus (Say) (<0. 01) (Table 1). To estimate vertebrate host competence, published data and unpublished data provided by Dr. John Morrill from RVFV experimental infections (Figure 1) [39], [40], [48]–[65] were inserted into a viremia-vector competence equation that describes the relative number of infectious mosquitoes resulting from feeding on a vertebrate host (Figure S1, C). Exposure viremia dosages ranged from 104. 3–10. 2 PFU/ml at an incubation temperature of 26°C. With this approach, 12 vertebrate species demonstrated reservoir competence by producing sufficient viremia titers to infect mosquitoes after exposure to RVFV, all of which were mammals (Figure 2) [38]–[40]. Vertebrate host species demonstrating competence for viral amplification were the following: sheep (Ovis aries, Class Artiodactyla), domestic cow (Bos taurus, Artiodactyla), domestic goat (Capra aegagrus hircus, Artiodactyla), mouse (Mus musculus, Rodentia); brown rat (Rattus norvegicus, Rodentia), the common marmoset (Callithrix jacchus, Primates); four-striped grass mouse (Rhabdomys pumilio, Rodentia); South African pouched mouse (Saccostomus campestris, Rodentia); Rhesus macaque (Macaca mulatta, Primates); Griselda' s striped grass mouse (Lemniscomys griselda, Rodentia); African buffalo (Syncerus caffer, Artiodactyla); and namaqua rock rat (Aethomys namaquensis, Rodentia). Many species were considered incompetent because they did not develop a sufficient viremia profile to infect mosquito vectors (≤104. 7 PFU/ml), such as the red rock rat (Aethomys chrysophilus, Rodentia), African grass rat (Arvicanthis niloticus, Rodentia), guniea multimammate mouse (Mastomys erythroleucus, Rodentia), natal multimammate mouse (Mastomys natalensis, Rodentia), Mongolian gerbil (Meriones unguiculatus, Rodentia), Atlantic canary (Serinus canaria, Passeriformes), domestic chickens (Gallus gallus, Galliformes) and the Bushveld gerbil (Taera leucogaster, Rodentia). The vertebrate host competence index averages based on taxonomy were the following: Class: Mammalian (0. 17), Aves (0. 00); Order: Primates (0. 25), Artiodactyla (0. 21), Rodentia (0. 05); Family: Bovidae (0. 21), Muridae (0. 05), Cricitidae (0. 05); Genus: Ovis (0. 29), Bos (0. 19), Capra (0. 15), Rattus (0. 04). Among mosquito species evaluated, the vector amplification fraction (ΣFvi) ranged from 0 to 0. 018 (Table 2). The resulting index was expressed as a weighted percentage relative to the total amplification fraction demonstrated by the 40 mosquito species included in this analysis, which ranged from 0% to 11. 7% (Table 2; See Table S3 for calculations). This index estimates the relative probability that a mosquito will feed on an infectious vertebrate host, develop a disseminated infection into the salivary glands, and ultimately transmit RVFV to a vertebrate host during a subsequent blood-feeding event. Mosquito species with the highest amplification fractions were: Ae. japonicus japonicus (Theobald) (11. 4%), Ae. thibaulti (Dyar and Knab) (8. 8%), Ae. canadensis (Theobald) (7. 4%), Culiseta inornata (Williston) (6. 7%), Wyeomyia mitchellii (Theobald) (6. 6%), Ae. sollicitans (Walker) (5. 4%), Cq. perturbans (5. 4%), Ae. sticticus (Meigen) (5. 4%), Ae. aegypti (5. 0%) and Ae. nigromaculis (Ludlow) (4. 4%) (Table 2). Overall four classes (Mammalia, Aves, Amphibia, and Reptilia), eight mammalian orders (Artiodactyla, Carnivora, Chiroptera, Didelphimorpha, Lagomorpha, Perissodactyla, Primates, Rodentia), six families (Bovidae, Cervidae, Cricitidae, Muridae, Sciuridae, Suidae) and seven genera (Bos, Capra, Dama, Homo, Odocoilius, Ovis, Rattus) of vertebrates were included in the model. As indicated by vertebrate competence studies, only mammals are competent hosts and are estimated to contribute 100% of theoretical RVFV amplification in the U. S. The order Artiodactyla is estimated to contribute 64. 3% of all theoretical mammalian RVFV amplification followed by the orders Lagomorpha (16. 8%), Primates (6. 8%), Carnivora (4. 4%), Rodentia (0. 8%), Perissodactyla (0. 4%), Didelphimorpha (0. 1%), and Chiroptera (0. 0%) (Table S3). Because some blood meal data was only specific to the taxonomic resolution of Class there were undefined mammalian hosts that represent 6. 3% of the risk, which means all % risk estimates are potentially underestimated (Table S3). Similarly, within the Artiodactyla order 10. 5% risk is undefined, therefore, the family Cervidae accounts for at least 56% of the theoretical RVFV amplification contributed to Artiodactyla, while Bovidae contributes 34%, and Suidae contributes <1% (Table S3). Rift Valley fever virus has been isolated from at least 40 African mosquito species and currently 19 North American species have been shown to be competent laboratory vectors of RVFV, several of which are known vectors of enzootic viruses of large mammals (e. g. , Cx. tarsalis and western equine encephalitis virus or Ae. taeniorhynchus (Wiedemann) and Venezuelan equine encephalitis). These data suggest that a suite of mosquito vectors could potentially transmit RVFV should the virus reach North America [21]. Overall, results from previous studies have indicated that vector competence for RVFV is variable between mosquito species and among different populations of the same mosquito species. These variations in vector competence within mosquito species could be due to differences in development temperatures, phenotype, or parasite interactions that facilitate or block viral transmission [25], [32], [66]–[68]. Viral infection, dissemination and transmission rates are also dependent on the titer of the viremic exposure [33]. Because mosquito control methods vary for different mosquito species, future RVFV transmission experiments are necessary to better understand variations in vector competence [32], [68]. The vertebrate host competence index value depends on the viral titer circulating in the blood and the duration of this infectious viremia [38]. As the classic RVFV transmission paradigm would hypothesize, which implicates peri-domestic livestock as important amplification hosts, the calculated vertebrate host competence index shows sheep, domestic cow, domestic goat, and African buffalo may potentially contribute to RVFV amplification (Figure 2) [69]. Primates from the new world also demonstrate a high competence suggesting humans may play a role in RVFV transmission. In the 1977 Egyptian outbreak of RVFV, Meegan et al. [6] demonstrated that humans produce a viremia of 10 4. 1–10 8. 6 LD50, but how this relates to vertebrate competence values of new world monkeys remains unclear. The vertebrate competence index indicates rodents can be competent amplification hosts, but their role in viral amplification may be limited as mosquitoes rarely use them as blood meal hosts. The lack of RVFV competence for parakeets, canaries, and pigeons has been described, however our analysis of the class Aves was limited to a study evaluating the Atlantic canary (S. canaria) [52] and an unpublished study by Turell et al. evaluating domestic chickens (G. gallus), both of which have a competence index of zero. It is apparent that RVFV viremia profiles vary between vertebrate hosts (Figure 1 and Figure 2). These variations emphasize the importance of characterizing RVFV viremia profiles of domestic and wild animals present in the U. S. , especially since their immune systems may be more susceptible to a foreign virus. Experimental infection studies evaluating vertebrate species from the U. S. with larger sample sizes will manifest in more accurate competence values and provide a finer set of data to better implicate important vertebrate hosts for RVFV amplification should the pathogen emerge in the U. S. Previous experimental transmission studies conclude that Cx. tarsalis and Ae. j. japonicus are the most competent vectors with the highest risk to transmit RVFV should it arrive in the U. S. ; however, vector competence does not directly imply a significant role in disease transmission [21], [30]–[33], [36], [68]. The vector amplification fraction provides a means to quantitatively compare theoretical risk of various mosquito species based on their potential to contribute to RVFV transmission in the U. S. Vector-host contact rates, as dictated by mosquito feeding patterns, is a key component to consider when evaluating the risk of a mosquito vector, as illustrated by the Cx. tarsalis mosquioto. Cx. tarsalis is one of the most competent vectors of RVFV in the U. S. (Table 1), which feeds mainly on avian hosts (Table S2), and therefore, is predicted to have a low amplification fraction in comparison to other vectors as seen in Table 2 (0. 2% of total risk). Recent transmission experiments by Turell et al. [30] suggest that Ae. j. japonicus mosquitoes are the most competent vector of RVFV in the U. S. (previously Cx. tarsalis). The vector amplification fraction calculated in this study further implicates Ae. j. japonicus as a high risk vector with the potential to contribute to RVFV transmission in the U. S. (11. 4%, Table 2). This invasive mosquito has a high vector competence (0. 37, Table 1), feeds heavily on competent hosts (Artiodactyla 80% and Primates 16%, Table S1), and is found in all U. S. states east of the Mississippi river except for Florida and Louisiana [70]. Should RVFV spread to the U. S. , Ae. j. japonicus populations should be carefully monitored for infection and potentially targeted for mosquito control [30]. Ae. sticticus and Cs. inornata both demonstrate varying degrees of transmission competency, but vector competence for these two species remains undetermined. In the study by Iranpour [68], RVFV was detected in the saliva of Ae. sticticus after experimental infection and Cs. inornata demonstrated both a high infection rate (100%; N = 5) and high dissemination rate after exposure to RVFV viremia between 107. 9 to 109. 4 PFU/ml (60%; N = 3). Considering both these species feed heavily on the order Artiodactyla (Ae. sticticus 94% and Cs. inornata 80%, Table S2) their role in RVFV transmission in the U. S. is uncertain and should be evaluated. Ae. trivittatus is another mammal-biting mosquito estimated to have a moderate role in transmission that occurs in large populations in the Eastern U. S. and is lacking experimental data. Among the top 10 mosquito species theoretically contributing to RVFV transmission in the U. S. , only five species (Ae. j. japonicus, Ae. sollicitans, Ae. canadensis, Cq. perturbans and Ae. aegypti) have data comprehensive enough for this analysis. This underscores the lack in data necessary to estimate the theoretical role of different mosquito vectors in RVFV transmission in the U. S. Of those ranking as high-risk for contributing to RVFV enzootic transmission, some are limited in geographic range within the U. S. (e. g. Wy. mitchellii) underscoring the importance for including spatial and temporal mosquito abundance data while evaluating local regions for RVFV transmission potential. These results indicate a gap in experimental transmission data and requisite further vector competence evaluations to properly evaluate the potential risk of mosquitoes contributing to RVFV transmission in the U. S. Future studies should pay particular emphasis on assessing and re-evaluating the regional transmission competence and population dynamics of Ae. j. japonicus, Cs. inornata, Ae. sollicitans, Ae. sticticus (only 13 individuals have been evaluated [70]), Ae. nigromaculis (all data from one study in 1988 [31]), and Ae. trivittatus because of their estimated risk and abundance in the Eastern U. S. Artiodactyla, Lagomorpha, Primates, and Carnivora are estimated to be theoretically involved in RVFV amplification in the U. S. , while the Mammalian orders Perissodactyla, Didelphimorpha and Chiroptera are not (Table S3). The order Chiroptera may deserve further investigation as a potential reservoir host as RVFV has been isolated from several bat genera [71] and even though antibodies against RVFV have been detected in horses, the family Equidae has demonstrated low viremic titers [72], [73]. Our results suggest that Artiodactyla contributes 64. 3% of the theoretical risk for RVFV transmission in the U. S. , which supports the currently held paradigm that Artiodactyla are the most important vertebrate host for RVFV amplification and transmission. Research and control efforts should place a particular emphasis on the families Cervidae and Bovidae as they account for at least 56% and 34% of the total risk contributed by the order Artiodactyla, respectively (Table S3). Based on the 2012 Census of Agriculture (USDA National Agriculture Statistics Service) there are about 90 million cattle, 5 million sheep, 3 million goats, and 300,000 captive cervids. There are an estimated 25 million white-tailed deer (Odocoileus virginianus) in the U. S. [74]. Throughout the U. S. captive and wild ruminants are widely available and heavily utilized by mosquitoes (Table S2) emphasizing their potential role in RVFV transmission. It is important to note that the role of the order Lagomorpha (17%) may be inflated by the vector amplification fraction because their estimated vertebrate competence was based on a mammalian average (0. 17). No studies provide evidence supporting that Lagomorphs are capable of producing an infectious viremia, but little research has evaluated their role in RVFV ecology [52]. Similarly, vertebrate competence of the order Carnivora is lacking. Studies demonstrate susceptibility in cats, dogs, ferrets and serological studies demonstrate antibodies against RVFV in lions (Panthera leo) and the polecat (Ictonyx striatus) [72], [75]–[77]. Experimental evaluation within the Order Carnivora should focus on the competence of dogs, cats, and raccoons because mosquito host-feeding is mainly associated with these species (Table S2). Arbovirus amplification in domestic and peridomestic animals and eventual spillover to humans is a well-documented phenomenon. However the permanent establishment of dengue and chikungunya viruses in urban, tropical environments demonstrates the ability for arboviruses to subsist through human reservoirs [2], especially important given the recent emergence of chikungunya in the Caribbean in 2013 [76]. The vertebrate amplification fraction estimates Primates will contribute about 7% of the theoretical RVFV amplification in the U. S. (Table S3). This estimate is based on the assumption that the human viremia profile is comparable to Rhesus macaques and common marmosets. Viremia data from new-world monkeys as a surrogate for human viremia may overstate the role of humans in RVFV transmission. In the 1977 Egyptian outbreak of RVFV, Meegan et al. [6] demonstrated that indeed humans produce a viremia of 10 4. 1–10 8. 6 LD50, however socio-economic factors in the U. S. may limit mosquito-human contact rates, and dampen any role in amplification of RVFV. As such, the role of humans as vertebrate hosts for RVFV amplification remains unknown. Hypotheses implicating rodents as important hosts for RVFV amplification started when high death rates of Arvicanthis abyssinicus and Rattus rattus coincided with sheep deaths caused by RVFV in 1932 [72]. Experimental studies demonstrate rodents can be competent amplification hosts for RVFV (Figure 1 & 2) depending on the viremic dose, age, and species [72]. However, results from the vertebrate amplification fraction suggest members of the order Rodentia are at low risk for contributing to RVFV transmission because of infrequent contact with mosquitoes (Table S2). Given the gaps in data preventing a complete analysis of the amplification fraction potentially produced by all mosquito and vertebrate hosts, we made several assumptions that limit the accuracy of these results. This analysis does not account for spatial or temporal variation in mosquito abundance or competence, both of which are known to be spatially heterogeneous and influence pathogen transmission dynamics [32], [77]. Many of the mosquito species and vertebrate hosts included in the analysis have no competence data and for these species we assigned taxonomic averages. It is important to note that taxonomic averages are not always appropriate and extrapolations based on taxonomic averages for both vectors and vertebrate hosts can lead to spurious results (e. g. disparate RVFV vector competence exists for several Culex spp.) [41]. By combining data on 39 studies reporting mosquito host-feeding patterns in different regions and landscapes across the U. S, we aim to incorporate a robust measure of vertebrate host utilization. However, the mosquito host-feeding patterns for several species are based on a single study, and given the importance of host availability [78], a single study might not be broadly representative of host feeding patterns. Despite these limitations, the results from this study highlight potentially important mosquito vectors and vertebrate hosts of RVFV that should be monitored in the event RVFV emerges in the U. S. Additionally, this study identifies knowledge gaps that can be filled by future experimental work on both vectors and vertebrate species. World-wide zoonotic disease emergence is an increasing phenomenon due to environmental changes, ecological disturbances, and globalization [79]. The U. S. has already been affected by the emergence of WNV, recently identified a new zoonotic disease (Heartland virus) [80], [81], and is threatened by the spread of chikungunya virus to the Caribbean [76]. During the initial epidemics of WNV in the U. S. in 2002 and 2003, many mosquito control programs did not have a strong focus on Culex spp. mosquitoes. As knowledge of the WNV transmission system increased, vector control has improved by targeting Culex species to reduce human exposure events. The delay of Culex spp. vector control might have allowed more human WNV disease and may have contributed to the rapid spread of the virus across the U. S. highlighting the importance of a priori response strategies for potential viral threats. RVFV is of particular concern in the U. S. because it causes disease in humans and economically important animals alike. Even more, its emergence throughout Africa and the Arabian Peninsula make it a conceivable threat for future geographic expansion. We combined published data to provide an estimate of each vector and vertebrate taxon' s contribution to RVFV amplification in the U. S. However, major gaps in knowledge exist preventing a comprehensive evaluation of potentially important vectors and vertebrate hosts to RVFV transmission in the U. S. Results, combined with information on abundance of vectors and vertebrate hosts, can provide guidance for proactive management programs and aid parameterization for further modeling efforts evaluating environmental receptivity of RVFV in the U. S. [22], [28]. Additionally, the framework of this analysis can also be applied to regions in Africa and the Arabian Peninsula with endemic RVFV transmission to help identify important vectors and vertebrate hosts for vector control and vaccination programs. Future research efforts should focus on: 1) further evaluating the dose-dependent nature of RVFV vector competence in geographically widespread mosquitoes quantified as high risk: Ae. j. japonicus, Ae. canadensis, Cs. inornata, Ae. sollicitans, Cq. perturbans, Ae. sticticus, Ae. nigromaculis, Ae. cantator and Ae. trivitattus 2) characterizing local vector competence in high risk areas for RVFV introduction, and 3) evaluating the RVFV viremia profiles of vertebrates in the U. S. with particular emphasis on the orders Artiodactyla (Cervidae, Bovidae, Suidae), Lagomorpha, and Carnivora (domestic dog, domestic cat, raccoon), respectively.

In anticipation of continued pathogen emergence in the U. S. due to globalization climate change, and other factors, the development of proactive management plans and interventions to predict and then intervene is going to be more efficient and effective than retrospective plans developed after pathogen emergence. Effective management of mosquito-borne pathogens like Rift Valley fever virus (RVFV) requires an understanding of the roles that different mosquito species and vertebrate hosts play in transmission. This study combines data on mosquito transmission efficiency, mosquito feeding patterns, and vertebrate infectiousness to quantitatively evaluate the relative importance of different mosquito species and vertebrate hosts to the amplification of RVFV in the U. S. We identify several species of floodwater Aedes spp. mosquitoes that would be the most likely vectors for RVFV, and hoofed ungulates (deer, cows, sheep) would be the most important amplifying vertebrate hosts. Although these data provide public and animal health agencies a priori knowledge on the primary mosquitoes that should be targeted for vector control and the highest priority animals to receive vaccines, this analysis reveals many gaps in knowledge reducing our ability to predict and then manage a potential invasion of RVFV.

Abstract Introduction Methods Results Discussion

invertebrates ecology and environmental sciences medicine and health sciences animals population modeling veterinary science infectious diseases veterinary diseases zoonoses epidemiology infectious disease modeling population ecology insects disease vectors arthropoda mosquitoes ecology veterinary virology biology and life sciences computational biology organisms

2014

Predicting the Mosquito Species and Vertebrate Species Involved in the Theoretical Transmission of Rift Valley Fever Virus in the United States

There is growing evidence that gene expression profiling of peripheral blood cells is a valuable tool for assessing gene signatures related to exposure, drug-response, or disease. However, the true promise of this approach can not be estimated until the scientific community has robust baseline data describing variation in gene expression patterns in normal individuals. Using a large representative sample set of postmenopausal women (N = 286) in the Norwegian Women and Cancer (NOWAC) postgenome study, we investigated variability of whole blood gene expression in the general population. In particular, we examined changes in blood gene expression caused by technical variability, normal inter-individual differences, and exposure variables at proportions and levels relevant to real-life situations. We observe that the overall changes in gene expression are subtle, implying the need for careful analytic approaches of the data. In particular, technical variability may not be ignored and subsequent adjustments must be considered in any analysis. Many new candidate genes were identified that are differentially expressed according to inter-individual (i. e. fasting, BMI) and exposure (i. e. smoking) factors, thus establishing that these effects are mirrored in blood. By focusing on the biological implications instead of directly comparing gene lists from several related studies in the literature, our analytic approach was able to identify significant similarities and effects consistent across these reports. This establishes the feasibility of blood gene expression profiling, if they are predicated upon careful experimental design and analysis in order to minimize confounding signals, artifacts of sample preparation and processing, and inter-individual differences. There is growing evidence that transcriptome analysis of peripheral blood cells is a valuable tool for determining signatures related to disease [1]–[5] and drug-response [6]. Differences in blood gene expression may also reflect the effects of a particular exposure, such as smoking [7], metal fumes [8], or ionizing radiation [9]. In our previous research, we studied gene expression profiles from whole blood related to hormone therapy (HT) use in postmenopausal women [10] and identified specific challenges raised by inter-individual variability when isolating signals associated with defined exposure levels. Although blood gene expression profiling promises molecular-level insight into disease mechanisms, there remains a lack of baseline data describing the nature and extent of variability in blood gene expression in the general population. Characterizations of this variation and the underlying factors that most influence gene expression amongst healthy individuals will play an important role in the feasibility, design and analysis of future blood-based studies investigating biomarkers for exposure, disease progression, diagnosis or prognosis [11]. Several studies [12]–[18] have reported that technical variables such as collection, transportation, storage of blood samples, RNA isolation method and choice of microarray platform, in addition to biological effects, can influence gene expression profiles. These technical factors associated with the processing and preparation of human blood and subsequent microarray hybridization represent significant challenges in the analysis of variability. Furthermore, a few previous studies have used microarrays to analyze blood from healthy volunteers and found that inter-individual sample variation was associated with sex [18], age [13], [18], the time of day the sample was taken [18], [19], and the proportion of the different cell populations comprising the blood sample [13], [18], [20]. However to date, all such studies have focused on gene expression profiles generated from a small set of samples not representative of the general population using different blood cell subtypes. For several reasons including the small sample sizes, these studies have been restricted to the analysis of a small number of variables simultaneously, thus ignoring possible interaction and confounder effects. Finally, an understanding of these causes of variability would represent a significant step forward in the identification and evaluation of the disease and disease risk biomarkers. Most if not all genes are involved in molecular pathways that provide mechanistic insight in response to exposure or disease development. Pathway depictions are usually simplified, ignoring interactions with other pathways, and we often have incomplete knowledge about the specific interplay of the many elements in almost any particular system. Using a large representative sample set of postmenopausal women in the Norwegian Women and Cancer (NOWAC) postgenome study [21], [22] processed via a standardized blood collection procedure and via an experimentally validated microarray platform [23], we investigate here the baseline variability of whole blood gene expression profiles. This represents the first comprehensive cross-sectional analysis of blood gene expression changes related to multiple inter-individual and exposure variables, and opens the new research discipline of systems epidemiology [24]. In this setting, we investigated blood gene expression changes due to technical variability, normal inter-individuality, and exposure variables at proportions and levels relevant to real life situations, and establish that these effects are mirrored in the blood transcriptome. In addition to technical variability, substantial differences in gene expression profiles were identified between individuals with respect to exposure. Overall, the functional enrichment of significant single genes and gene set enrichment analyses show that high-throughput gene expression studies implicate similar (although not identical) underlying biology across several studies. Whereas age did not induce a large effect in blood gene expression for our cohort of postmenopausal women aged from 48 to 62 years, pathways and gene sets affected by smoking and, to a lesser extent both BMI and fasting, are numerous and interconnected. Some expression profiles associated with these variables may also be associated with other factors (e. g. , lower levels of exercise, age). A host of new candidate genes for regulation by inter-individual (fasting, BMI) and exposure (smoking) factors were identified which could be used as a basis for hypothesis development. Several processes associated with smoking were involved in cardiovascular regulation by G-coupled receptors (i. e. purinergic, adrenergic beta-1, urotensin II or thromboxan A2 receptors) or protein activity (i. e. thrombospondin type-1, fibronectin type-3). Consistent with previous observations that smoking reduces olfactory sensitivity in a dose- and time-dependent manner [39], [40], we find that smoking significantly impairs blood gene expression of olfactory receptors. We also observed that smokers have deregulated gene expressions of several P450 cytochromes which catalyse mono-oxygenase activity that can both toxify and detoxify carcinogenic compounds. As established in normal lung [41] and rats [42], smokers tend to have a small increase in NAD (P) H: (quinone-acceptor) oxidoreductase compared to non-smokers. Two previous studies [7], [31] have examined the effects of cigarette smoking on leukocyte gene expression in circulation and both of the associated signatures had the most significant enrichment scores over all gene sets considered here. Inflammatory responses previously associated with smoking [7] were up-regulated in the blood expression of smokers in our dataset. Lending support that smoking has immune and inflammatory effects, specific blood cell gene signatures [13], [18] (i. e. increased monocytes and decreased red blood cell and natural killer cell signalling) were differentially expressed according to smoking status. This is consistent with previous observations showing that the total numbers of peripheral leukocytes differ by smoking status [43], [44]. Core genes up-regulated in non-smokers from the enriched hormone-related gene sets [10], [33] were predicted to be involved in neuroactive ligand-receptor interactions like prostaglandin receptors. Elevated prostaglandin E2 synthesis has been previously reported in smokers in comparison with non-smokers [45], [46]. The predicted gene network also reflects the effect of smoking on hormone levels with increased secretion of prolactin and glucagon [47]. Two pathways related to exercise [32] were also found up-regulated in non-smokers, which may simply be due to an underlying prevalence of active exercisers in non-smokers [48]. In our study, we found BMI class associated with blood gene expression changes involved in several immune processes including diabetes type I. It has been reported that several immune functions are dysregulated in obesity [49], [50] and both genetic and environmental factors such as obesity have been implicated as triggers in the pathogenesis of diabetes. The role of autoimmunity in the origins of type I diabetes is well-known, including a role in latent autoimmune diabetes in adults [51] and several observations suggest that autoimmunity may be part of type II diabetes [52]–[55]. Finally, two pathways related to exercise [32] were also up-regulated in women with normal BMI which may be due to a higher prevalence of physical exercise than in overweight/obese women. Of all the variables considered, fasting was associated with the largest number of genes, but few genes were identified as core genes possibly due to the limited number of fasting women (N = 28) at the time of blood sampling. Selection of core genes aims to select a subset of true positives which work together (possibly in similar pathways) towards significance of the set. The significant core genes associated with fasting were generally involved in gene expression regulation and chromatin modification [56]–[58]. Much of our understanding of the effects of nutrition on chromatin structure has been gleaned from model organisms, especially S. cerevisiae, C. elegans, Drosophila, and mice [59]. In humans, two previous studies were unable to characterize acute effects of food intake in blood gene expression profiles [13], [18]. One putative 784-gene signature exists [34], however only 49 genes associated with fasting overlap with this signature. This may simply be due to chance. Due to a significant interaction between HT and MED within our profiles, further analyses with a larger sample size are needed in order to investigate the different categories of medications, HT regimens and hormone levels, as well as their interactions in blood. Differences between the genes identified and the interpretation of results in the various studies discussed here are likely to have resulted from technical differences in the array platforms used, the subset of blood cells analyzed, and the chosen analytical procedures. Several studies [12]–[18] examined how gene expression profiles of blood samples are affected by technical variables. Specific blood sample collection methods result in the isolation of different blood cell subpopulations. White blood cells have been defined as the most transcriptionally active of all cell types in blood and may give the most sensitive gene expression profiles in response to defined factors [60]. In large epidemiological studies, RNA stabilization is compulsory and PAXgene tubes have been found satisfactory to stabilize and enable RNA extraction from whole blood cells [61]. While high proportions of globin RNA could reduce sensitivity with respect to certain microarray platforms [60], [62], [63], we previously investigated two globin reduction protocols and determined that they were not beneficial when Applied Biosystems (AB) microarrays are used [23]. We found that RNA extraction and one variable related to RNA degradation (i. e. time between blood collection and freezing) had a significant global effect on blood gene expression profiles. In addition to normalization preprocessing, our results suggest that technical variability should not be ignored and possible adjustment for technical sources of variability should be considered in any analysis. Techniques such as surrogate variable analysis [64] may adjust for hidden sources of heterogeneity and large-scale dependence in gene expression studies [65]. As an example in our study, 25 significant surrogate variables were highly correlated to the strongest identified technical sources of noise, array lot number (canonical correlation r2 = 0. 95), time between blood collection and freezing (canonical correlation r2 = 0. 62) and RNA extraction (canonical correlation r2 = 0. 43). After adjustment for technical variability, our analysis demonstrates the ability to find significant similarities between studies by focusing on the biological implications of the gene sets from each individual study, rather than the specific single genes that met the criteria for significant differential expression in each individual study. They lend support to the idea that blood gene expression studies can indeed detect exposure-specific differences and that failure to consider this type of biological variation can result in the misidentification of genes when investigating predictive, diagnostic or prognostic signatures in blood. In conclusion, this study extends the limited baseline information currently available that describes normal patterns of variation in blood gene expression. The data generated have been made freely available and should represent a useful resource for the design of future studies including power calculations. Our results confirm the feasibility of identifying signatures of inter-individual factors (e. g. fasting, BMI) and exposure factors (e. g. smoking) in blood-based gene expression profiles, and reinforces the need for proper study design, sample preparation, and technical analysis. We have received approval from the Regional Committee for Medical Research Ethics for the collection and storing of questionnaire information and blood samples. The informed consent formula explicitly mentions that the blood samples can be used for gene expression analyses as well as large-scale genotyping. All data are stored and handled according to the permission given by the Norwegian Data Inspectorate. The Directorate of Health and Social affairs (SHD) has given us an exemption from the confidentiality of information in national registers. Before use of the biological material, a request has been sent to the regional ethical committee for Northern-Norway. Use of biological material requires permission according to laws pertaining to biotechnology and gene technology, both of which are administered by the SHD. The women are part of the Norwegian Women and Cancer (NOWAC) study (http: //uit. no/kk/NOWAC/) consisting of 172471 women who were 30 to 70 years of age at recruitment from 1991 to 2006 [22]. The NOWAC postgenome cohort study [21] consists of approximately 50,000 women born between 1943 and 1957, randomly drawn in groups of 500 from the NOWAC registers, who gave blood samples between 2003 and 2006 and filled in a two-page questionnaire. The two-page questionnaire included questions regarding menopausal status, weight, height; past week exposure to smoking, HT, oral contraceptives, other MED, omega-3 fatty acid, soy or other dietary supplements; and details concerning blood specimen collection (date, hour, posture). Women included in the present study received a blood collection kit and an accompanying two-page questionnaire by mail in April 2005. Among the group of 500 women, 444 (89%) returned both citrate and PAXgene blood RNA (PreAnalytiX GmbH, Hembrechtikon, Switzerland) tubes; 3. 3% declined to participate, 0. 7% had died or migrated and 7% did not respond. Samples were included in the study according to the following inclusion criteria: the donor was postmenopausal (99 donors excluded), blood was successfully collected in one PAXgene tube and in two plasma collection tubes (8 donors excluded), and the samples were frozen within 3 days from blood collection (9 donors excluded). Based on these criteria, 328 PAXgene blood samples were included for RNA extraction. PAXgene blood RNA tubes were thawed at room temperature for 4 h. 500 µL of blood was removed and stored on −70°C for future use. Total RNA was isolated using the PAXgene Blood RNA Isolation Kit, according to the manufacturer' s manual. RNA quantity and purity was assessed using the NanoDrop ND-1000 spectrophotometer (ThermoFisher Scientific, Wilmington, Delaware, USA). The absorbance ratio of 260 nm and 280 nm (A260/A280) was between 1. 93 and 2. 1 for all samples included for further analysis. The Experion automated electrophoresis system (BioRad, Hercules, CA, USA) and the RNA StdSens Analysis Kit was used to evaluate RNA integrity of a randomized 32% of the samples, according to the instruction manual. The electropherograms were inspected for clear ribosomal peaks. We were not able to analyze any numerical criteria corresponding to electrophoresis patterns, because this information was not available. Thirty nine samples were excluded from further analysis due to insufficient RNA purity, yield or integrity. RNA samples were kept at −70°C until further use. After exclusion based on study design and RNA quality and quantity criteria, samples were analyzed using the Applied Biosystems (AB) expression array system (Foster City, Lousiana, USA). 500 ng total RNA was used for amplification by the NanoAmp RT-IVT labeling kit from AB for one round of amplification, in accordance with the manufacturer' s manual. Briefly, the 1st strand of cDNA was synthesized by reverse transcription using the T7-oligo (dT) primer, followed by 2nd strand synthesis. The double-stranded cDNA was purified, and used as template for in vitro transcription (IVT). During IVT, digoxigenin (DIG) -labeled UTP was incorporated into the cRNA. The quantity and purity of the cRNA was measured on the NanoDrop ND-1000, and the cRNA was stored on −70°C until further use. 10 µg of DIG-labeled cRNA was fragmented and hybridized to AB Human Genome Survey Microarray V2. 0, in accordance with the Chemiluminescence Detection Kit Protocol. The AB Human Genome Survey Microarray V2. 0 contains 277 control probes and 32,878 probes for the interrogation of 29,098 genes. AB Expression System software was used to extract signal intensities, signal to noise ratios (S/N) and flagging. A total of 304 arrays including 15 technical replicates were analyzed. Data analysis was performed using R (http: //cran. r-project. org), an open-source-interpreted computer language for statistical computation and graphics, and tools from the Bioconductor project (http: //www. bioconductor. org), adapted to our needs. Using R, we set the expression intensity to “missing” for genes with flagging value >8191 (threshold recommended by the microarray manufacturer). For a set of technical replicate arrays from the same subject, we excluded the array with the least number of probes that had a S/N exceeding 3. Furthermore, arrays (N = 3) where less than 40% of the probes had a S/N≥3 were also removed from the analysis. Individual probes were not considered, if the S/N exceeded 3 in less than 50% of the samples. After sample and probe filtration, we proceeded with a log2 transformation, quantile normalization and imputation of missing values using 10-nearest neighbourhood method [66]. A total of 286 arrays and 16185 probes are analyzed. Microarray data have been deposited at Gene Expression Omnibus (GEO; http: //www. ncbi. nlm. nih. gov/geo) accession number GSE15289. The global ANCOVA [25] was carried out by comparison of linear models via the extra sum of squares principle to test for the univariate and multivariate association between global expression values and technical variables. All significant technical variables with a permuted p-value <0. 001 identified in the ANCOVA multivariate analysis were included in the gene-wise linear model selection as random (array lot number, RNA extraction date) and fixed (time between blood collection and freezing) variables. Forward-backward variable selection was used to select gene-wise model based on BIC. Linear mixed models were used to test the association of each gene with the significant technical and all biological variables. The z-score from the global test [26] was used to select core probes that most strongly explain the difference between groups setting a FDR [67] threshold which maximizes the discovery of true positives (weight = 2) versus false positives (weight = 1) associated with each variable. Gene set enrichment analysis was conducted using the global test [26], which offers the opportunity to compare two or more groups while taking into account the association between probe sets as well as their individual effects. When testing several gene sets curated from the literature, we adjusted for multiple testing using FDR [67]. Functional clustering and gene networks prediction were performed with the Database for Annotation, Visualization, and Integrated Discovery (DAVID) at http: //david. abcc. ncifcrf. gov/ [27], and the Human Experimental/Functional Mapper (HEFalMp) [28] at http: //function. princeton. edu/hefalmp, respectively.

As a major defence and transport system, blood cells are capable of adjusting gene expression in response to various clinical, biochemical, and pathological conditions. Here, we expand our understanding about the nature and extent of variation in gene expression from blood among healthy individuals. Using a large representative sample of postmenopausal women (N = 286) in the Norwegian Women and Cancer (NOWAC) postgenome study, we investigated blood gene expression changes due to normal inter-individuality (age, body mass index, fasting status), and exposure variables (smoking, hormone therapy, and medication use) at proportions and levels found in real life situations. Host genes were found to vary by inter-individual (i. e. fasting, BMI) and exposure (i. e. smoking) factors, and these gene lists may be used as a basis for further hypothesis development. Our study also establishes the feasibility of blood gene expression profiling for disease prediction, diagnosis, or prognosis, but underscores the necessity of care in study design and analysis to account for inter-individual differences and confounding signals.

Abstract Introduction Discussion Methods

genetics and genomics/genomics public health and epidemiology/epidemiology genetics and genomics/physiogenomics genetics and genomics/gene expression

2010

Deciphering Normal Blood Gene Expression Variation—The NOWAC Postgenome Study

Plants produce large amounts of secondary metabolites in their shoots and roots and store them in specialized secretory structures. Although secondary metabolites and their secretory structures are commonly assumed to have a defensive function, evidence that they benefit plant fitness under herbivore attack is scarce, especially below ground. Here, we tested whether latex secondary metabolites produced by the common dandelion (Taraxacum officinale agg.) decrease the performance of its major native insect root herbivore, the larvae of the common cockchafer (Melolontha melolontha), and benefit plant vegetative and reproductive fitness under M. melolontha attack. Across 17 T. officinale genotypes screened by gas and liquid chromatography, latex concentrations of the sesquiterpene lactone taraxinic acid β-D-glucopyranosyl ester (TA-G) were negatively associated with M. melolontha larval growth. Adding purified TA-G to artificial diet at ecologically relevant concentrations reduced larval feeding. Silencing the germacrene A synthase ToGAS1, an enzyme that was identified to catalyze the first committed step of TA-G biosynthesis, resulted in a 90% reduction of TA-G levels and a pronounced increase in M. melolontha feeding. Transgenic, TA-G-deficient lines were preferred by M. melolontha and suffered three times more root biomass reduction than control lines. In a common garden experiment involving over 2,000 T. officinale individuals belonging to 17 different genotypes, high TA-G concentrations were associated with the maintenance of high vegetative and reproductive fitness under M. melolontha attack. Taken together, our study demonstrates that a latex secondary metabolite benefits plants under herbivore attack, a result that provides a mechanistic framework for root herbivore driven natural selection and evolution of plant defenses below ground. Plants produce over 200,000 different metabolites that are not directly needed for their growth and development [1]. Many of these so-called secondary metabolites have a negative impact on insect herbivores [2–6], leading to the hypothesis that they evolved as defenses against the latter [7]. Indeed, recent studies demonstrated that leaf secondary metabolites reduce herbivore damage and thereby counteract the negative impact of herbivores on plant growth, that herbivore abundance covaries with secondary metabolites across different environments, that the exclusion of herbivores leads to rapid changes in genotype frequencies and associated metabolites, and that genes encoding for defensive metabolites can be under differential selection [8–12]. Together, these studies provide strong evidence for the hypothesis that above ground herbivores drive the evolution of leaf secondary metabolites. In contrast to the leaves, less is known about the role of secondary metabolites in root–herbivore interactions. Roots are often attacked by below ground herbivores, and root herbivore infestation can strongly reduce plant growth and reproduction [13–15]. Furthermore, roots produce diverse and abundant blends of secondary metabolites [16,17], many of which can affect root herbivore behavior and reduce their performance [18]. Furthermore, root secondary metabolites can determine host species ranges in below ground feeding insects [19]. However, if root secondary metabolites enable plants to maintain growth (i. e. , vegetative fitness) and reproduction (i. e. , reproductive fitness) under root herbivore attack remains unclear. Common milkweed (Asclepias syriaca) families with high and low root cardenolides, for instance, did not differ in their above-ground biomass accumulation when attacked by Tetraopes tetraophthalmus below-ground [20]. Maize lines with high root benzoxazinoid concentrations on the other hand suffered less root damage by Diabrotica virgifera virgifera and had higher yields than lines with low benzoxazinoid concentrations [21]. However, follow-up experiments conducted under more controlled conditions failed to confirm this pattern [5,22]. The lack of knowledge regarding fitness benefits of root secondary metabolites makes it difficult to understand their role in the evolution of plant–herbivore interactions. In both leaves and roots, secondary metabolites often accumulate in specialized structures including laticifers [23,24], which are among the most common secretory structures of flowering plants [25–27]. Laticifers are elongated individual or interconnected cells whose cytoplasm is called latex [28,29]. Laticifers are often under pressure and release large quantities of latex upon wounding, which can deter or even kill insect herbivores [28,30]. Surprisingly, however, direct evidence that laticifers are defensive, i. e. , that they are positively associated plant vegetative or reproductive fitness in the presence but not in the absence of herbivory, is virtually absent [28,31,32]. A study by Agrawal [31] showed that latex exudation is under positive selection in common milkweed under ambient insect pressure. However, whether this pattern is herbivore dependent remains to be elucidated. One of Europe’s most prevalent native latex-producing plants is the common dandelion (T. officinale agg.) (Flora Helvetica, 5th edition). T. officinale is a species complex consisting of sexual, outcrossing diploids that are native to central and southern Europe and a multitude of apomictic, clonal triploids that are spreading across the globe [33–35]. Similar to many other perennials in temperate ecosystems, the plant relies on its roots for resprouting and flowering in spring. As a perennial plant, both vegetative and reproductive performance contribute to the fitness of the plant. T. officinale produces latex in all major organs, with the highest amounts exuding from wounded tap roots [36]. The latex is dominated by three classes of secondary metabolites: phenolic inositol esters (PIEs), triterpene acetates (TritAcs) and the sesquiterpene lactone taraxinic acid β-D-glucopyranosyl ester (TA-G) [36]. Each compound class accounts for 5%–7% of latex fresh mass [36]. Sesquiterpene lactones and TritAcs can have deterrent and toxic effects against a wide range of organisms [37–40]. In its native range, T. officinale is frequently attacked by the larva of the common cockchafer (also called May bug), M. melolontha (Coleoptera: Scarabaeidae). M. melolontha is among Europe’s largest and most prevalent native root-feeding insects and periodically causes devastating damage to crops and pastures [41–43]. Although the larvae are highly polyphagous, they preferentially feed on T. officinale [44,45]. In this study, we explored the putative defensive function of T. officinale latex secondary metabolites against M. melolontha larvae. First, we investigated which latex secondary metabolites are likely to be involved in root herbivore defense using a correlative approach. Second, we decreased the production of the major candidate compound TA-G by identifying the gene encoding the first committed biosynthetic step and silencing it by RNA interference (RNAi), which allowed testing the effect of TA-G deficiency on plant and insect performance. Third, we purified TA-G to investigate its impact on M. melolontha in vitro. Fourth, we performed a common garden experiment with different T. officinale genotypes to determine whether TA-G reduces the negative impact of M. melolontha on plant vegetative and reproductive performance in the field. Through the above approaches, we demonstrate that TA-G protects the roots and thereby benefits plant fitness in the presence of root herbivores. Three classes of secondary metabolites dominate the latex of T. officinale: PIEs (Fig 1A, left panel), the sesquiterpene lactone TA-G (Fig 1A, left panel), and TritAcs (Fig 1A, right panel) [36]. We measured the concentrations of the major latex secondary metabolites in 40 triploid T. officinale genotypes collected across central and northern Europe and selected 17 genotypes that displayed maximal variation in latex traits, but minimal variation in growth (S1 Text, S1 Table) to correlate latex secondary metabolites with herbivore performance. M. melolontha larval mass gain was negatively correlated with the concentration of TA-G (Fig 1B, left panel, p = 0. 007, r2 = 0. 40, linear model), with TA-G accounting for 26% of the observed variance. By contrast, larval mass gain was not correlated to the total concentrations of PIEs or TritAcs (Fig 1B, middle and right panel, p = 0. 58 for PIEs; p = 0. 53 for TritAcs, n = 17, linear models). Also, the concentration of TA-G was not correlated to the amount of latex that was released from wounded roots (S1 Fig). Surprisingly, latex mass was positively correlated to larval mass gain when analyzed together with TA-G concentration using multiple linear regression (S2 Table, p (TA-G) = 0. 003, p (latex mass) = 0. 03, linear model). The total amount of TA-G (latex mass * TA-G concentration), on the other hand, was not correlated to larval mass gain (S3 Table). Across the different genotypes, TA-G was constitutively produced and not induced by M. melolontha attack. On the contrary, we observed a trend for a reduction of TA-G concentration in the latex of M. melolontha-attacked roots (S2 Fig, p = 0. 08 t-test,). The magnitude of this effect was similar across genotypes (S3 Fig, p = 0. 0004, linear model). To test if TA-G predominantly accumulates in laticifers and to what extent this accumulation is responsible for the overall TA-G concentration in the roots, we measured TA-G concentrations in latex-drained and latex-containing main roots, as well as latex-free root cortex cells. Draining latex from the roots decreased TA-G concentration by a factor of four (S4 Fig, p = 4x10-6, one-way ANOVA). TA-G concentration in the root cortex was as low as in drained roots (S4 Fig, p = 4x10-6, one-way ANOVA). Across the 17 different genotypes, TA-G concentrations in the latex and in the entire main roots were strongly positively correlated, with TA-G concentrations being about 100-fold higher in latex than in main roots (S5 Fig, p = 0. 004, linear model). Together, these experiments show that TA-G is predominantly stored in the laticifers, and that latex TA-G is responsible for the overall concentration of TA-G in T. officinale roots. To investigate the effect of TA-G on M. melolontha preference and T. officinale performance, we identified and silenced a gene that encodes for a germacrene A synthase, the enzyme that mediates the first committed step of TA-G biosynthesis, by RNAi (Fig 2A). To identify germacrene A candidate genes in T. officinale, we sequenced a transcriptome of the main root and the latex and constructed a reference transcriptome with the pooled reads. Putative germacrene A synthases were identified based on amino acid sequence similarity with two known germacrene A synthases from chicory [46]. Through this approach, we obtained full-length sequences of two putative germacrene A synthase genes, ToGAS1 and ToGAS2, which share 71% identity at the amino acid level. Phylogenetic comparison with other Asteraceae terpene synthases revealed that ToGAS1 belongs to the larger of two germacrene A synthase clusters, while ToGAS2 belonged to the smaller cluster (Fig 2B). Heterologous expression in Escherichia coli showed that both recombinant proteins produced (+) -germacrene A when incubated with the substrate farnesyl diphosphate (FDP) (Fig 2C, S6 Fig). To further characterize the two genes, we analyzed their expression in the outer root cortex, latex, and the entire main root. As ToGAS1 was more strongly expressed than ToGAS2 in both latex and entire main roots (Fig 2D), we targeted ToGAS1 through RNAi by expressing a 191 base pair fragment of this gene under the control of the constitutive 35S promoter. A reduction of TA-G by over 90% compared to wild type was observed in three independently transformed lines: −1, −12b, and −16 (“TA-G-deficient lines”). No reduction in TA-G concentration was found in two other lines, −9 and −15, compared to wild type (all designated as “control lines”) (Fig 2E). The amount of exuded latex did not differ between TA-G-deficient and control lines (S7 Fig). ToGAS1 was suppressed by more than 90% in the TA-G deficient lines compared to control lines, whereas ToGAS2 expression was not affected (S8 Fig). These results show that ToGAS1 is involved in TA-G biosynthesis in T. officinale latex. To test the function of TA-G in planta using the transgenic lines, we first measured the effect of M. melolontha attack on 8 wk-old TA-G-deficient and control T. officinale lines. As noninfested TA-G-deficient and control lines differed in their growth (S9 Fig), we expressed the biomass of herbivore-infested plants relative to the mean biomass of control plants of each genotype. After herbivory, TA-G-deficient lines had lower main and side root mass (Fig 3A, main roots: p = 0. 04; side roots: p = 0. 01, Kruskal-Wallis rank sum test), but not leaf mass (S10 Fig, p = 0. 8, Kruskal-Wallis rank sum test), expressed relative to noninfested plants of each genotype, showing that TA-G-deficient lines suffered a higher percentage of root biomass reduction than control lines. To exclude the possibility that the observed effects are due to differences in root growth, we performed a choice experiment with the TA-G-deficient and control lines using 5 wk-old plants, which did not show any differences in growth or biomass accumulation (S11 Fig). M. melolontha larvae preferred to feed on TA-G-deficient rather than on control lines (Fig 3B, top panel, p = 0. 03, binomial test), resulting in three times higher root mass loss in the TA-G deficient than in the control lines under M. melolontha attack (Fig 3C, p = 0. 04, paired Student’s t test). Additional metabolic profiling revealed that TA-G-deficient and control lines differed in total root protein levels (S12–S14 Figs). However, no correlation of this trait with M. melolontha behavior was found (S15 Fig). To specifically test the effect of TA-G silencing on latex bioactivity, we painted 6 wk-old carrot seedlings with latex from TA-G-deficient and control plants. M. melolontha preferred to feed on carrots painted with latex from the TA-G-deficient lines compared to that from the control lines as measured three hours after the start of the experiment (Fig 3B, lower panel, p = 0. 01, binomial test). Latex profiling revealed that TA-G-deficient lines also had lower PIE levels, suggesting an interaction between the two pathways (S16 and S17 Figs). To test whether TA-G alone is sufficient to reduce larval consumption, we isolated and purified TA-G by preparative chromatography and performed a feeding experiment with M. melolontha larvae feeding on artificial diet containing TA-G. To determine physiologically relevant TA-G concentrations, we first quantified TA-G in different T. officinale tissues. Latex contained 75 μg TA-G per mg per fresh mass, and the main roots, side roots and leaves contained 0. 2–0. 7 μg TA-G per mg fresh mass (Fig 4A). For the artificial diet experiment, we used a concentration of 3 μg TA-G per mg diet to represent a natural situation in which M. melolontha feeds on a root that accumulates latex at the site of wounding. Over 24 h, M. melolontha larvae consumed 40% less TA-G containing diet than control diet (Fig 4B, p = 0. 045, Student’s t test). To investigate whether TA-G benefits vegetative and reproductive fitness under M. melolontha attack in the field, we grew 2,040 T. officinale individuals of the experimental population (consisting of the 17 genotypes as described above) in a common garden. We established 20 circular plots, each of them containing 6 individuals of each genotype, and infested half of the plots with 72 M. melolontha larvae (23 larvae per m2) each (S18 Fig), a density similar to the damage threshold in pastures [47]. In the first year during which most plants did not flower, we measured the length of the longest leaf (“maximal leaf length”) —a reliable predictor for leaf and root mass under greenhouse conditions (S19 Fig) —and correlated this parameter with latex secondary metabolite concentrations. To standardize growth rates, we expressed the size increase of the longest leaf of the herbivore-infested plants relative to the size increase of the longest leaf of control plants of the same genotype (“relative leaf growth”). Shortly after infestation of the plants in June, no reduction in leaf growth was observed in the infested plants, and relative leaf growth was not correlated with the concentration of the three latex secondary metabolite classes (Fig 5A, p (June) = 0. 38, Pearson’s product–moment correlation). In the course of the growing season, M. melolontha infestation reduced overall plant growth, and a positive correlation between relative leaf growth and TA-G concentration emerged, suggesting that TA-G reduced the negative impact of M. melolontha on plant performance (Fig 5A, p (September) = 0. 01, Pearson’s product–moment correlation). In absolute terms, TA-G concentration and leaf growth tended to be positively correlated under M. melolontha attack and negatively correlated in the absence of M. melolontha (S20 Fig). No correlation between relative leaf growth and the total concentrations of PIEs, TritAcs, latex mass, or the total amount of TA-G (latex mass * TA-G concentration) was observed throughout the entire growing season (S21 Fig, S5 Table). Similarly, latex mass did not significantly account for relative leaf length when analyzed in a multiple regression together with TA-G concentration (S6 Table). Leaf length of the herbivore-infested plants was proportional to leaf length of noninfested plants, indicating that plant size did not affect the degree of damage (S22 Fig). To assess whether TA-G also benefits plant reproductive fitness, we correlated the number of flowers to latex secondary metabolite concentration in the following year. At the beginning of the flowering season, TA-G was positively correlated with the relative number of flowers (number of flowers of the herbivore-infested plants expressed relative to noninfested plants of each genotype) in the genotypes that flowered at this time point (Fig 5B, left panel). Genotypes that flowered did not differ in their TA-G concentration from genotypes that did not flower at this time point (p = 1, Wilcoxon rank sum test). No correlation between the relative number of flowers and the total concentrations of PIEs and TritAcs were observed (Fig 5B, middle and right panel). The positive correlation between the relative number of flowers, and TA-G disappeared at the end of the flowering period (p = 0. 33, Pearson’s product–moment correlation), likely because almost all M. melolontha larvae had stopped feeding by this time (S7 Table). Together, these data strongly suggest that TA-G reduces the negative effect of root herbivore attack on plant vegetative and reproductive fitness. In this study, we demonstrate that the sesquiterpene lactone TA-G, a major secondary metabolite of T. officinale, protects the plant against its major native root herbivore M. melolontha. TA-G deters M. melolontha larvae from feeding and thereby directly protects the roots, resulting in a reduction of the negative impact of the root feeder on vegetative and reproductive fitness. The observed pattern indicates that root herbivores may exert positive selection pressure on latex secondary metabolites and may thereby drive their evolution. Our experiments involving natural variation, chemical manipulation, and genetic modification provide parallel lines of evidence for a negative effect of TA-G on M. melolontha larvae. First, across different T. officinale genotypes, TA-G concentration was negatively correlated with M. melolontha growth. Surprisingly, latex mass was positively associated with larval mass gain. Other unidentified plant traits that benefit the larvae and covary with latex exudation may account for this pattern. Second, purified TA-G reduced food consumption in vitro. Third, TA-G suppression through ToGAS1-silencing increased the attractiveness and consumption of T. officinale roots and decreased the deterrent effect of T. officinale latex towards M. melolontha. Interestingly, silencing ToGAS1 not only affected sesquiterpene lactone biosynthesis, but also plant growth and the accumulation of PIEs. Germacrene A synthases convert FDP into germacrene A. The substrate FDP is a common precursor for sesquiterpenes, triterpenes, and phytosterols [48,49], and the farnesyl residue can bind to growth-regulating proteins of the ras family [50]. It is therefore possible that ToGAS1 silencing affects other branches of the metabolism of T. officinale by changing FDP pool sizes. These observations illustrate the limitations of transgenic approaches as a stand-alone method and highlight the power of combining genetic manipulation, natural variation, and chemical complementation to elucidate the role of plant secondary metabolites in plant–herbivore interactions. Many studies demonstrate that plant secondary metabolites are toxic to root and leaf herbivores [2–6]. Surprisingly, however, the benefits for the plant often remain unclear, especially for root herbivores [28,51,52]. Plant secondary metabolites may reduce food quality for herbivores and may thereby trigger compensatory feeding, leading to higher plant damage [53]. In addition, herbivore-imposed loss of biomass can lead to an overcompensation of plant growth and sexual reproduction, which may mask the fitness benefits of resistance factors [54–56]. Secondary metabolites can also reduce plant performance in the field by attracting specialized herbivores that use the chemicals as oviposition [57] and foraging cues [22,58,59]. All these factors may constrain the fitness benefits of bioactive secondary metabolites. Finally, the heterogeneity of natural environments, including varying herbivore communities and abiotic factors, can render the detection of fitness benefits difficult [10–12]. We manipulated the abundance of a major root herbivore within artificial populations consisting of plant genotypes that differ substantially in their capacity to produce plant secondary metabolites. This approach allowed us to demonstrate herbivore-dependent vegetative and reproductive fitness benefits under field conditions. Similar experimental designs could be used in combination with transgenic or genetic mapping populations to quantify the contribution of individual herbivore species and herbivore communities to secondary metabolite-dependent fitness benefits in heterogeneous environments [10–12]. So far, clear evidence for the fitness advantage of particular metabolites under insect attack has remained particularly scarce for below-ground plant–herbivore interactions. Vaughan et al. [60] showed that silencing the production of a semivolatile diterpene increased root damage of Arabidopsis thaliana by the opportunistic fungus gnat Bradysia spp. However, it remains unknown to what extent this effect translates into improved plant performance in nature. The lack of knowledge regarding the benefits of secondary metabolites under root herbivory limits our understanding for the evolution of root metabolites. Eschscholzia californica (Fabaceae) mainland populations that are exposed to pocket gopher herbivory had 2. 5 times higher root alkaloid concentrations than island populations that are free from this herbivore pressure [61], suggesting that pocket gophers may exert positive selection on this metabolite. We show here that high TA-G concentration benefits plant vegetative and reproductive performance in the presence of T. officinale’s major native root herbivore, M. melolontha, thus providing an evolutionary framework for root herbivore-driven natural selection. TA-G-deficient T. officinale lost more root mass than control lines upon feeding by M. melolontha. In a common garden experiment, TA-G concentration was positively correlated with leaf growth and flower production across natural T. officinale genotypes. While our data provides evidence that TA-G benefits the plants under M. melolontha attack, we did not obtain strong evidence for costs of TA-G production. Although TA-G concentration tended to be negatively correlated to plant growth across the 17 genotypes in the common garden experiment in the absence of M. melolontha, the correlation was not significant, possibly due to the relatively low number of genotypes that were used for the experiment. Experiments that evaluate putative fitness costs of TA-G production in different environments may provide further insights into the varying fitness effects of TA-G. Laticifers are commonly assumed to be defensive [28,30]. Toxic metabolites or proteins in the latex can reduce herbivore performance [28], while the stickiness of latex can trap entire insects or glue their mouthparts together [30,62]. Despite the overwhelming evidence that latex reduces herbivore performance, experimental validation that latex benefits plant fitness under herbivore attack remains scarce [31,32]. We show that a toxic metabolite in the latex benefits the plant in the presence but not in the absence of an herbivore and thereby provide an experimental validation of the assumption that microevolutionary processes govern intraspecific variation in plant defense traits. These microevolutionary processes are consistent with the observed macroevolutionary patterns in which latex represents a key innovation that has spurred the evolution of the angiosperms [25]. Taken together, our results furnish an ecological and evolutionary explanation for the high concentrations of root and latex secondary metabolites and highlight the potential of soil-dwelling insects to shape the chemical defenses of their host plants. All indoor experiments were performed in a climate chamber operating under the following conditions: 16 h light 8 h dark; light supplied by a sodium lamp NH 360 FLX SUNLUX ACE Japan; light intensity at plant height: 58 μmol m2 s−1; temperature: day 22°C; night 20°C; humidity: day 55%, night 65% (unless specified otherwise). Plants were potted in 0. 7–1. 2 mm sand and watered with 0. 01%–0. 05% fertilizer with N-P-K of 15-10-15 (Ferty 3, Raselina, Czech Republic). M. melolontha larvae (S23 Fig) were collected from meadows in Switzerland and Germany. Experiments were performed with larvae in the third larval stage (L3) unless indicated otherwise. Insects were reared individually in 200 ml plastic beakers filled with a mix of potting soil and grated carrots in a phytotron operating under the following conditions: 12 h day 12 h night; temperature: day 13°C, night 11°C; humidity: 70%; lighting: none. All statistical analyses were performed in R version 3. 1. 1 [63]. Pairwise comparisons were performed with the agricolae [64] and lsmeans [65] package. Results were displayed using ggplot2 [66] and gridExtra [67]. More details on the individual statistical procedures are given in the experimental sections below. To investigate the effects of latex secondary metabolites on M. melolontha performance, we measured growth of M. melolontha larvae on 17 T. officinale genotypes. To establish an experimental T. officinale population, we screened 40 triploid genotypes from central and northern Europe [68] for secondary metabolite concentrations and growth rates. Twenty genotypes were selected based on maximal difference of latex chemistry with minimal variation in plant growth rate using cluster analysis (S1 Table, S1 Text). Among these 20 genotypes, three genotypes completely lacked TA-G but were later found to contain other unidentified sesquiterpene lactone glycosides (S24 Fig). These genotypes were subsequently excluded from analysis. The remaining 17 genotypes were used to correlate larval growth with latex secondary metabolite concentrations. For each genotype, 12 plants were infested at an age of 7 wk with one preweighed M. melolontha larva, while 12 plants were left herbivore-free. Eleven days after infestation, M. melolontha larvae were recovered and larval mass difference was determined. To measure the concentration of latex secondary metabolites, main roots were cut 1 cm below the tiller and exuding latex collected into Eppendorf tubes and glass vials, immediately flash-frozen in liquid nitrogen and stored at −80°C until extraction. For extraction, 1 ml methanol was added to the plastic tubes, and 1 ml hexane containing 0. 1 mg*ml−1 cholesteryl acetate as internal standard was added to the glass vials. Both types of vessels were vortexed for 5 min, centrifuged, and the supernatant was stored at −80°C until analysis. Methanol samples were measured on a high pressure liquid chromatograph (HPLC 1100 series equipment, Agilent Technologies), coupled to a photodiode array detector (G1315A DAD, Agilent Technologies) and an Esquire 6000 ESI-Ion Trap mass spectrometer (Bruker Daltonics, Bremen, Germany). For quantification, peak areas were integrated at 245 nm for TA-G and at 275 nm for PIEs, and quantified using external standard curves. Hexane samples were analyzed on an Agilent series 6890 gas chromatograph coupled to a flame ionization detector (GC-FID). Individual TritAcs were quantified based on the internal standard. Methodological details for the analytical procedure have previously been described [36]. Correlations between TA-G, total PIE, total TritAc concentrations and total amount of TA-G (TA-G concentration * latex mass) and M. melolontha mass gain, as well as between TA-G concentration and latex fresh mass, were analyzed using linear models on the mean values of each of the 17 genotypes using the metabolite concentration and latex mass of the noninfested plants, as these measurements were not confounded by differential larval feeding activities. The combined effect of TA-G concentration and latex fresh mass on M. melolontha growth was analyzed with a multiple regression based on the mean values of the 17 genotypes of the control treatment. Differences in TA-G concentration between M. melolontha-infested and control plants were analyzed with Student’s t tests. The correlation across the 17 genotypes between TA-G concentrations of the control, and M. melolontha-infested plants were analyzed with a linear model. The assays were performed in two blocks within two months, and latex for GC analysis was collected from a third batch of plants grown in the same growth chamber under identical conditions. To investigate to which extent latex contributes to TA-G measured in the main roots, we measured the correlation between TA-G concentration in the latex and main roots, as well as the difference in TA-G concentration between main roots, latex-drained main roots, and largely latex-free outer main root cortex. To analyze the correlation of TA-G concentration in the latex and main roots, we grew 12 plants of each of the above-mentioned 17 genotypes. Main roots of 9 wk-old plants were cut 1 cm below the tiller and the exuding latex was collected. Main roots were separated from side roots and flash-frozen in liquid nitrogen. Latex was extracted using 1 ml methanol containing 10 μg*ml−1 loganin as an internal standard and analyzed on HPLC-DAD as described above. Peak area was integrated at 245 nm for TA-G and normalized to loganin as an internal standard. Main roots were ground to a fine powder, and 100 mg tissue was extracted with 1 ml methanol containing 10 μg*ml−1 loganin, vortexed, centrifuged, and the supernatant transferred to HPLC vials. Main root samples were analyzed the same way as the latex samples, except that the mobile phases consisted of 0. 1% acetic acid (A) and acetonitrile (B) using following gradient: 0 min: 5% B, 18 min: 43% B, followed by column reconditioning. Peak area was integrated at 245 nm for TA-G and normalized to loganin as internal standard. The correlation between TA-G concentration in the latex and main roots was analyzed with a linear model. To investigate the TA-G concentration of main roots, latex-drained main roots and laticifer-free main root cortex, we grew 16 plants from genotype A34 for 12 wk. To harvest drained and nondrained roots, the main roots of 8 plants were cut 2 cm below the tiller into two 1 mm slices. From one slice, the latex was collected using filter paper (Whatman 40) before freezing it in liquid nitrogen (“drained”). The other slice was flash-frozen without collecting latex (“nondrained”). To harvest the laticifer-free cortex tissue, the outer cortex zones of the main roots of 8 plants were dissected with a razor blade and frozen in liquid nitrogen. All samples were ground to a fine powder, weighed, extracted, and analyzed for TA-G concentrations as described above. Differences in the TA-G concentration between root samples were analyzed with a one-way ANOVA. Pairwise comparisons were performed with a Tukey posthoc test. To identify putative germacrene A synthases, we sequenced the transcriptome of T. officinale main root and latex using Illumina HiSeq 2500. The main roots of six 10 wk-old plants from genotype A34 were cut, the exuding latex was collected into 100 μl homogenization buffer [69] (4 M guanidine isothiocyanate, 100 mM Tris-HCl, pH 7. 0, and 5 mM dithiothreitol) and the latex as well as main root samples were frozen in liquid nitrogen. Main root samples were ground to a fine powder and RNA was extracted from 100 mg tissue with the RNAeasy plant mini kit (Qiagen) following the manufacturer’s instructions. For latex RNA extraction, 900 μl QIAzol lysis reagent was added to the latex samples, vortexed, and RNA isolated using RNAeasy Plant Lipid Tissue Mini Kit (Qiagen) following the standard procedure. On-column DNA digestion for main root and latex samples was performed using DNase free RNase (Qiagen). The six samples for main root and latex were pooled equimolarly. TruSeq RNA-compatible libraries were prepared and PolyA enrichment was performed before sequencing the two transcriptomes on an Illumina HiSeq 2500 with 20 Mio reads per library, 100 base pair, paired end. De novo transcriptome assembly on pooled reads from main root and latex sample was performed using Trinity (version Trinityrnaseq_r20131110) [70,71] running at default settings. Raw reads were deposited in the NCBI Sequence Read Archive (SRA) (BioProject Accession PRJNA301484). To identify putative germacrene A synthase genes in the T. officinale transcriptome, we performed a BLAST analysis using the amino acid sequences of two known germacrene A synthases from chicory as templates [46]. Two putative germacrene A synthase genes were identified and designated as ToGAS1 and ToGAS2. Sequences were deposited in GenBank with the accession numbers KT898039 (ToGAS1) and KT898040 (ToGAS2). For the estimation of a phylogenetic tree of ToGAS1, ToGAS2, and characterized terpene synthases from other Asteraceae (S4 Table), we used the MUSCLE algorithm (gap open, −2. 9; gap extend, 0; hydrophobicity multiplier, 1. 5; clustering method, upgmb) implemented in MEGA5 [72] to compute an amino acid alignment using a neighbor-joining algorithm (Poisson model). All positions with less than 80% site coverage were eliminated. A bootstrap resampling analysis with 1,000 replicates was performed to evaluate tree topology. The two putative germacrene A synthases were heterologously expressed in E. coli to verify their biochemical function. The complete open reading frames (S2 Text) encoding putative proteins with 559 amino acids for ToGAS1 and 583 amino acids for ToGAS2 could be amplified from root cDNA using the primers GAS1fwd (ATGGCAGCAGTTGAAGCCAATGGG) and GAS1rev (TTACATGGGCGAAGAACCTACA) for ToGAS1 and the primers GAS2fwd (ATGGCTCTAGTTAGAAACAACAGTAG) and GAS2rev (TCAGTTTTCGAGACTCGGTGGAGGAC) for ToGAS2. The genes were cloned into the vector pET100/D-TOPO (Invitrogen, Carlsbad, CA, USA) and an E. coli strain BL21 Codon Plus (Invitrogen) was used for heterologous expression. Expression was induced by addition of isopropyl-1-thio-D-galactopyranoside to a final concentration of 1 mM. The cells were collected by centrifugation at 4,000 g for 6 min and disrupted by a 4 × 30 sec treatment with a sonicator in chilled extraction buffer (50 mM Mopso, pH 7. 0, with 5 mM MgCl2,5 mM sodium ascorbate, 0. 5 mM phenylmethanesulfonylfluoride, 5 mM dithiothreitol and 10% v/v glycerol). The cell fragments were removed by centrifugation at 14,000 g, and the supernatant was desalted into assay buffer (10 mM Mopso, pH 7. 0,1 mM dithiothreitol, 10% v/v glycerol) by passage through an Econopac 10DG column (BioRad, Hercules, CA, USA). Enzyme assays were performed in a Teflon-sealed, screw-capped 1 ml GC glass vial containing 50 μl of the bacterial extract and 50 μl assay buffer with 10 μM (E, - E) -FPP, 10 mM MgCl2,0. 2 mM NaWO4 and 0. 1 mM NaF. An SPME (solid phase microextraction) fiber consisting of 100 μm polydimethylsiloxane (SUPELCO, Belafonte, PA, USA) was placed into the headspace of the vial for 60 min incubation at 30°C and then inserted into the injector of the gas chromatograph for analysis of the adsorbed reaction products. GC-MS analysis was conducted using an Agilent 6890 Series gas chromatograph coupled to an Agilent 5973 quadrupole mass selective detector (interface temp, 250°C; quadrupole temp, 150°C; source temp, 230°C; electron energy, 70 eV). The GC was operated with a DB-5MS column (Agilent, Santa Clara, USA, 30 m x 0. 25 mm x 0. 25 μm). The sample (SPME) was injected without split at an initial oven temperature of 50°C. The temperature was held for 2 min, then increased to 240°C with a gradient of 7°C*min−1, and further increased to 300°C with a gradient of 60°C*min−1 and a hold of 2 min. For the GC-MS analysis with a cooler injector, the injector temperature was reduced from 220°C to 150°C. Chiral GC-MS analysis was performed using a R-βDEXsm-column (Restek, Bad Homburg, Germany) and a temperature program from 50°C (2 min hold) at 2°C*min−1 to 220°C (1 min hold). A (+) -germacrene A synthase (MrTPS3) from chamomile (Matricaria recutita) [73] was used to prepare an authentic (+) -germacrene A standard. To measure the expression of ToGAS1 and ToGAS2, we harvested latex, main roots and outer cortex cells of 8 wk-old A34 plants. Plants we cultivated in a growth chamber at 18°C and 75% humidity with a 16-h photoperiod (250 μmol m−2 s−1) in 50% Ricoterlanderde (RICOTER Erdaufbereitung AG, Aarberg, Switzerland), 40% sphagnum peat and 10% sand. Plants were fertilized every week with 0. 1% Plantaktiv 16 + 6 + 26 Typ K (Hauert HBG Dünger, Grossaffoltern, Switzerland) according to the manufacturer`s instructions. Total RNA was isolated from roots using the GeneJET Plant RNA Purification Mini Kit (Thermo Scientific) according to the manufacturer’s instructions. Total RNA was isolated from latex by dissecting the main root with a razor blade and harvesting 10 μl of expelling latex in 100 μl homogenization buffer (see above). After the addition of 900 μl QIAzol Lysis Reagent, RNA was isolated using the RNeasy Lipid Tissue Mini Kit (Qiagen) according to the manufacturer’s instructions. All RNA samples were treated on column with RNase-free DNase I (Qiagen), and the RNA quality and quantity was determined on agarose gels as well as by spectrophotometric analysis using a ND-1000 spectrophotometer (NanoDrop Technologies). From each sample, 1 μg total RNA was used for reverse transcription using oligo (dT) primers and SuperScript II Reverse Transcriptase (Invitrogen) according to the manufacturer’s instructions. The cDNA quality was determined by PCR using the primer combination ToActin-fwd (5`-CGTGACATCAAGGAGAAGC-3`) und ToActin-rev (5`-GCTTGGAGATCCACATCTG-3`). Quantitative real-time PCR (qRT-PCR) was performed according to the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines [74]. All primer sequences were validated in silico (Oligo Proberty Scan, Eurofins MWG, http: //www. mwg-biotech. com) and accepted when they yielded single amplicons as it was proven by melt curve analysis, agarose gel electrophoresis, and sequencing. qRT-PCR primers for ToGAS1 (ToGAS1-fwd, 5`-AAATTTCCCTCCTTCAGTATGGGG-3`; ToGAS1-rev, 5`-CTTATTGGAATCCATGGTTGGATCTAC-3`) and ToGAS2 (ToGAS2-fwd, 5`-CTGATACTACCATTGATGCAACCAC-3`; ToGAS2-rev, 5`-CAGCATCAATCTCTTCTGGATAAAG-3`) were designed to anneal at positions of significant sequence divergence between these two GAS genes to yield specific products. The T. officinale transcription elongation factor encoding EF-1α gene was used as a reference and amplified with the primer combination ToEF1α-fwd (5`-ACTGGTACTTCCCAGGCCGATTGC-3`) and ToEF1α-rev (5`-TTGTTTCACACCAAGGGTGAAGGCG-3`). qRT-PCR experiments were carried out with the LightCycler 96 Real-Time PCR System (Roche Diagnostics International Ltd) using the KAPA SYBR FAST qPCR Kit (Kapa Biosystems) according to the manufacturer’s instructions. For each experiment, three biological replicates were performed with two technical replicates for each biological triplicate. Relative gene expression levels were calculated with the LightCycler 96 Application Software (Version 1. 1. 0. 1320, Roche Diagnostics International Ltd). Expression between different tissues and genes was analyzed with a two-way ANOVA, and pairwise comparison of the expression levels of the two genes performed with Tukey posthoc test. Based on the transcriptome data, we targeted ToGAS1 by RNAi. For silencing, we used the triploid genotype A34 from the above-mentioned 17 T. officinale genotype based on transformation compatibility and intermediate levels of TA-G concentration. A34 is a triploid, synthetic apomict created by crossing a diploid mother from France with diploid pollen from a triploid apomict from the Netherlands [75]. For the construction of the germacrene A synthase RNAi vector, a 191-bp germacrene A synthase PCR fragment was amplified from T. officinale leaf cDNA using the RNAi-dicer optimized primers ToGermA-RNAi-BamHI_fw (5’-aaaGGATCCGGGATAGAGTACCAGAGATT-3’) and ToGermA-RNAi-XhoI_rev (5’-aaaCTCGAGGGCACTAATGTCCCACCTA-3’). This fragment was digested with BamHI and XhoI and inserted into the respective sites of the Gateway vector pENTR4 (Invitrogen). The resulting vector was used for LR recombination (mediated by LR clonase, Invitrogen) with the GW-compatible destination vector pFGC5941 (http: //www. chromDB. org), which contains the CaMV 35S promoter and the chalcone synthase intron from Petunia hybrida. The integrity of the constructs was verified by sequencing and subsequently used for Agrobacterium tumefaciens-mediated stable transformation of the T. officinale A34 genotype using the same method as described previously [76]. The T1 generation of 13 transformed lines was screened for latex secondary metabolite concentrations using three individuals of each line. Main root latex of 8 wk-old T. officinale was collected into Eppendorf tubes and frozen in liquid nitrogen. Latex was extracted as described above using 1 ml methanol containing 10 μg*ml−1 loganin as internal standard. Samples were analyzed on HPLC-DAD as described above. Five lines were selected for further molecular and phenotypic characterization. First, the transgenic lines were confirmed to be triploid by flow cytometry. Second, the insertion of the transgene was verified by PCR and sanger sequencing on genomic DNA using the primer combination P2 + ToGermA-RNAi-XhoI_rev and P3 + ToGermA-RNAi-XhoI_rev, with 5‘-TACCTTCCCACAATTCGTCG-3‘f for P2,5‘-CAGGTATTGGATCCTAGGTG-3‘ for P3 and 5‘-AAACTCGAGGGCACTAATGTCCCACCTA-3’ for ToGermA-RNAi-XhoI_rev. Third, transcript levels of ToGAS1 and ToGAS2 were determined in the T2 generation by qPCR using the primers described above. Four individuals of the TA-G-deficient (RNAi-1, -12b, -16) and control (RNAi-9, -15, WT) lines grown in soil were harvested at an age of 8 wk. Main root tissue was frozen in liquid nitrogen and ground under liquid nitrogen to a fine powder. RNA was extracted using the GeneJET Plant RNA Purification Mini Kit (Thermo Scientific) according to the manufacturer’s instructions. RT-qPCR was performed for ToGAS1, ToGAS2 and ToEF1α as described above (n = 4). Relative gene expression levels were calculated with the LightCycler 96 Application Software. Gene expression was analyzed with generalized linear models using a gamma error distribution for ToGAS1 and a Gaussian error distribution for ToGAS2. Forth, we determined latex fresh mass, latex secondary metabolites, total protein, amino acid and sugar concentrations in the roots. To analyze concentration of TA-G and total PIEs in the transgenic plants, we harvested six individuals of three TA-G-deficient (RNAi-1, -12b, -16) and three control (wild type, RNAi-9, -15) lines in the T2 generation at an age of 8 wk. Main root latex was collected into Eppendorf tubes, which were flash-frozen in liquid nitrogen. 1 ml methanol containing 10 μg*ml−1 loganin and 100 μg*ml−1 salicin as internal standards for TA-G and PIEs, respectively, were added to the Eppendorf tubes. Samples were extracted and analyzed as described above. Differences in the latex fresh mass, as well as in the concentration of TA-G and total PIEs between TA-G-deficient and control lines, were analyzed with one-way ANOVAs. To determine whether total TritAc concentration was affected by silencing, we collected main root latex from 6 individuals of 9 wk-old TA-G-deficient (RNAi-1, -12b, -16) and control lines (wild type, RNAi-9,15) into glass vials, which were immediately frozen in liquid nitrogen. Samples were extracted with 1 ml hexane containing 100 μg*ml−1 cholesteryl acetate as internal standard. Samples were processed and analyzed on GC-FID as described above. Differences in the concentration of total TritAcs were analyzed with a one-way ANOVA. To determine soluble protein, free amino acid and soluble sugar concentrations in the roots of the TA-G-deficient (RNAi-1, -12b, -16) and control (wild type, RNAi-9, -15) lines, we harvested 5 individuals of each line at an age of 12 wk. Root systems were exposed, washed, and main and side roots frozen in liquid nitrogen. Root tissue was ground under liquid nitrogen to a fine powder. For extraction, 1 ml 0. 1 M TRIS-HCl, pH = 7. 0 was added to 100 mg ground tissue, vortexed and centrifuged at room temperature at 17,000 g for 10 min. The supernatant was stored at −20°C until analysis. Soluble protein concentration was determined using the Bradford assay and quantified using a standard curve of albumin [77]. Differences in soluble protein concentrations between TA-G-deficient and control lines, as well as between root tissues, were analyzed with two-way ANOVAs. To determine free amino acid concentrations, 10 μl of the diluted samples were mixed with 90 μl 13C, 15N labelled amino acid mix (20 μg amino acids*ml−1) (Isotec, Miamisburg, USA) and 100 μl borate buffer (0. 9 M, pH = 10. 0). To derivatize amino acids, 22 μl 30 mM fluorenylmethoxy-carbonyl chloride was added and samples were vortexed. After 5 min, 800 μl hexane was added to stop the reaction; the samples were vortexed and placed at room temperature until phases had separated. The lower, aqueous phase was analyzed on an Agilent 1200 HPLC system coupled to an API 5000 tandem mass spectrometer according to [78]. To determine soluble sugar concentrations, main root samples were diluted 1: 10 and side root samples 1: 5 in 0. 1 M TRIS-HCl, pH = 7. 0. Samples were analyzed on an Agilent 1200 HPLC system (Agilent Technologies, Germany) coupled to an API 3200 tandem mass spectrometer (Applied Biosystems, Germany) equipped with a turbospray ion source operating in negative ionization mode. Injection volume was 1 μl. Metabolite separation was accomplished by an apHera NH2,15 cm x 4. 6 mm x 3 μm. The mobile phase consisted of water (A) and acetonitrile (B) utilizing a flow of 1 ml*min−1 with the following gradient: 0 min: 20% A, 0. 5 min: 20% A, 13 min: 45% B, followed by column reconditioning. The column temperature was maintained at 20°C. The ion spray voltage was maintained at −4. 5 keV. The turbo gas temperature was set at 600°C. Nebulizing gas was set at 50 psi, curtain gas at 20 psi, heating gas at 60 psi and collision gas at 5 psi. Multiple reaction monitoring (MRM) was used to monitor analyte parent ion → product ion: m/z 178. 9 →89 (collision energy (CE) −10 V; declustering potential (DP) −25 V), for glucose; m/z 178. 9 →89 (CE −12V; DP −25V) for fructose; m/z 341. 03 →58. 96 (CE -52V; DP -45V) for sucrose; Both Q1 and Q3 quadrupoles were maintained at unit resolution. Analyst 1. 5 software (Applied Biosystems, Darmstadt, Germany) was used for data acquisition and processing. All compounds were identified based on comparison of retention times and mass spectra to those of commercial standards. Glucose, fructose, and sucrose concentrations were quantified using external standard curves obtained from commercial standards. Differences in the sugar concentrations between TA-G-deficient and control lines, as well as between root tissues, were analyzed with two-way ANOVAs. To investigate whether silencing of ToGAS1 affects plant performance, we measured root and leaf mass of three TA-G-deficient (RNAi-1, -12b, -16) and three control (wild type, RNAi-9, -15) lines. For each line, 24 plants of the T2 generation were cultivated for 8 wk. Half of the plants were infested with one preweighed M. melolontha larva. One week after infestation, plants were separated into side roots, main roots and leaves, and plant material was dried for three days at 60°C before weighing. As TA-G-deficient lines had 50% lower root mass than control lines, resistance was expressed relative to the control plants of each genotype (100* (1 − (mass herbivore plant / mean mass control plants of its genotypes) ) ) and analyzed with Kruskal-Wallis rank sum tests. In order to test M. melolontha preference for and plant resistance of TAG-deficient and wild type T. officinale plants, three TA-G-deficient (RNAi-1, -12b, -16) and three control (wild type, RNAi-9, -15) lines were tested in a choice experiment with M. melolontha larvae. Larvae were starved for three days prior to the experiment. Each larva was placed into a 180 ml plastic beaker, which was filled with 2–3 mm vermiculite. The roots of 5 wk-old T. officinale seedlings of the T2 generation (grown in soil in seedling trays) were washed, briefly dried with a tissue and the mass of the plants determined. One TA-G-deficient and one control plant was embedded into the vermiculite-filled beaker at opposite edges, with 9 replicates of each possible pairwise combination. Larval feeding site was scored visually after 3 h by inspecting the beakers from the outside. To determine root mass consumption, plants were recovered after 4 h. The plants were separated into shoots and roots and dried for three days at 60°C. Fresh mass was calculated from dry mass using a common conversion coefficient based on the fresh/dry mass ratio of five non-manipulated seedlings of each genotype. Root mass consumption was analyzed using paired Student’s t tests. To obtain sufficiently large sample sizes for a binomial test, larval preference was analyzed by pooling the data for the three TA-G-deficient and control lines. In order to test whether differences in primary metabolites affected M. melolontha choice, we correlated M. melolontha preference and root mass consumption to total main root protein concentration as determined from 12 wk-old plants as described above. Data were analyzed with Pearson’s product–moment correlation. To test whether M. melolontha preference for TA-G-deficient T. officinale is mediated by latex metabolites, we recorded larval choice among carrot seedlings painted with latex of three TA-G-deficient (RNAi-1, -12b, -16) and three control (RNAi-9, -14, -15) lines. M. melolontha larvae were starved for two days. Each larva was placed into the center of a 180 ml plastic beaker, which was filled with 2–3 mm vermiculite. The roots of the 6 wk-old carrot seedlings were completely covered with latex of 5 mo-old TA-G-deficient and control T. officinale of the T1 generation, cultivated in 21 pots in soil (identical growth conditions as described above, except light source from NH 360 FLX SUNLUX ACE Japan). Seedlings painted with TA-G-deficient and control latex were pairwise arranged on opposite edges of the beaker, resulting in 6–11 replicates of each possible pairwise combination. Larval feeding site was visually scored after three hours. Larval preference was analyzed based on pooled data for the three TA-G-deficient and control lines using a binomial test. To determine physiologically relevant TA-G concentrations for bioassays, we analyzed TA-G concentration from latex, main, and side roots and leaves from three wild type A34 plants. Main root latex of 11 wk-old T. officinale was collected into Eppendorf tubes, frozen in liquid nitrogen and extracted with 1 ml methanol containing 10 μg*ml−1 loganin as an internal standard as described above. Main roots, side roots, and leaf tissues were flash-frozen in liquid nitrogen and ground to fine powder. 100 mg tissue was extracted with 1 ml methanol containing 10 μg*ml−1 loganin, vortexed, centrifuged, and the supernatant transferred to HPLC vials. All samples were analyzed as described above for the analysis of TA-G in the main roots. Peak area was integrated at 245 nm for TA-G and normalized to loganin as internal standard. To test whether TA-G deters M. melolontha, we isolated TA-G from latex and added it to artificial diet at a concentration of 3 μg TA-G*mg−1 diet. TA-G was isolated using 300 ml latex methanol extracts obtained from 300 A34 plants grown in the greenhouse. 10 ml water was added to the methanol extract before methanol was completely evaporated using rotary-evaporation. The aqueous solution was loaded on a Sephadex LH20 (GE-Healthcare, Germany) column with 2. 5 cm x 30 cm dimensions. The compounds were eluted from the column using water at a flow speed of 1 ml*min−1. 15 ml fractions were collected and analyzed for TA-G on an Agilent 1200 HPLC system (Agilent Technologies, Germany) coupled to an API 3200 tandem mass spectrometer (Applied Biosystems, Germany) equipped with a turbospray ion source operating in negative ionization mode. Injection volume was 5 μl using flow injection analysis. The mobile phase consisted of 0. 05% formic acid (A) and acetonitrile (B) utilizing a flow of 1 ml*min−1. 50% A was maintained for 0. 5 min. The column temperature was kept at 20°C. The ion spray voltage was maintained at −4. 5 keV. The turbo gas temperature was set at 600°C. Nebulizing gas was set at 50 psi, curtain gas at 30 psi, heating gas at 60 psi and collision gas at 5 psi. Multiple reaction monitoring (MRM) was used to monitor analyte parent ion → product ion: m/z 423 →261 (collision energy (CE) −14 V; declustering potential (DP) -40 V), for TA-G; m/z 447 →151 (CE -26V; DP -100V) for di-PIEs; m/z 581 →151 (CE -38V; DP -140V) for tri-PIEs. Both Q1 and Q3 quadrupoles were maintained at unit resolution. Analyst 1. 5 software (Applied Biosystems, Darmstadt, Germany) was used for data acquisition and processing. Pure fractions were pooled and lyophilized using an Alpha 1–4 LD plus freeze dryer (Martin Christ GmbH, Germany). 30 M. melolontha larvae were starved for two days before providing them 300 mg artificial diet supplemented with either 30 μl 30 mg*ml−1 TA-G or with 30 μl water as solvent control (artificial diet: 25 g bean flower, 2. 4 g Agar from Roth, Agar-Agar bacteriologist, 105 ml tap water, 33. 3 g cooked and mashed carrots). Larvae were allowed to feed for 24 h inside a 180 ml plastic beaker covered with a moist tissue before the remaining food was weighed. Food consumption was analyzed using Student’s t test. Larvae that consumed less than 30 mg diet were considered inactive and were excluded from the analysis. In order to examine the effects of latex secondary metabolites on plant resistance in the field, we cultivated 2,400 T. officinale from the above-mentioned 20 genotypes in a common garden with and without M. melolontha infestation over one year at a field site in Jena, Germany (50°54' 34. 8" N; 11°34' 00. 1" E). Seeds were surface sterilized, germinated on moist filter paper in petri dishes in spring 2013, and the emerging seedlings were transferred onto peat balls after 10 d. One month after germination, seedlings were conditioned outside for one week before planting them into the field site. At the field site, the top 50 cm soil layer was removed and a metal mesh installed on the ground to confine vertical M. melolontha movement. Experimental units (“plots”) were set up using 20 circular plastic tubes (50 cm depth, 2 m diameter) that were placed on the top of the mesh and filled up with the original soil. One wheelbarrow of peat was mixed with the top 20 cm of soil to facilitate plant growth. In each plot, 6 replicates of all 20 genotypes were placed randomly in a quadratic grid with 10 cm distance between plants. To buffer edge effects, these experimental plants were surrounded with an additional row of T. officinale plants, which were excluded from data analysis. Plants were watered as necessary during the first two months after planting and plots weeded monthly. Three weeks after planting, the length of the longest leaf (“maximal leaf length”) was measured for each plant. Subsequently, 72 late L2 or early L3 M. melolontha larvae were homogenously distributed in half of the plots (“herbivory”), while the remaining plots were not manipulated (“control”). As a nondestructive measurement of plant performance, we measured maximal leaf length—a reliable predictor for above and below ground biomass under greenhouse conditions (S19 Fig) —of each plant every month until the end of the growing season. For statistical analysis, the length of the longest leaf at the beginning of the experiment was subtracted from the maximal leaf length measured each month to reduce the impact of initial differences in plant size (“leaf growth”). To normalize between genotypes, leaf growth of herbivore-treated plants was expressed relative to control plants of the same genotype (“relative leaf growth”). Relativeleafgrowth (j) =Mean (MaxleaflengthH (ij) −InitialmaxleaflengthH (ij) ) Mean (MaxleaflengthC (ij) −InitialmaxleaflengthC (ij) ) with H = herbivore-infested plants C = control plant i = individual plant j = genotype Initial max leaf length = maximal leaf length in June before infestation Correlations between relative leaf growth and TA-G, total PIEs, total TritAcs, latex fresh mass and total TA-G (latex mass * TA-G concentration) were performed based on mean values for each genotype for each month separately with Pearson’s product–moment correlations in R. The combined effect of latex mass and TA-G concentration on relative leaf growth was analyzed with a multiple linear regression. Secondary metabolite concentrations and latex fresh mass were obtained from the experiment with the 20 genotypes in the greenhouse as described above. Three genotypes lacking TA-G were excluded from the analysis due to the presence of unknown and thus unquantifiable sesquiterpene lactone glycosides. In order to test whether damage caused by M. melolontha in the field is proportional to plant size, we assessed the correlation between leaf length of herbivore-infested individuals and leaf length of non-infested individuals of the 17 genotypes with Pearson’s product-moment correlations. To correlate secondary metabolite concentrations to reproductive plant fitness, the number of flowers was counted every month in the following year. Correlations between relative number of flowers (number of flowers of the herbivore-infested plants expressed relative to noninfested plants of each genotype) and TA-G, total PIEs, and total TritAcs were analyzed with linear models and Pearson product–moment correlations based on the mean value of each of the 17 genotypes. Difference in TA-G concentration between genotypes that flowered and genotypes that did not flower at the beginning of the flowering season was analyzed with a Wilcoxon rank sum test based on the mean value of each of the 17 genotypes.

Plant roots produce diverse and abundant blends of bioactive metabolites. One potential function of these compounds is to protect roots against the devastating effects of below ground herbivore attack. However, examples demonstrating such a protective function in native plant-herbivore systems are lacking. Here, we investigated the interaction between the dandelion (T. officinale) and its native root feeding enemy, larvae of the common cockchafer beetle (M. melolontha). Dandelion is known to release secondary metabolite-rich latex from wounded roots, thus we specifically focused on the potential defensive role of these metabolites. By combining natural variation, genetic manipulation and in vitro assays, we demonstrate that taraxinic acid glucoside, a highly concentrated chemical, deters cockchafer larvae and thereby protects the roots. Dandelion plants with high levels of taraxinic acid benefited from this protection in terms of both vegetative and reproductive fitness. Our study demonstrates that a latex metabolite benefits plant fitness under root herbivore attack, a result which provides a mechanistic framework for root herbivore driven natural selection and the evolution of root defenses.

Abstract Introduction Results Discussion Materials and Methods

2016

A Latex Metabolite Benefits Plant Fitness under Root Herbivore Attack

The cochlea not only transduces sound-induced vibration into neural spikes, it also amplifies weak sound to boost its detection. Actuators of this active process are sensory outer hair cells in the organ of Corti, whereas the inner hair cells transduce the resulting motion into electric signals that propagate via the auditory nerve to the brain. However, how the outer hair cells modulate the stimulus to the inner hair cells remains unclear. Here, we combine theoretical modeling and experimental measurements near the cochlear apex to study the way in which length changes of the outer hair cells deform the organ of Corti. We develop a geometry-based kinematic model of the apical organ of Corti that reproduces salient, yet counter-intuitive features of the organ’s motion. Our analysis further uncovers a mechanism by which a static length change of the outer hair cells can sensitively tune the signal transmitted to the sensory inner hair cells. When the outer hair cells are in an elongated state, stimulation of inner hair cells is largely inhibited, whereas outer hair cell contraction leads to a substantial enhancement of sound-evoked motion near the hair bundles. This novel mechanism for regulating the sensitivity of the hearing organ applies to the low frequencies that are most important for the perception of speech and music. We suggest that the proposed mechanism might underlie frequency discrimination at low auditory frequencies, as well as our ability to selectively attend auditory signals in noisy surroundings. Our ability to hear is due to an intricate mechanotransduction process that takes place inside the inner ear. Sound-evoked waves on the basilar membrane, an elastic structure stretching along the cochlear canal, cause the deflection of mechanosensitive hair bundles of the sensory cells, thus gating ion channels in the cell membrane and producing electrical signals that are ultimately transmitted to the brain [1]. The transfer of basilar-membrane motion to deflection of the hair bundles is shaped by the structurally complex organ of Corti (Fig 1 (A) ), the outer hair cells of which can provide mechanical force [2]. Changes in transmembrane voltage cause these cells to change length, a phenomenon referred to as electromotility [3,4]. Furthermore, the hair bundles of outer hair cells can also generate mechanical force [5,6]. Both mechanisms may contribute to an active modulation of the sound-evoked motion of the organ of Corti [7–9]. The active mechanical feedback by outer hair cells is essential for the extraordinary sensitivity, tuning, and dynamic range of mammalian hearing organs, and damage to the outer hair cells consequently results in hearing loss [10–12]. Although this feedback presumably operates on a cycle-by-cycle basis, outer hair cells can also exhibit quasi-static length changes that occur on much slower timescales. In particular, outer hair cells respond to acoustic stimulation through static contraction and sometimes elongation. Moreover, the static length change is largest when the frequency of the stimulation matches the characteristic frequency of the cochlear location of the outer hair cell [13,14]. Although discovered almost thirty years ago, the biophysical relevance of such static length changes of outer hair cells for the functioning of the organ of Corti remains unresolved. Another main uncertainty regarding the feedback by outer hair cells concerns the micromechanics of the organ of Corti in the low-frequency apical region of the cochlea that is responsible for detecting frequencies below a few kHz that are most important for speech and music [15,16]. Recent in vitro experimental studies have indeed shown that the apical organ of Corti deforms in a complex and unexpected way [16–21]. When stimulated electrically, the outer hair cells contracted and pulled the reticular lamina, in which the hair bundles of outer hair cells are anchored, towards the basilar membrane. Surprisingly, the lateral portion of the organ of Corti composed of the Hensen cells moved in the opposite direction, away from the basilar membrane, at an amplitude larger than that of the reticular lamina [20]. No vibration could be detected from the adjacent portion of the basilar membrane [16]. The mechanisms producing this complex motion of the organ remain unclear. Here we set out to identify the origin of the complex internal motion of the organ of Corti at the cochlear apex and the influence of static length changes of outer hair cells. We show that a plausible assumption about the apical organ of Corti, namely that each cross-section is incompressible, highly constrains the organ’s internal motion. The deformation of the organ of Corti that results from length changes of the outer hair cells can then be described through a mathematical model that is based on the organ’s geometry. We develop this model and verify it through comparison with existing [16,20] as well as newly acquired experimental data, where length changes of the outer hair cells were induced by current injections inside scala media. Our results reveal that static length changes of the outer hair cells can sensitively determine how much of the sound-evoked motion is transferred to the reticular lamina, thus providing a novel mechanism for outer hair cells to regulate hearing sensitivity. Sound elicits a traveling wave on the basilar membrane which triggers the deflection of hair bundles and thus the electromotile response of the outer hair cells. As the outer hair cells contract, the reticular lamina and basilar membrane are pulled towards each other [7] (Fig 1). This can potentially reduce the cross-sectional area of the fluid-filled space of Nuel, causing fluid inside the organ of Corti to be displaced longitudinally, that is, along the cochlear canal. The volume of displaced fluid is proportional to the change in cross-sectional area, multiplied by the longitudinal extent l of the organ that contracts. For a traveling wave, this longitudinal extent l is approximately half the wavelength, and the amplitude of the evoked fluid velocity is proportional to the displaced volume and thus to the wavelength. Near the cochlear apex, low-frequency sound elicits a wave with a long wavelength of several millimeters [2,22]. The longitudinal extent over which the organ of Corti deforms similarly thus far exceeds the width and the height of the space of Nuel, which are of the order of 100 μm. Longitudinal fluid flow inside the organ of Corti would thus require velocities much larger than the velocity of the length-changing outer hair cells, and would hence be counteracted by viscous friction. We conclude that such longitudinal flow is suppressed and that the cross-sectional area of the organ of Corti in the apex is accordingly conserved when the outer hair cells change length. The same reasoning holds for in vitro experiments using electrical stimulation, due to the long effective range of the electrodes, but not for deformation of the organ of Corti near the cochlear base where the wavelengths in the peak region of a traveling wave can be much shorter, below one millimeter [2]. The cross-section of the organ of Corti can be divided into two components, a fluid-filled space on the neural side of the organ, and a portion representing the body of Hensen cells on the abneural side (see Fig 1 (B) ). The cross-sectional area of each component needs to be conserved separately: the fluid space because of the argument above, and the Hensen cells because their cytoplasm cannot escape longitudinally. The motion of the cochlear partition can be decomposed into a passive component, where all structures follow the sound-evoked displacement of the basilar membrane [23], and an active component that involves internal deformation of the organ of Corti caused by outer hair cell forces. Here we seek to determine the motion of various structures of the organ of Corti—in particular of the Hensen cells, the reticular lamina, and the outer hair cells—relative to the basilar membrane. We use the constraint of a conserved cross-sectional area to estimate the active deformation of the organ of Corti from its geometry. The length change of outer hair cells is characterized by its relative contraction ϵ, such that the length of an outer hair cell is given by LOHC (ϵ) = (1 − ϵ) LOHC, 0, where LOHC, 0 is the resting length of the cell (Fig 1 (B) ). A length change of the outer hair cells can result from electromotility as well as from hair bundle motility that can exert force on the reticular lamina [24–26]. Experiments using electrical stimuli on isolated and unloaded outer hair cells indicate that the magnitude of the outer hair cell contraction is |ϵ| ≲ 0. 02 [4]. Other anatomical elements of the organ of Corti are assumed to have constant length, except for the Deiter’s cells and the contour of the Hensen cells. Motion of the reticular lamina can be approximated as pivoting about the top of the pillar cells, which is why we lump the three rows of outer hair cells and Deiter’s cell in a single, effective row, located near the third row of outer hair cells. Since we consider small deformations only, we assume linear relationships between the length change of the outer hair cells and the length LDC (ϵ) of the Deiter’s cells as well as the length LHC (ϵ) of the contour of the Hensen cells. We can therefore write LDC (ϵ) = (1 + ϵΔ) LDC, 0 and LHC (ϵ) = (1 + ϵΓ) LHC, 0 with the resting lengths LDC (ϵ = 0) = LDC, 0 and LHC (ϵ = 0) = LHC, 0. The two parameters Δ and Γ quantify the extent to which the Deiter’s cells and the Hensen cell contour change their length as a result of an outer hair cell length change, respectively. In the following, we will refer to them as extensibilities. Note that we have introduced Δ and Γ as purely geometrical parameters; they do not correspond in a simple way to material properties of the Deiter’s or Hensen cells alone. Rather, they are the result of the complex interplay of the material properties and the geometrical arrangement of all the different elements that comprise the cochlear partition and resist deformation through outer hair cell forces. Their values are therefore a priori unknown. Furthermore, we assume that Δ and Γ are constants and, in particular, independent of the frequency of outer hair cell deformations. We thus consider elastic deformations only and neglect effects from viscosity or inertia in the system. However, at the low frequencies considered here, the latter are indeed relatively less important [27]. We assume that the Deiter’s cells can pivot around their attachment on the basilar membrane and that they do not bend. The arc of Hensen cells is treated as a thin elastic body that deforms around a preferred shape, characterized by its local curvature along its length. Details of the model calculations are given in the Materials and Methods. We characterize the motion of the deforming organ of Corti by the motion of specific points along the arc of Hensen cells (Fig 2). Our model shows that, depending on the values of the Deiter’s cell extensibility Δ and the extensibility Γ of the Hensen cell contour, different types of motion can occur. Which of these does occur in experiments? Having constrained both free parameters of our model, we compared the resulting model predictions to additional known features of apical micromechanics. Recent in vitro experiments have shown that outer hair cells essentially pivot around their attachment at the reticular lamina when stimulated electrically [20]. Outer hair cells were first subjected to a negative current, yielding a reference state, and then to a positive current of equal magnitude. The change in current leads to contraction of the outer hair cells which were found to rotate the cell’s base towards the stria vascularis (Fig 5 (A) ). The angle of this rotation was quantified for different amplitudes of electrical stimulation and was found to increase linearly for small stimulation amplitudes but to saturate at larger ones [20] (Fig 5 (B) ). Our model shows that the reticular lamina moves much less than the length change of the outer hair cells which is consistent with the essentially rotational motion of the outer hair cells found experimentally. Direction and amount of the rotation depend on the size of the outer tunnel of Corti as parametrized by the angle φ between the outer hair cells and the arc of the outer tunnel (Figs 1 (B) and 5 (C) ). For simplicity, we here consider the reticular lamina as fixed and regard the organ at the hyporpolarized state of the outer hair cell as the reference position. The amount of contraction considered in Fig 5 (C) therefore ranges from ϵ = 0 to approximately ϵ = 0. 04, rather than from ϵ = −0. 02 to ϵ = 0. 02 as before. Our model correctly predicts the direction of outer hair cell rotation when the outer tunnel is large, which agrees with the geometry commonly seen in micrographs: the realistic geometry is arguably the one where the outer tunnel arc lies in almost tangential continuation of the reticular lamina. Furthermore, the amplitude of the applied current can be related to the amplitude of the length change of the outer hair cells if we assume that the saturation observed for high currents corresponds to the maximal contraction of the outer hair cells of about 4% [4,19]. The predicted rotation angles for a realistic geometry (the bluest curve in Fig 5 (C) ) are then in good quantitative agreement with the experimental data (Fig 5 (B) ). While our model does not explicitly describe displacements of internal points of the Hensen cells, it suggests a motion pattern in which the entire body of Hensen cells is essentially displaced as one by the contracting outer hair cells with little internal deformation. As a consequence, structures at different depths within the organ are expected to show approximately constant vertical displacement, and the displacement decreases only close to the basilar membrane. In particular, the direction of displacement remains the same throughout the entire height of the organ. In contrast, if the cross-section of the organ of Corti were to change, such as through fluid being pressed into the outer tunnel, vertical displacement would vary markedly and change direction as a function of depth. We interferometrically determined current-evoked displacements from positions at different depths. We found that the direction of Hensen-cell displacement, as well as the displacement amplitude, vary little with depth (Fig 6 (A) ). While the direction of the displacements with respect to the applied current was consistent in all preparations, the amplitude of the evoked displacements varied considerably between preparations, as well as with time in a given preparation. For this reason, results shown in Fig 6 (B) have been normalized to the average displacement at the surface of the Hensen cells. The displacement amplitude exhibited a small but significant decrease with increasing depth (13% on average; a linear mixed model reveals a negative slope of -0. 0008/μm in normalized displacement units, p = 0. 0014, t = −3. 22, d. f. = 532; data from 540 measurements from 7 preparations). This agrees with our model that revealed that the counter-intuitive direction of Hensen-cell motion under electrical stimulation is due to large motion at the bases of outer hair cells. As the measurement became increasingly noisy with increasing depth inside the tissue, we were not able to determine the location of the basilar membrane. The fact that large displacement amplitudes persist with depth suggests, however, that some basilar membrane motion occurs underneath the Hensen cells. In contrast, such motion was not detectable in the portion of the basilar membrane lateral to the organ of Corti. This is consistent with recent in vivo measurements obtained using optical coherence tomography [16]. Inner hair cells are responsible for detecting the mechanical sound vibrations and transducing them into electrical signals that are then forwarded to the brain. The hair bundles of the inner hair cells are deflected by oscillatory fluid flow between the reticular lamina and the tectorial membrane, whose magnitude is dependent on the vibration amplitude of the reticular lamina, at least for frequencies up to 3 kHz [17]. Therefore, the nonlinear reticular-lamina displacement upon length change of the outer hair cells that is predicted by our model has striking consequences for inner hair cell stimulation (Fig 4 (B) and 4 (C) ). Sound vibration at a frequency f leads to an oscillating length change of the outer hair cells around some resting position ϵ (0): ϵ (t) = ϵ (0) + ϵ (osc) sin (2 π f t). (1) This length change elicits an oscillating reticular-lamina motion DRL (t) at an amplitude D R L (osc) around the steady displacement D RL (0) that is set by the outer hair cell’s steady contraction ϵ0: D RL (t) = D RL (0) + D RL (osc) sin (2 π f t). (2) The amplitude of an oscillating length change of an outer hair cell for sound pressures in the hearing range is small [28], |ϵ (osc) | ≪ 0. 02. The amplitude of the resulting reticular-lamina vibration D RL (osc) can thus be approximated by a linear expansion around the resting amplitude D RL (0): D RL (osc) = d D RL d ϵ | ϵ (0) ϵ (osc). (3) For the Deiter’s cell extensibility Δ = 1. 15 that we identified above, the derivative of reticular-lamina displacement with respect to hair-cell contraction, dDRL/dϵ, varies monotonically from approximately zero at a resting length change ϵ (0) = −0. 02 of the outer hair cells to a value of approximately -7 at ϵ (0) = 0. 02. As a result, an oscillating length change of the outer hair cells around a maximally elongated resting length defined by ϵ (0) = −0. 02 produces virtually no oscillation of the reticular lamina. On the other hand, a vibration of the outer hair cell length around a maximally contracted resting length defined by ϵ (0) = 0. 02 leads to a seven-fold larger vibration of the reticular lamina (Fig 7). The resting length of the outer hair cells can thus sensitively determine how much vibration of the reticular lamina is elicited by an oscillating length change of the outer hair cells at low frequencies. We have developed a model for the deformation of the organ of Corti that is based on the organ’s geometry as well as on the plausible assumption that the organ of Corti near the cochlear apex is incompressible. The model involves only two parameters that are not derived from the geometry, namely the extensibility of the Deiter’s cells and of the outer edge of the Hensen cells. Qualitative comparison of model predictions with experimental data highly constrains these parameters, and the resulting model predictions agree excellently with further data on the displacement of the outer hair cells and the vertical vibration at different depths in the organ of Corti. A limitation of our model stems from the fact that we have considered the organ of Corti as composed only of spring-like elastic elements and neglected viscous and inertial impedances. However, as the deformation frequency is lowered, the relative importance of the latter components decreases. For frequencies up to several hundred Hertz, the organ of Corti’s impedance (albeit with the tectorial membrane removed) is found to be dominated by stiffness, rather than viscosity [27]. This corresponds to the characteristic frequencies found at the apical locations studied here, which is why our results can provide a valid approximation also for the acoustic response. We also note that our assumption of unimportant viscous and inertial impedance regards only the part of the organ of Corti between the basilar membrane and the reticular lamina, but not the subtectorial space. The latter presumably contributes the major viscous damping to the cochlear partition, and this damping may be counteracted by a cycle-by-cycle length change of outer hair cells [29–31]. Our model generically produces the non-intuitive counterphasic motion of the reticular lamina and the Hensen cells that was recently observed experimentally [17,18,20,21]. Importantly, our analysis suggests that this behaviour does not result from perilymph being pressed against the Hensen cells, as hypothesized recently [21]. Instead, our model and our measurements evidence that the entire body of Hensen cells is being pulled upwards by the contracting outer hair cells. Generally, the experimental data are reproduced if the base of the outer hair cell is allowed to move somewhat more than its apex, such that the largest displacements then occur inside the organ of Corti. Intriguingly, this is corroborated by our own recent in vivo measurements using optical coherence tomography [16]. What is the origin of this internal motion? In our model, the vibrational pattern is achieved through Deiter’s cells that are fairly compliant, at least in response to quasi-static or low-frequency forcing by outer hair cells [32,33]. Alternatively, or in addition, large displacements at the bases of outer hair cells could also occur as a consequence of a locally very compliant basilar membrane [16]. We have not detected basilar-membrane motion lateral of the organ of Corti in response to current stimulation [16,34]. However, our interferometric measurements from different depths inside the Hensen cells indicate that some basilar-membrane motion may be present in a limited region underneath the organ, while the decrease in amplitude with depth suggests that some stretching occurs as well. Conservation of the cross-sectional area of the organ of Corti may then require counterphasic displacement of the arcuate zone of the basilar membrane, as observed by Nuttall et al. in more basal regions in response to electrical stimulation [35]. We did not include this mode of deformation in our model, as no corresponding data are available for the cochlear apex. Current theories of cochlear function suggest that the mechanical activity of outer hair cells serves to amplify the motion of the basilar membrane [2] or the reticular lamina [36,37] in order to render faint sounds more easily detectable for the stereocilia of inner hair cells. In this light, it seems surprising that the largest motion would occur in the interior of the organ. However, our geometrical analysis and experiments suggest that this motion pattern is associated with a nonlinear dependence of the reticular-lamina motion on the resting length of the outer hair cells. In consequence, we find that the resting length of the outer hair cells can control the magnitude of vibration of the reticular lamina that is evoked by an oscillating length change of the outer hair cells. Experimental evidence for this effect comes from the observed nonlinear dependence of sound-evoked motion on an imposed endocochlear potential in vitro [20]. It has been suggested previously that static length changes of the outer hair cells might influence the operating point of hair bundles [38], or of the micromechanics of the organ of Corti as a whole [39], but the details of such a mechanism have remained unclear. Our analysis shows that the incompressibility of the organ of Corti together with a high level of compliance at the base of outer hair cells yields a novel and intriguingly simple mechanism for the outer hair cells to regulate hearing sensitivity through their static length change. While we have thus shown the availability of such a mechanism, further experimental work and improved imaging techniques are needed to verify it in the living cochlea. Our geometrical model quantifies the internal motion of the organ of Corti. The actual sound-evoked and active motion of the cochlear partition is a linear combination of the internal deformation and an overall net displacement. While internal motion is due to active amplification by outer hair cells, the net displacement of the organ can be caused both by sound stimulation as well as by the mechanical activity of outer hair cells. In a recently proposed ratchet mechanism, or unidirectional amplification, the outer hair cells may cause only internal deformation of the organ of Corti without displacement of the basilar membrane [24], in agreement with some recent experimental observations [16]. Further modeling that integrates the geometric model presented here with an analysis of the different forces produced by outer hair cells and their effects on the overall motion of the organ of Corti, as well as further experimental results on the linear or nonlinear response of the reticular lamina and the basilar membrane to varying sound intensity, are needed to clarify these issues. Our findings are particularly relevant for two lines of further research. First, our results could shed new light on the role of the static and frequency-dependent motile response of outer hair cells to acoustic stimulation whose biophysical origin and function in the cochlea remain poorly understood [13,14]. Because our model shows that a sustained length change of outer hair cells can sensitively regulate the reticular lamina’s vibration, the tuned sound-evoked static length changes of outer hair cells can serve as an effective tuning mechanism that can circumvent the poor mechanical tuning of the basilar membrane in the cochlear apex [2]. As set out above, elongated outer hair cells will transfer only little of their oscillating length change to the reticular lamina. The mechanical sound signal elicited by a pure tone may, however, cause outer hair cells at the characteristic position to contract such that their additional oscillatory response to sound is leveraged into a large vibration of the reticular lamina and thus of the hair bundles of the inner hair cells. This effect can thus endow the motion of the reticular lamina with a frequency selectivity that is independent of the mechanical tuning of the basilar membrane which is comparatively poor in the cochlear apex [2]. Second, the discussed principle could present a potential mechanism for efferent medial olivocochlear (MOC) nerve fibers that innervate the outer hair cells to modulate the auditory stimulus [40]. This efferent feedback is thought to play an important role, for instance, in our ability to understand speech in noisy environments. Our results show that efferently-mediated static length changes of the outer hair cells can modify the transfer of outer hair cell activity to reticular-lamina motion. Experiments have indeed observed efferently-induced modifications in the auditory nerve signal that is not found in the mechanics of the basilar membrane, suggesting that inner hair cell stimulation is in part directly due to outer hair cell activity [41]. This effect was present throughout the cochlea, and was particularly prominent in low-frequency regions. A mechanism as the one described here could underlie these observations. Experiments were performed on guinea pigs. All experimental procedures were approved by the local ethics committee (permit N32/13). We used detailed morphometric data on the guinea pig’s organ of Corti in the cochlear apex [42–44] in conjunction with high-quality micrographs [45] as a basis for the geometrical model (Fig 1). Relative sizes and orientations of different structures in the organ of Corti show a high level of consistency between the different data sources. The contour of the Hensen cells is represented by a polynomial curve approximating the shape seen in micrographs. Since we assume the reticular lamina to pivot as a stiff rod around its attachment near the inner hair cell [8,9], we have for simplicity lumped the three rows of outer hair cells and Deiter’s cells into a single one, located at the position of the outermost row. Young pigmented and albino guinea pigs of both sexes weighing 200 to 400 g were used in the current study. The animals were housed at six animals per cage and a 12-hour light/dark cycle. Using procedures approved by the local ethics committee (permit N32/13), the temporal bones were removed, attached to a custom holder, and the bulla opened to expose the cochlea. The preparation was then immersed in oxygenated tissue culture medium (Minimum Essential Medium, Invitrogen, Carlsbad, CA, USA) and a small opening created over scala vestibuli in the apical turn. This opening provided optical access to the organ of Corti and also allowed the tip of a beveled glass microelectrode to be pushed through the otherwise intact Reissner’s membrane. The electrode was used throughout the experiment to monitor the sound-evoked potentials produced by the sensory cells. Data collection was aborted if these potentials underwent sudden changes, or if their initial amplitude was abnormally low. The electrode was also used for injecting electrical currents into scala media. The currents were generated by an optically isolated constant current stimulator (A395, World Precision Instruments, Sarasota, FL, USA). For our different experiments, we used either linear current ramps or current steps as stimuli, with durations between 50 ms and 100 ms and amplitudes of up to 30 μA. Scala tympani was continuously perfused with oxygenated tissue culture medium at a rate of ∼ 0. 6 ml/h, starting within 10 minutes of decapitation, and the perfusion system was also used to introduce the dye RH795 (5 micromolars, Biotium, Howard, CA, USA), which provides fluorescent labeling of the cell membranes of sensory cells and neurons. All experiments were performed at room temperature (21–24°C).

Outer hair cells are highly specialized force producers inside the inner ear: they can change length when stimulated electrically. However, how exactly this electromotile effect contributes to the astonishing sensitivity and frequency selectivity of the inner ear has remained unclear. Here we show for the first time that static length changes of outer hair cells can sensitively regulate how much of a sound signal is passed on to the inner hair cells that forward the signal to the brain. Our analysis holds for the apical region of the inner ear that is responsible for detecting the low frequencies that matter most in speech and music. This shows a mechanisms for how frequency-selectivity can be achieved at low frequencies. It also opens a path for the efferent neural system to regulate hearing sensitivity.

Abstract Introduction Results Discussion Materials and methods

medicine and health sciences classical mechanics engineering and technology vibration ears signal processing neuroscience surgical and invasive medical procedures labyrinth supporting cells epithelial cells outer hair cells organ of corti functional electrical stimulation hensen cells animal cells biological tissue head inner hair cells physics cellular neuroscience speech signal processing cell biology anatomy neurons epithelium biology and life sciences cellular types afferent neurons deiter's cells physical sciences

2018

Static length changes of cochlear outer hair cells can tune low-frequency hearing

Probability matching is a classic theory of decision making that was first developed in models of cognition. Posterior probability matching, a variant in which observers match their response probabilities to the posterior probability of each response being correct, is being used increasingly often in models of perception. However, little is known about whether posterior probability matching is consistent with the vast literature on vision and hearing that has developed within signal detection theory. Here we test posterior probability matching models using two tools from detection theory. First, we examine the models’ performance in a two-pass experiment, where each block of trials is presented twice, and we measure the proportion of times that the model gives the same response twice to repeated stimuli. We show that at low performance levels, posterior probability matching models give highly inconsistent responses across repeated presentations of identical trials. We find that practised human observers are more consistent across repeated trials than these models predict, and we find some evidence that less practised observers more consistent as well. Second, we compare the performance of posterior probability matching models on a discrimination task to the performance of a theoretical ideal observer that achieves the best possible performance. We find that posterior probability matching is very inefficient at low-to-moderate performance levels, and that human observers can be more efficient than is ever possible according to posterior probability matching models. These findings support classic signal detection models, and rule out a broad class of posterior probability matching models for expert performance on perceptual tasks that range in complexity from contrast discrimination to symmetry detection. However, our findings leave open the possibility that inexperienced observers may show posterior probability matching behaviour, and our methods provide new tools for testing for such a strategy. Human decision making is partly random, in the sense that a person can make different decisions on different occasions based on the same information. Probability matching is a theory of decision making that aims to account for this randomness. Suppose a person believes that response A has a 70% probability of being correct, and response B has a 30% probability of being correct. A person who exhibits probability matching chooses response A with 70% probability and response B with 30% probability. This is a surprising decision strategy, because it means that the person sometimes chooses the response that is less likely to be correct according to the available evidence. Nevertheless, many studies support probability matching as a model of decision making in cognitive tasks such as probability learning [1–2]. Probability matching originated in models of cognition, and variants of probability matching have also been used in models of perception. For example, Mamassian and Landy [3] had observers judge the three-dimensional shapes depicted in line drawings. In their model of this task, subjects used the intrinsically ambiguous shape information from line drawings along with assumptions about the statistical distribution of shapes in the real world to estimate the probabilities that a line drawing depicted an elliptical shape or a saddle shape. Subjects then used probability matching to choose their response, either “elliptical” or “saddle”. Acerbi et al. [4] call such a strategy “posterior probability matching, ” because the subject matches their response probabilities to the posterior probability of each response being correct, instead of the prior (i. e. , baseline) probability as in classic probability matching models. Posterior probability matching and related approaches have become increasingly common in models of perceptual decision making [3–8]. Posterior probability matching has some appealing features for models of perception. It offers an explanation of why perceptual decisions have a random component at all. Furthermore, it has no free parameters, which one might hope could explain why subjects’ decisions show a limited range of randomness across many perceptual tasks [9,10]. However, little is known about whether posterior probability matching is consistent with well-supported models of perception that have been developed within signal detection theory, which take a very different approach to modelling randomness in decision making. Signal detection models typically assume that observers’ decisions depend both on information received from stimuli, and on random fluctuations in perceptual mechanisms, i. e. , internal noise. If the same stimuli are repeated on different trials, the subject may make different responses, because the internal noise contributions may be different. A large literature supports signal detection models of perceptual and cognitive decision making [11]. Furthermore, posterior probability matching models represent an unusual mix of optimal and suboptimal behaviours: these models state that observers calculate the posterior probabilities that would enable them to make optimal responses based on the available stimulus information (e. g. , maximizing payoff according to some utility function), but then instead behave suboptimally by making stochastic responses that match their response probabilities to the posterior probabilities. Here we compare posterior probability matching and signal detection models of perceptual decision making. We begin by defining two very general classes of these models, and we examine the models’ behaviour using two psychophysical methods. First, we use the two-pass response consistency method, which quantifies the randomness in an observer’s decisions by examining how consistent decisions are across repeated presentations of identical trials [9,12]. In a two-pass response consistency task the observer views one of two possible signals shown in noise, and attempts to identify the signal. This continues for some number of trials. The observer then sees the identical sequence of trials a second time (i. e. , the same signals in the same samples of noise), without knowing that they are being repeated, and again attempts to identify the signal on each trial. The experimenter measures the proportion of correct responses, PC, and also the proportion of consistent responses, PA, i. e. , the proportion of repeated trials on which the observer gives the same response twice. (Here the subscript “A” stands for “agreement”.) Burgess and Colborne [12] show how to use PC and PA to calculate the relative amounts of internal and external noise, σI/σE, in the decision variable of an observer. As a second test of posterior probability matching models, we use ideal observer analysis, which measures an observer’s efficiency at a task by comparing the observer’s performance to the best performance that is theoretically possible on the task [13,14]. We examine posterior probability matching and signal detection models' predictions as to how response consistency and efficiency should vary as a function of proportion correct. To test these models we compare their predictions to human observers’ behaviour on a two-pass response consistency task, and to results of previous experiments on response consistency and efficiency. In Experiment 1 we compare human observers' response consistency on repeated trials in a two alternative forced choice (2AFC) discrimination task to the predictions of posterior probability matching and signal detection models. In Experiment 2 we test a larger number of inexperienced observers on a shorter version of the same task in order to sample a wider range of observers and examine the effect of practice. We examined the performance of posterior probability matching and signal detection models in a task where there are two possible signals (A and B) and two possible responses. This includes two-alternative identification tasks, and it also includes 2AFC tasks because we can take “signal A” to mean the two stimulus intervals in one order, and “signal B” to mean the other order. To make the models as general as possible, we modelled observers at the level of the decision variable instead of choosing specific visual or auditory stimuli. The decision variable was D = E+I, where E is a normal random variable representing the contribution of the stimulus to the decision variable, and I is a normal random variable representing internal noise. E had mean zero on signal A trials, mean μE on signal B trials, and standard deviation σE on both types of trials. The internal noise I had mean zero and standard deviation σI, and was statistically independent of E. Signals A and B were equally likely. We used the decision variable D to represent the information that the observer computes from a complete 2AFC trial. The most common model of 2AFC tasks, the difference model [11], assumes that observers compute one decision variable D1 from the first stimulus interval and another decision variable D2 from the second interval, and base their decisions on D = D1−D2. Our approach does not rely on the difference rule, and we do not need to consider the single-interval decision variables D1 and D2. On each trial the model observers received a sample d from the decision variable D, and calculated the likelihoods that the value d would be generated on signal A and signal B trials: P (d|A) =ϕ (d, 0, (σE2+σI2) 1/2) (1) P (d|B) =ϕ (d, μE, (σE2+σI2) 1/2) (2) Here ϕ (x, μ, σ) is the normal probability density function. The observers used these likelihoods in Bayes’ theorem to find the posterior probability that the signal was A or B: P (A|d) =P (d|A) P (A) P (d|A) P (A) +P (d|B) P (B) =P (d|A) P (d|A) +P (d|B) (3) P (B|d) =1−P (A|d) (4) Here we have used the fact that P (A) = P (B). The posterior probability matching observer chose response A with probability P (A|d) and response B with probability P (B|d). We call this' veridical posterior probability matching' (VPPM), because the observer makes veridical estimates of the posterior probabilities given the value of the decision variable, and uses these probabilities in the probability matching rule. In the rest of this article, when we speak of' posterior probability matching' we mean the VPPM model unless we specify otherwise. The signal detection observer used a maximum a posteriori (MAP) decision rule, and chose the response that had the greater posterior probability, P (A|d) or P (B|d). When the goal is to maximize the number of correct responses, the MAP rule is the statistically optimal strategy in this task. We modelled the VPPM and MAP observers in a two-pass task as follows. On the first pass of each trial, the decision variable was D = E+I, with E and I drawn from their respective distributions. On the second pass the external component E of the decision variable was the same as on the first pass, and the internal component I was a new, independent sample. In S1 Text we show that the posterior probability matching observer' s proportion correct in this task is pC=∫−∞∞ϕ (u) 2ϕ (u) +ϕ (u−d′D) du (5) Here ϕ (x) is the standard normal probability density function, and d′D is the signal-to-noise ratio of the decision variable D, d′D=μD/ (σE2+σI2) 1/2. We also show that with ρ = σI/σE, the probability of two correct responses on repeated trials is pCC= (1+ρ−2) (1+ρ2) 1/2∫−∞∞ (∫−∞∞ϕ (u+v) ϕ (u+v) +ϕ (u+v−d′D) ϕ (v (1+ρ−2) 1/2) dv) 2ϕ (u (1+ρ2) 1/2) du (6) and the probability of two incorrect responses is pII= (1+ρ−2) (1+ρ2) 1/2∫−∞∞ (∫−∞∞ϕ (u+v−d′D) ϕ (u+v) +ϕ (u+v−d′D) ϕ (v (1+ρ−2) 1/2) dv) 2ϕ (u (1+ρ2) 1/2) du (7) The probability of consistent responses across two repeated trials is PA = PCC+PII. Green and Swets [11] show that the unbiased MAP observer' s proportion correct is pC=Φ (d′D/2) (8) Here Φ (x) is the standard normal cumulative distribution function. In S1 Text we show that the probability of the MAP observer making two correct responses on repeated trials is pCC=ρ∫−∞∞Φ (d′D2 (1+ρ−2) 1/2−u) 2ϕ (ρu) du (9) and the probability of two incorrect responses is pII=ρ∫−∞∞ (1−Φ (d′D2 (1+ρ−2) 1/2−u) ) 2ϕ (ρu) du (10) Again, the probability of consistent responses is PA = PCC+PII. We used Eqs (5) to (10) to find the proportion correct and proportion of consistent responses for the VPPM and MAP model observers at several signal-to-noise ratios (d′D) and internal-to-external noise ratios (σI/σE). We used eight values of d′D, evenly spaced from 0 to 2. 6. We used σI/σE = 0,1, and 2, which spans the range of internal-to-external noise ratios typically found with human observers [10]. Then, following Burgess and Colborne [12], we calculated the model observers’ apparent internal-to-external noise ratios from their proportion correct and proportion of consistent responses. Eqs (8), (9), and (10) give the proportion correct PC (d′D, ρ) and proportion of consistent responses PA (d′D, ρ) for a MAP observer as a function of d′D and ρ = σI/σE. We found the apparent internal-to-external noise ratio for the model observers by numerically minimizing the following sum-of-squares error: (d′D^, ρ˜) =argmin (d′D, ρ) (pC^−pC (d′D, ρ) ) 2+ (pA^−pA (d′D, ρ) ) 2 (11) Burgess and Colborne’s method estimates the internal-to-external noise ratio of the decision variable of a MAP observer. Thus when we applied Burgess and Colborne’s method to the MAP model observer, we simply recovered the internal-to-external noise ratio that we had used to calculate PC and PA in the first place. When we applied the same method to the posterior probability matching observer, though, we did not recover the internal-to-external noise ratio that we had used to calculate PC and PA, because this observer has an additional source of internal variability, namely the probability matching rule. We emphasize this point: we applied Burgess and Colborne’s method to the posterior probability matching model observer, even though this observer does not satisfy that method’s assumption that additive internal noise is the only source of randomness in the observer’s decisions. We did this in order to discover what internal-to-external noise ratios Burgess and Colborne’s method would attribute to a posterior probability matching observer. We then used Burgess and Colborne’s method with human observers (see below) to see whether they showed the pattern of internal-to-external noise ratios that is predicted by the VPPM or MAP models. We will use σI/σE (or sometimes ρ for brevity) to denote the internal-to-external noise ratios that we used to calculate PC and PA for the model observers, and we will use ρ˜ to denote the apparent internal-to-external noise ratios that we calculated from PC and PA using Burgess and Colborne' s method. An observer’s efficiency at a task can be defined as the squared ratio of their d′ to the ideal observer’s d′: η= (d′/d′ideal) 2 [15,16]. The ideal observer is a theoretical model observer that achieves the best possible performance on the task. The VPPM observer’s proportion correct as a function of d′D is given by Eq (5), and the observer is unbiased, so its sensitivity is d′ = 2Φ−1 (PC). (However, this method of calculating d′ is really only meaningful when the observer chooses responses by comparing the decision variable to a criterion, and the VPPM observer does not do this. Consequently, this is the apparent d′ that an experimenter who assumes signal detection theory would attribute to the VPPM observer, based on its proportion correct. This is fine for our purposes, since our goal is simply to find the VPPM observer’s efficiency.) To examine the best-case scenario, we assumed that the VPPM observer had no internal noise and that the decision variable captured all relevant information from the stimulus. In this case the ideal observer’s d′ is the signal-to-noise ratio of the decision variable, d′ideal = d′D. Thus the VPPM observer’s highest possible efficiency is η = (2Φ−1 (PC) /d′D) 2. We calculated the VPPM observer' s efficiency with σI/σE = 0 and several values of d′D evenly spaced from 0. 3 to 4. 7. Fig 1a–1c shows the model observers’ proportion correct and proportion of consistent responses at several signal levels and internal-to-external noise ratios σI/σE. At any given level of signal and noise, the posterior probability matching observer has lower proportion correct than the MAP observer. Furthermore, at any given proportion correct the posterior probability matching observer’s responses are less consistent than the MAP observer’s, i. e. , they have lower PA. Fig 1d shows the model observers’ apparent internal-to-external noise ratios ρ˜, calculated from the proportion correct and proportion of consistent responses shown in Fig 1a–1c. The posterior probability matching observer has very high apparent internal-to-external noise ratios at low proportion correct, and lower but still quite high ratios at higher proportion correct. The MAP observer has constant apparent internal-to-external noise ratios across all performance levels, as expected, since as explained earlier the apparent internal-to-external noise ratio ρ˜ simply recovers the internal-to-external noise ratio σI/σE we used to calculate the proportion correct PC and the proportion of consistent responses PA. Fig 2 shows the posterior probability matching observer’s efficiency as function of proportion correct. Posterior probability matching is very inefficient at low performance levels, and even at a typical threshold performance level of 75% correct it reduces an otherwise ideal observer’s efficiency to around 50%. Efficiencies of 50% have been found in perceptual tasks at 75% threshold [13], and to be this efficient a posterior probability matching observer would have to make optimal use of the stimulus in all other respects, aside from probability matching. Furthermore, in a contrast increment detection task, Burgess, Wagner, Jennings, and Barlow [15] found efficiencies as high as 83% (standard error 15%) at 69% correct performance, and Fig 2 shows that at this proportion correct the posterior probability matching observer' s maximum efficiency is 44%. Even in a task as complex as symmetry detection, Barlow [16] found efficiencies of 50% at 60% correct performance, whereas the posterior probability matching observer' s maximum efficiency at this proportion correct is just 26%. Many authors have noted that probability matching is a suboptimal strategy, but perhaps it has not been realized previously how very inefficient it actually is. In contrast, the MAP observer makes optimal use of the decision variable. To calculate the posterior probability matching observer' s efficiency (Fig 2), we assumed that the internal-to-external noise ratio was zero, and that the decision variable captured all task-relevant information. Under these conditions the MAP observer is the ideal observer, so its efficiency is 100% at all values of proportion correct, and naturally no human observer can be more efficient than this. Fig 3a shows the results of the two-pass response consistency experiments with human observers, and also a section of the lowest red curve in Fig 1d, which represents the posterior probability matching observer with no internal noise (solid red line). For reference, the dashed red lines show internal-to-external noise ratios at twice the value and half the value of the solid red line. The observers’ apparent internal-to-external noise ratios (coloured circles) are approximately constant across performance levels, and do not show the sharp increase at low proportion correct that is predicted by the posterior probability matching model. For the three observers, least-squares linear regressions of the apparent internal-to-external noise ratio against proportion correct have slopes and bootstrapped 95% confidence intervals of -0. 37 (-2. 85,2. 03) (red data points), 0. 28 (-1. 97,2. 21) (green data points), and -0. 46 (-4. 24,2. 36) (blue data points). Observers' performance approximately spans the range 58% to 75% correct, and at these values the noiseless posterior probability matching observer (solid red line) has apparent internal-to-external noise ratios of 1. 89 and 0. 93, respectively, corresponding to a slope of -5. 6, which is well outside the 95% confidence intervals of the linear regression slopes for all three observers. Furthermore, at low proportion correct (<70%) the observers’ apparent internal-to-external noise ratios are lower than is ever possible according to the posterior probability matching model. These results show decisively that observers did not follow a posterior probability matching strategy in this task. Observers clearly did not use a posterior probability matching strategy in Experiment 1. However, only three observers ran in the experiment, and they ran in almost 10,000 trials. Previous studies on classic probability matching (i. e. , prior probability matching) have found large individual differences, with only some observers showing probability matching behaviour [1,2]. Furthermore, previous studies have found that classic probability matching behaviour decreases with practice, e. g. , Shanks et al. [1] found that probability matching declined over the course of 1800 trials, with less than half the participants showing probability matching by the end of the experiment. These findings raise concerns that in Experiment 1 we may simply have chosen three observers who did not happen to exhibit probability matching behaviour, or that any probability matching behaviour may have been eliminated over the course of the experiment. To address these concerns, in Experiment 2 we ran a larger number of observers in a shorter version of the same task. Fig 3b shows observers’ apparent internal-to-external noise ratios as a function of proportion correct. Each observer is represented by a different coloured symbol, so each coloured symbol appears twice, once for the observer’s low performance trials (at the estimated 65% threshold) and once for the observer’s high performance trials (at the estimated 80% threshold). We clipped apparent internal-to-external noise ratios at a maximum of 4. 0 in order to make them visible on the plot. The internal-to-external noise ratios are higher than in Experiment 1, probably because observers were less psychophysically experienced and did not run in the task as long. Confidence intervals are also much larger than in Experiment 1, for two reasons. First, there were fewer trials, which increased the standard error of the estimates of proportion correct and proportion of consistent responses. Second, internal-to-external noise ratios were higher, and Fig 1a–1c shows that iso-ρ lines are closer together at higher values of ρ, meaning that standard errors on estimates of proportion correct and proportion of consistent responses translate into larger confidence intervals on estimates of ρ. Although the data is noisy, we can still test the VPPM model’s predictions. As explained earlier, the VPPM model predicts that apparent internal-to-external noise ratios increase sharply at low performance levels (Fig 1d). We will denote each observer’s apparent internal-to-external noise ratio at the low performance level by ρ˜low, and the apparent internal-to-external noise ratio at the high performance level by ρ˜high. Fig 4 shows the ratio ρ˜low/ρ˜high for each observer, versus the ratio ρ˜low/ρ˜high predicted by the VPPM model with no additive decision noise. We calculated the VPPM model’s predictions by looking up the predicted ρ˜low and ρ˜high for the noiseless VPPM model in Fig 1d (lowest red line), at each observer’s lower and higher proportion correct, and taking the ratio ρ˜low/ρ˜high. We clipped the ratios at a maximum of 4. 0 to make them visible on the plot. Most data points in Fig 4 fall below the main diagonal, indicating that the ratio ρ˜low/ρ˜high is not as high as predicted by the noiseless VPPM model. A sign test shows that ρ˜low/ρ˜high is significantly lower than predicted by the noiseless VPPM model (17 of 21 data points below the diagonal, p<0. 001). However, we can also run less stringent tests of the VPPM model. The sign test reported in the previous paragraph tests the noiseless VPPM model, but observers may have sources of internal noise besides the posterior probability matching rule. If we re-run the sign test, comparing the actual ratio ρ˜low/ρ˜high for human observers to the ratio predicted by the VPPM model with an internal-to-external noise ratio of σI/σE = 1 (Fig 1d, middle red line), we find that the human observers’ ratio is lower than predicted in only 10 of 21 cases, which is not statistically significant according to a sign test (p = 0. 50). Similarly, we can test the VPPM model simply by asking whether ρ˜low>ρ˜high for human observers, as is predicted by the VPPM model with any value of σI/σE. The prediction ρ˜low>ρ˜high is true for 12 of 21 observers, which is not a significant difference (p = 0. 19). Thus our results with unpractised observers can rule out the noiseless VPPM model (which is what has been tested in previous studies of posterior probability matching in perceptual decision making), but they cannot rule out the VPPM model with additional sources of internal noise. Would minor adjustments to the VPPM model produce better matches to human performance? A simple argument shows that many details of the model are unimportant, and that our findings are very general. In the VPPM model, the observer calculates veridical posterior probabilities P (A|d) and P (B|d) on each trial. On trials where PC = 0. 5, the observer’s posterior probability estimates are P (A|d) = P (B|d) = 0. 5, since any other values imply a higher proportion correct. On these trials the observer chooses responses A and B with probability 0. 5, and so the proportion of consistent responses across repeated trials is also 0. 5. However, PC = 0. 5 and PA = 0. 5 imply an apparent internal-to-external noise ratio of ρ˜=∞ (i. e. , the lower end of the blue line labelled ∞ in Fig 1a–1c passes through the point (0. 5,0. 5) ). This means that when the VPPM observer is uncertain (P (A|d) ≃P (B|d) ≃0. 5), it must also be highly inconsistent across repeated trials. This is very different from a MAP observer, which may be uncertain and yet highly consistent if it has little or no internal noise. Consequently, if we use Burgess and Colborne' s method to interpret randomness in terms of internal noise, we must attribute a near-infinite internal-to-external noise ratio to a posterior probability matching observer as its performance approaches chance, so long as the observer bases their responses on veridical estimates of posterior probabilities. In contrast, a non-veridical posterior probability matching observer, i. e. , one that makes inaccurate estimates of P (A|d) and P (B|d), could produce very different results from those in Fig 1. Suppose that a posterior probability matching observer always assigns zero or one to P (A|d) and P (B|d), even when performing at chance, and suppose that the observer has no additive internal noise (σI/σE = 0). This observer’s two-pass responses will be completely consistent, and the observer will have an apparent internal-to-external noise ratio of zero at all levels of proportion correct. It may also be possible to construct a non-veridical posterior probability matching observer that produces a more psychophysically plausible internal-to-external noise ratio. However, it is not clear what would be gained by constructing such a model, where probability matching is applied to inaccurate estimates of posterior probabilities that are, in effect, artificially chosen to avoid the high variability of posterior probability matching responses at low confidence levels. Nevertheless, this observation helps to define the range of models that our findings rule out. The VPPM model should not be confused with an earlier probability matching model that was developed within signal detection theory [21–23]. In the earlier model the observer’s responses depend on whether the decision variable is greater than or less than a criterion. The observer chooses the criterion that makes their long-term response probabilities match the baseline signal probabilities, i. e. , P (R = A) = P (A) and P (R = B) = P (B), instead of choosing the criterion that maximizes either the number of correct responses, or the expected payoff based on the signal probabilities and a payoff matrix. This is simply a standard signal detection model with a suboptimal criterion, and is quite different from the intrinsically random VPPM model that we have tested here and that has appeared in recent models of perceptual decision making. Furthermore, VPPM differs from prior probability matching models such as the one considered (and rejected) by Eckstein et al. [24], in which observers' response probabilities on each trial are simply matched to the baseline probabilities of the various signals, and the observer' s response on a given trial does not even depend on the stimulus shown on that trial. A similar model has also been tested (and rejected) as an account of how observers distribute attention across possible signal locations [25]. It may be that people exhibit posterior probability matching in some types of perceptual tasks but not others. For example, there is at least one notable difference between our contrast discrimination task, where some observers did not exhibit posterior probability matching, and Gifford et al. ’s [6] task, where model fitting suggests that they did. In our task, subjects discriminated between black-disk-first and white-disk-first stimuli, and the performance-limiting factors were the faintness of the signals and the power of the external noise. In Gifford et al. ’s task, subjects categorized auditory signals of various frequencies as having been generated from one of two broad, overlapping frequency ranges. Subjects’ ability to perceive the signal frequency precisely was not the performance-limiting factor. Rather, the task was difficult because the two categories overlapped, and contained many of the same frequencies. Thus, even though this was a simple frequency categorization task, it was not limited by low-level sensory information, but instead by the fact that clearly perceived signals gave ambiguous information about the correct response. In this way it is similar to the cued probability learning tasks that have supported prior probability matching models in the past [26], i. e. , models in which observers match their response probabilities to the baseline probabilities of the signals being viewed. We suggest that human observers may be more likely to show posterior probability matching in tasks where performance is limited by the ambiguity of clearly perceived signals, than in simple discrimination tasks that are limited by the perceptual signal-to-noise ratio. In summary, posterior probability matching causes highly inconsistent responses, especially at low performance levels, and it is also inefficient. We can definitively rule out posterior probability matching models for practised observers in four typical perceptual tasks, where human observers are more consistent or more efficient than such models allow: the contrast polarity discrimination task reported here, sine wave detection [12], sine wave contrast discrimination [15], and symmetry detection [16]. We also find evidence that less practised observers are more consistent at low performance levels than noiseless posterior probability matching models predict, but these results are consistent with a posterior probability matching model with additional internal noise. We conclude that posterior probability matching is not a promising general-purpose model of expert perceptual decisions that are limited by low-level perceptual information. However, it may be viable as a theory of untrained perceptual decision making, and of decision making in tasks where performance is not limited by low-level perceptual information. The methods we have introduced also provide new tools for testing posterior probability matching models in more complex perceptual and cognitive tasks.

Decision making is partly random: a person can make different decisions at different times based on the same information. The theory of probability matching says that one reason for this randomness is that people usually choose the response that they think is most likely to be correct, but they sometimes intentionally choose the response that they think is less likely to be correct. Probability matching is a theory that was developed to describe how people try to predict the outcome of a partly random event, e. g. , whether a patient has some medical condition, given the result of a medical test that does not provide perfectly accurate information. Recently, modified probability matching theories have been used to understand perceptual decision making, e. g. , judging whether a sound and a visual flash were produced by the same event or by different events. We show that probability matching predicts that peoples’ perceptual decisions on difficult tasks are highly random and make poor use of the available information. We show experimentally that expert perceptual decisions are less random and more efficient than probability matching predicts. These findings help us understand how people perform a wide range of important real-world perceptual tasks, such as evaluating medical images and detecting targets in airport screening scans.

Abstract Introduction Models Results Discussion

2015

Posterior Probability Matching and Human Perceptual Decision Making

Computational protein design has been successful in modeling fixed backbone proteins in a single conformation. However, when modeling large ensembles of flexible proteins, current methods in protein design have been insufficient. Large barriers in the energy landscape are difficult to traverse while redesigning a protein sequence, and as a result current design methods only sample a fraction of available sequence space. We propose a new computational approach that combines traditional structure-based modeling using the Rosetta software suite with machine learning and integer linear programming to overcome limitations in the Rosetta sampling methods. We demonstrate the effectiveness of this method, which we call BROAD, by benchmarking the performance on increasing predicted breadth of anti-HIV antibodies. We use this novel method to increase predicted breadth of naturally-occurring antibody VRC23 against a panel of 180 divergent HIV viral strains and achieve 100% predicted binding against the panel. In addition, we compare the performance of this method to state-of-the-art multistate design in Rosetta and show that we can outperform the existing method significantly. We further demonstrate that sequences recovered by this method recover known binding motifs of broadly neutralizing anti-HIV antibodies. Finally, our approach is general and can be extended easily to other protein systems. Although our modeled antibodies were not tested in vitro, we predict that these variants would have greatly increased breadth compared to the wild-type antibody. Computational design has been used successfully by protein engineers for many years to alter the physicochemical properties of proteins [1,2]. In the simplest case, protein design involves optimizing the amino acid sequence of a protein to accommodate a desired 3-D conformation. This approach has been extended to related tasks such as protein-protein interface design, de novo design of protein binding molecules, design of self-assembling protein nano-cages, etc. [3–6]. Each of these examples involves the straightforward application of design methodologies to a single, static protein conformation. However, there is a need to extend protein design to apply to several conformations simultaneously. These approaches, referred to as multistate design (MSD), can be used to modulate protein specificity, model protein flexibility, and engineer proteins to undergo conformational changes [7–13]. Several methods have been developed to enable computationally expensive multistate design [14,15]. However, these methods all suffer from large energetic barriers that limit sampling in sequence space, resulting in sub-optimal designs [14]. In addition, these methods are severely limited in scale by the size and number of states that can be included. To address these limitations, we have developed a method that integrates structural modeling with integer linear programming to enable a fast global search through large ensembles of target states. Our design algorithm, which we call BROAD (BReadth Optimization for Antibody Design) incorporates Rosetta-based structural modeling with integer linear programming to more easily traverse boundaries in the energy function (Fig 1). The experimental workflow involves generating a large training set of randomly mutated proteins, fitting a linear model (described below) to predict binding, and using integer linear programming to find an optimal antibody sequence balancing stability and binding with respect to a collection of target virus epitopes. We applied this method to the problem of designing broadly binding anti-HIV antibodies. We modeled anti-HIV antibody VRC23 [16] against a set of 180 diverse viral proteins, creating antibody variants that were mutated randomly in the paratope region. The viral panel used was derived from Chuang G-Y, et al [17]. Based on known binding patterns of VRC23 we calculated the predicted binding energy that corresponds to observable binding, and searched antibody space using integer linear programming to optimize stability of the unbound antibody while achieving predicted 100% binding breadth to the 180 target viral proteins. We then used a non-linear Support Vector Machine classifier, trained on the entire dataset produced by Rosetta, to identify top sequences. Finally, we entered the top scoring sequences back into Rosetta structural modeling to measure the predicted breadth of antibody variants. Our end goal is to design broadly binding and stable antibodies by searching the sequence space, i. e. , to optimize the amino acids at each binding position of the antibody. The key challenge for this approach is that an exhaustive search in the combinatorial sequence space is intractable. To address this issue, we first propose to learn sequence-based linear classification and regression models to predict binding and stability from data. Building on these models, we formulate an integer program to accomplish global search in the antibody sequence space. To generate our training set, we determined three contiguous stretches on the antibody that are in contact with the viral protein. These positions were determined to be residues 46–62, spanning FR2-CDR2-FR3; residues 71–74 in FR3; and residues 98-100b in CDR3 (S1 Fig). We then created randomly mutated antibody variants, modeled their binding poses using Rosetta, and used this data to train a binding classifier to predict Rosetta score and binding energy from amino acid composition. The binding classifier is based on the assumption that the amino acids at the binding positions of the antibody interact with those on the binding positions of the virus. In particular, this model assumes that binding between an antibody and a viral protein is determined by two factors: a) the individual amino acids in each binding position of the antibody and the virus respectively and b) the effects of the pairwise amino acid interactions between the antibody and the virus respectively. To capture these, we construct a sequence-based binary feature vector from the input antibody and virus pair, which explicitly represents the individual and pairwise amino acid contributions. Let the input antibody-virus pair represented as vectors of amino acids, be denoted by (a, v). Let b (a, v) denote the Rosetta predicted binding energy for (a, v) and let Φ (a, v) denote the binary binding decision. We chose a threshold θ such that Φ (a, v) = +1 if b (a, v) ≤ θ (i. e. , a and v bind) and Φ (a, v) = −1 otherwise. For evaluation of our approach, we choose the value of θ based on experimental neutralization data. This data is available as the experimental neutralization IC50 (in units of μg/ml) of VRC23 with the 180 virus sequences in the panel [17]. Lower values represent better neutralization potency and values that have ‘>50’ concentration represent a virus that is not neutralized by VRC23. Accordingly, VRC23 has a neutralization breadth of 63. 5% on this panel. We set θ = -28. 5 such that the VRC23 breadth of binding computed on the Rosetta generated data (sequences and the corresponding Rosetta binding scores) is consistent with the above experimental neutralization data. We learn the classifier Φ (a, v) as a linear Support Vector Machine (SVM) [18] using the binary feature set comprised of actual antibody and virus sequences along the corresponding binding sites, as well as all pairwise interactions of antibody and virus amino acids. The SVM classifier uses the Rosetta binding energy as the ground truth, and allows more efficient sampling by approximating the Rosetta score function by sequence alone. To optimize the L2 regularization parameter of the SVM, we performed 10-fold cross-validation on the full dataset, using 80% of the data for training and 20% for testing. Smaller parameter values enforce higher regularization and higher values lead to overfitting. The average prediction accuracy is shown in Fig 2A for different values of the L2 regularization parameter. We also plot the prediction error on the two classes: binders (+1) and non-binders (-1). The prediction accuracy is 67% on the test set using the optimized parameter (a random predictor would be at 50%). We observe that even if the prediction accuracy is relatively low, it provides reasonable signal within the subsequent breadth optimization step (discussed in the results section). Since the final decision is determined by solving the breadth optimizing integer linear program, our approach does not rely on a highly accurate classification model. In previous research [19], a similar model was introduced to predict ΔG values for interaction between PDZ domains and peptide ligands. The result was a 0. 69 correlation coefficient in 10-fold cross validation. This model can also be interpreted to identify the important binding position pairs that contribute significantly to the final prediction. We plot this interaction strength for each pairwise interaction in Fig 2C (please refer to the methods section for details). Next, we learned a linear regression model to predict the thermodynamic stability, using only the antibody amino acids as features. The prediction of thermodynamic stability is necessary to ensure that our designed antibodies can be expressed stably. To simplify the approach, we predicted the stability of the antibody-virus complex as a function of the antibody sequence only (note that we do not make this assumption during evaluation). Specifically, we constructed a binary feature vector restricted to amino acids in the antibody binding positions. Let s (a, v) denote the Rosetta stability for the pair (a, v). We learn a linear model Ψ (a) to predict s (a, v) for an antibody a (i. e. , independent of the virus). To measure the accuracy of prediction, we computed the correlation coefficient between the true scores and the predicted scores. Interestingly, our assumption that stability scores are only weakly dependent on the virus protein sequence is borne out: we found a correlation of 0. 85 between the predicted and actual stability energy score on the test set (Fig 2B). Given the classification and regression model learned from data, we formulate an integer linear program (ILP) to optimize the amino acids in the antibody sequence space to achieve both breadth and stability. The variables are the amino acids in the antibody binding positions. The objective function optimizes the predicted stability score (i. e. , minimizes Ψ (a) ). The constraints represent the condition that the designed antibody should bind to all the viruses in the panel, using binding predictions from Φ (a, v). We found that this problem was always feasible: there always existed some antibody sequence that could bind to all viral proteins based on our learned binding model. More generally, we can impose a minimal binding breadth criterion. This algorithm is outlined in S2 Fig. Armed with these tools, we used the following protocol to generate a collection of candidate antibodies to be evaluated using Rosetta. First, we took a random subsample of the full training data corresponding to 100 out of the 180 virus sequences. Using only this subsample, we trained the binding and stability models, Φ (a, v) and Ψ (a) respectively. We then solved the ILP described above to compute a stable, broadly-binding antibody sequence, considering only the 100 out of 180 selected virus sequences (that is, we only constrain the ILP to bind to these 100 virus proteins, rather than the full set of 180). We repeated this procedure 50 times, to obtain 50 candidate antibody sequences. To validate these optimized antibody candidates, we predicted binding and stability scores using a model trained on all the data. In case of stability prediction, we used a linear model as described above (since the model is reasonably accurate). For binding prediction however, we trained a non-linear (radial basis function kernel) SVM for improved prediction accuracy. Each of the 50 candidate antibodies were scored using these models trained on all data, in terms of predicted binding breadth and stability, and 10 best candidates were then chosen for Rosetta evaluation using the full panel of 180 virus proteins. This procedure is outlined in S3 Fig. After generating redesigned antibody sequences with predicted increases in breadth, we threaded these sequences onto the VRC23-gp120 complexes and subjected them to structural modeling to measure the change in predicted breadth. We refined the complexes using the Rosetta relax protocol–to test the accuracy of the Rosetta relaxed models, we compared the relaxed models to solved structures of gp120 viral variants and computed the root mean squared deviation (RMSD) over Cα atoms on gp120. We observed that the relax protocol recapitulates the gp120 conformations with an average RMSD of 2. 2 Å, whereas the pairwise RMSD between gp120 conformations, representing the intrinsic flexibility of these molecules, is 1. 8 Å (S1 Table). Considering that we substituted only residues at the binding site of the gp120 variants, and not the entire gp120 sequence, we consider that the variant gp120 conformations are recapitulated with sufficient accuracy for this experiment. As a control, we generated sequences using structure-based multistate design with the RECON method [14]. The RECON method uses Rosetta design combined with coordination between differing states to generate an antibody sequence with increased affinity for all target states. Using RECON to redesign antibody-antigen complexes has been benchmarked and been shown to generate germline-like, broadly binding antibodies [14]. We compared the 10 sequences created by BROAD to 10 sequences generated by RECON multistate design to compare the change in breadth to alternate approaches. We found that the BROAD method resulted in a significant increase in predicted breadth over the RECON multistate design method (Fig 3A). The BROAD-designed antibodies were able to achieve predicted breadth ranging from 86. 1–100% of viruses, whereas multistate designed antibodies reached a predicted breadth of 62. 8–85. 6% of viruses. Notably, both methods were able to increase predicted breadth from the starting value of 53. 3% for wild-type VRC23. This finding suggests that the wild-type VRC23 sequence is sub-optimal for breadth, which is supported by the observation that other known broadly neutralizing antibodies bind in a similar mode to VRC23 but with breadths exceeding 85% [20–23]. In addition, we observed that the BROAD method samples sequence space that is not sampled in multistate design (Fig 3B). We hypothesize that the BROAD method is able to cross energetic barriers that restrict sampling in traditional structure-based design methods, and is thereby able to generate antibodies with greater predicted breadth and lower energy. To support this hypothesis we analyzed the difference in score and binding energy for antibodies designed by BROAD and multistate design over the panel of viral proteins (Fig 4). BROAD was consistently able to generate lower energy antibody-antigen complexes, with a marked decrease in binding energy. This finding supports the hypothesis that BROAD is able to search sequences that are unavailable to multistate design, and that these new sequences have favorable score and binding energy. A frequent problem in computational protein design is false positives–that is, sequences that are predicted to be favorable according to the score function, but are unable to recapitulate that activity in vitro. The Rosetta score function uses many approximations of energetic terms to enable faster simulations, and these approximations can introduce inaccuracies [24,25]. To reduce the possibility that the redesigned VRC23 variants are scored favorably due to inaccuracies in the score function, we compared the designed residues introduced by BROAD to structural motifs of known broadly neutralizing antibodies (Fig 5). In several cases, the residues introduced by BROAD mimicked a known interaction of an existing antibody. For example, position 61 was mutated from proline in VRC23 to arginine (Fig 5, top left). The broadly neutralizing antibody VRC01 has an arginine that occupies similar space to the designed arginine [20]. This phenomenon can be observed for several different broadly neutralizing antibodies, such as VRC-CH31,3BNC117, and NIH45-46, all of which target the CD4 binding site, but at slightly different orientations [20–22,26]. We observed several examples of this type of recapitulation. Mutation Q62R on VRC23 placed an arginine residue to fill space that is occupied by a tyrosine on VRC-CH31 (Fig 5, top right) —this mutation fills a void at the interface to improve antibody-antigen packing. Mutation L73Y places an aromatic group overlapping with the position of a tyrosine in antibody 3BNC117, which also improves packing with the antigen (Fig 5, bottom left). Lastly, the D102E mutant on the CDRH3 places a carboxylic acid group in the same position as a glutamic acid on NIH45-46, improving electrostatic interactions with the antigen (Fig 5, bottom right). This observation is remarkable due to the fact that the antibody loops occupy different space, but redesigned residues are able to mimic the interactions of the broadly neutralizing antibody side chains. In addition, it is worthwhile to note that out of these four mutants that recapitulate known broad motifs, three were unobserved in the sequences sampled by multistate design (Fig 3B). As an additional comparison, we identified 1,041 sibling sequences of known broadly neutralizing antibody VRC01, that were isolated in a previous study [27]. These siblings presumably represent the sequence space accessible to VRC01, and are a good test case to compare how well our design algorithms are capturing natural sequence variation in a broad HIV antibody. Since these sequences have CDRH3 loops of different lengths we were not able to include the portion of the binding site corresponding to the CDRH3 loop–however we compared the rest of the binding site to the sequences seen in the VRC01 lineage (Fig 6). We observe that at several positions, BROAD samples sequences that are present in the VRC01 lineage but absent from MSD-sampled sequences (Fig 6, blue boxes). For example, at the third position in the binding site isoleucine is sampled at a high frequency in BROAD and VRC01 lineage sequences, but is never sampled by MSD (Fig 6). We highlight a total of five positions where BROAD outperforms MSD in sampling sequences that are seen in the VRC01 lineage. To quantify the sequence similarity we computed a sum of squared difference between the two matrices and normalized the values to 100% [14,28]. According to this metric the sequences sampled by BROAD are 79. 5% similar to those from the VRC01 lineage, whereas those sampled by MSD are only 76. 3% similar. We conclude that BROAD more accurately recapitulates motifs known in broadly neutralizing antibodies. In this paper we describe the development of a new protein design method that we call BROAD. This method uses structural modeling with Rosetta combined with integer linear programming optimization techniques to rapidly search through sequence space for broadly binding antibodies. We validated this method by computationally optimizing the amino acid sequence of the broadly neutralizing anti-HIV antibody VRC23. After modeling VRC23 variants in silico we were able to generate VRC23 variants with a predicted breadth of 100% over the simulated viral panel, compared to a predicted 53% breadth for the wild type antibody. This outcome represents a substantial step forward in protein design, and our methodologies can be used to address a wide variety of protein design problems in which traditional structural models are insufficient. Although we did not test antibody variants in vitro in this study, we predict that the computationally designed variants will have greater breadth against the HIV viral panel. However, we note several caveats with respect to experimental validation of these antibodies. Since this experiment was designed as a computational proof of principle, we modeled only the amino acids at the antibody binding interface of gp120, and not the entire gp120 sequence. This led to gp120 models with ~2 Å accuracy (S1 Table), which we consider sufficient for validating our design principles but not necessarily for experimental validation. Future directions in this work include optimizing protocols for gp120 homology modeling to reduce this discrepancy and enable experimental validation. A distinct advantage of the BROAD method is the ability to truly incorporate backbone movement into protein design. Many protein design methods have been developed that incorporate backbone ensembles to some degree [11,14,29,30]–however, this work typically involves either pre-generating large backbone ensembles, many of which may be redundant, or introducing backbone movement iteratively after steps of sequence design. In our approach, since we are relaxing the backbone of all mutants before fitting the sequence-based predictor, we were able to design sequences that may be slightly sub-optimal on the starting backbone coordinates, but can be highly favorable when a slight backbone relaxation is applied. This approach allows us to search sequence space that is not accessible to other methods, which are highly constrained to the initial backbone coordinates. We observed that the BROAD-generated sequences are not sampled by Rosetta design using the RECON method, and indeed are more favorable according to the Rosetta energy function. Therefore, we conclude that we are searching a “blind spot” in the sequence space that is missed by traditional design. This approach to research could be of great utility to the field of HIV immunology. A longstanding goal of the field is discovering broadly neutralizing antibodies as the basis of a rational structure-based vaccine strategy [31–33]. Much work has gone into redesigning existing antibodies to increase their breadth and potency [3,21]. However, HIV is known for its variability, and with this variability comes a difficulty in generating a single antibody with potent neutralization against all possible variants. The BROAD method addresses this problem by enabling rapid redesign of known antibodies against viral panels of arbitrary size. This technology can be used in the future as part of the antibody discovery and characterization process, by rapidly searching sequence space for variants for greater breadth. In addition, protein design also has been used on the reverse side of the vaccination problem, namely, to design a vaccine with high affinity for antibodies of interest [34–36]. We can foresee the application of the BROAD method to this problem as well, by optimizing immunogens for recognition of germline precursors of known broadly neutralizing antibodies. The VRC23-gp120 complex used for modeling was from the Protein DataBank (PDB ID: 4j6r). The structure was downloaded from the PDB (www. rcsb. org) and processed manually to remove water and non-protein residues. The CH1 and CL1 domains of the antibody structure were removed from the structure manually, and the structure was renumbered starting from residue 1. To select binding sites on the antibody and virus, we applied a distance cutoff of 4 Å from the opposing protein chain, where any residue with a heavy atom within 4 Å of a heavy atom on the opposing protein was considered to be at the binding site. Distance calculations were done using PyMol visualization software [37]. We expanded this binding site to several neighboring residues to include contiguous stretches of at least four residues to constitute a binding site. A total of 27 residues on the antibody were included in the binding site. We similarly determined a viral binding site to use for structural modeling. This site included 5 contiguous stretches that were determined to be in contact with VRC23 (32 positions total). These positions were 276–282; 365–371; 425–430; 455–462; and 473–476 (HXB2 numbering). To model gp120 variants, we performed a multiple sequence alignment using ClustalW [38] of the variant sequences with the gp120 in the crystal structure (Q23. 17), and substituted the corresponding amino acids at the binding site using Rosetta side chain optimization [24]. To generate a training set of structural models, we made random antibody substitutions in the previously defined binding site. Each antibody variant had five randomly selected amino acid mutations. Viral variants were taken from a set of 180 known HIV gp120 sequences [17]. We chose random combinations of antibody variants and viruses, as well as the native antibody sequence with all 180 viruses, for a total of 2200 antibody-virus pairs to serve as the training set. All antibody-virus pairs were subjected to an energy minimization via the Rosetta relax protocol, which involves iterative rounds of side chain repacking and backbone minimization with an increasing repulsive force [39]. 50 models of each antibody-virus pair were generated by Rosetta relax, and the lowest scoring model was used for further evaluation. The talaris2013 score function was used for all Rosetta simulations. Our data-driven sequence-based model to learn amino acid contributions to binding and stability is similar to the graphical model approach proposed in [19]. Let Na and Nv denote the number of binding positions on the antibody and the virus respectively. Let A={A1, A2…ANa} be a set of discrete variables representing the amino acids in the binding positions of the antibody. Each Ai takes values in the set of M = 20 amino acids. Similarly, let V={V1, V2…VNv} represent the variables for the virus-binding positions. The inputs for binding prediction are the antibody sequence a={a1, a2…aNa} and virus sequence v={v1, v2…vNv} where ai and vj are the amino acid values for the variables Ai and Vj. Amino acid contributions to binding can be modeled as a bipartite graph in which nodes for A and V represent the amino acids and the edges Ω ⊆ A × V represent the pairwise amino acid interactions. Each node ai and vj has associated weight vector xi and yj∈RM. The edge (i, j) between nodes ai and vj has an associated weight matrix Qij∈RM×M to represent the position specific contribution to binding for each amino acid pair, where qklum is the umth entry of matrix Qij. Consequently, given a and v, the binding score varies as the sum of individual amino acids and pairwise interaction effects. Given this setting, a and v are predicted to bind, i. e. , Φ (a, v) = +1 (b (a, v) ≤ θ), if ∑i=1Na∑j=1Mxijaij+∑i=1Nv∑j=1Myijvij+∑k=1Na∑l=1Nv∑u=1M∑m=1Makuqklumvlm+c≤0 (1) where c is the intercept term and aij and vij are binary indicator variables that take the value 1 if amino acid j is present at position i (∑j aij = 1, ∑j vij = 1 ∀ i). The qklum term represents Qkl (u, m). These weights can be learned efficiently using a linear support vector machine (SVM) classifier. The feature vector f consists of Na × M binary antibody features, Nv × M binary virus features and Na × Nv × M × M binary pairwise interaction features corresponding to x, y and Q respectively. Given a set of d training instance-label pairs (fi, li), i = 1 … d, li = {+1, −1}, a L2-regularized linear SVM generates a weight vector w by solving the following unconstrained optimization: minw⁡12wTw+λ∑i=1d (max⁡ (1−liwTfi, 0) ) 2, where λ > 0 is the L2 regularization parameter. Smaller λ values enforce higher regularization. The second term is the squared hinge loss function. The decision function is given by sign (wTf). We used the LIBLINEAR SVM implementation [40] to learn the classifier. Finally, the weights x, y and Q are retrieved from the combined weight vector w. On each training set of the viruses, we trained this classifier and saved the weights and the intercepts for future use in optimization. In our example, Na = 27 and Nv = 32. To tune the regularization parameter λ of SVM, we performed 10-fold cross-validation on the full dataset, using 80% of the data for training and 20% for testing. The average prediction accuracy is shown in Fig 2 for different values of the L2 regularization parameter λ. As expected, higher λ values lead to overfitting. We simultaneously plot the prediction error on the two classes: binders (+1) and non-binders (-1). We chose λ = 0. 001 for our experiments based on the bias-variance trade-off (corresponding to 33% test error). The above model can be interpreted to identify the important binding positions on the antibody and the virus side, i. e. , the pairs that contribute significantly to the final prediction. Specifically, we denote the Euclidean norm of the coefficient matrix of interactions Qij, for each position pair as the strength of interaction between those positions. We plot this interaction strength for each pairwise interaction in Fig 2C. The linear regression model Ψ (a) predicts the stability scores as a function of the antibody sequence features: Ψ (a) =∑i=1Na∑j=1Mxijsaij+cs (2) where xs∈RM is the weight vector in regression and cs is the intercept. Given a set of d training instance-score pairs (ai, si) i = 1 … d, (si = s (ai, vi), so there are multiple scores for the same antibody feature vector), the regression objective with l1 (sparse) regularization is given by: minxs⁡12d (‖ (xs) Tai+cs−si‖2) 2+α‖xs‖1, where the first term is the least squares penalty, α is the regularization parameter and ∥ xs ∥1 is the l1-norm of the weight vector. We used the Lasso implementation in scikit-learn [41] to learn this model. To measure the effectiveness of the prediction, we computed the correlation coefficient between the Rosetta calculated stability scores (in Rosetta energy units, or REU) and the scores predicted by regression. We performed a 10-fold cross validation experiment similar to linear classification, with 80% of the data for training and 20% for testing. Based on this parameter tuning, we chose α = 0. 01 with an average correlation of 0. 85 between predicted and actual stability energy score. Again, for each training set of viruses, we learn this model and save the weights and the intercept for the optimization in the next step. We leverage the weights in the binding and stability prediction models Φ (a, v) and Ψ (a) to formulate an ILP for optimization in the antibody sequence space. The objective is to minimize stability score. The constraints enforce the condition that the designed antibody should bind to each virus sequence in the training set. Finally, we add the constraint that the binary variables at each antibody binding position should sum to 1, i. e. , each position admits one amino acid. The ILP is given by the following: minimize∑k=1Na∑u=1M (xkus) akusubjectto∑k=1Na∑u=1M (∑l=1Nv∑m=1Mqumklvlmn+xku) aku+∑i=1Nv∑j=1Myijvijn+c≤−ϵ, ∀n∈1, …, t∑u=1Maku=1, ∀k, aku∈{0,1} where ϵ = 0. 0001 (which constrains that the antibody binds to all virus variants in the dataset, with a slight margin to ensure that binding is strictly below the 0 threshold). We used CPLEX version 12. 51 to solve the above ILP. We solve this optimization problem for each binding and stability model learned for data obtained from randomly chosen 100 virus variants (from the dataset in which all 180 are represented). Our final step is to take 50 antibodies generated using the integer program above from 50 random subsets of data, and choose the top 10 candidates to evaluate with Rosetta. This decision is based on a non-linear model of binding learned on the full dataset which includes all 180 viral variants, combined with a full-dataset linear model of stability. The top 10 most stable antibodies from all which are predicted to have 100% binding breadth are then chosen for evaluation. The linear model of stability is identical to what we had described above. For the non-linear model of binding we use a kernel support vector machine with the radial basis function (RBF) kernel. This model uses the same feature set as the linear model. The kernel function enables learning in a high-dimensional, implicit feature space without explicitly computing the coordinates of the data in that space. The RBF kernel of two feature vectors f and f′ is defined as: K (f, f′) =exp⁡ (−∥f−f′∥22σ2), where ∥ f – f′ ∥2 is the squared Euclidean distance between the two feature vectors, and σ is a free tunable parameter. Consequently, we have two free parameters to tune: the regularization parameter λ, and the RBF kernel parameter σ. Similar to the earlier set-up, we used 80% data for training and 20% for testing in a 10-fold cross validation experiment to tune these. We performed a grid-search over all pairwise combinations of σ and λ values in 10−2 to 102. The LIBSVM implementation in scikit-learn was used to train the RBF SVM. We chose the model with σ = 0. 01 and λ = 1 corresponding to the prediction accuracy of 68%. All learning and ILP experiments were performed on a 2. 4GHz hyper threaded 8-core Ubuntu Linux machine with 16 GB RAM. VRC23 was placed in complex with all 180 viruses and designed via RECON multistate design to increase predicted breadth across the panel. Models of viral variants were created as previously described, by substituting amino acids at the binding site. All VRC23-gp120 pairs were refined by Rosetta relax with constraints to the starting coordinates to prevent the backbone from making substantial movements. Constraints were placed on all Cα atoms with a standard deviation of 0. 5 Å. All residues at the binding site of VRC23 were included in design, for a total of 27 residues. The RECON protocol was run in parallel over 180 processors (manuscript describing parallelization in preparation), with four rounds of design and a ramping convergence constraint [14]. The binding sites on both the antibody and gp120 chain was subjected to backrub movements between rounds of design to increase sequence diversity [42]. A total of 100 designs were generated. Sequences generated by both BROAD and RECON methods were visualized using the WebLogo tool [43]. To compare sequences generated by BROAD optimization and RECON multistate design, we threaded the optimized antibody sequences over the unprocessed VRC23-gp120 complexes, and subjected these complexes to Rosetta relax to determine the score and binding energy of optimized antibodies vs. wild-type. 50 models were generated for each complex, and the lowest scoring model was used for evaluation. To compare native and optimized VRC23 sequences, we compared the total energy of the VRC23-gp120 complex as well as the binding energy (DDG), defined below: DDG=Ecomplex− (EAb+EAg) where EAb and EAg are the energies of the antibody and antigen alone, respectively. Structures of modeled VRC23-gp120 complexes were visualized using Chimera software [44]. VRC01 lineage sequences were derived from a previous study [27]. The 1,041 curated heavy chain sequences we used in this analysis are available in GenBank with accession numbers KP840719–KP841751. To compare sequence profiles we used a modified Sandelin-Wasserman similarity score, as described in [14,28]. Briefly, this score was calculated by computing the sum of squared difference for each amino acid frequency at each position, which was then subtracted from two and normalized to yield a percent similarity for each position and summed over all designed positions to give an overall similarity score.

In this article, we report a new approach for protein design, which combines traditional structural modeling with machine learning and integer programming. Using this method, we are able to design antibodies that are predicted to bind large panels of antigenically diverse HIV variants. The combination of methods from these fields allows us to surpass protein design limitations that have been seen up to this point. We predict that if we tested these modified antibodies against HIV variants they would have greater neutralization breadth than any antibodies seen to this point.

Abstract Introduction Results Discussion Materials and methods

biotechnology sequencing techniques medicine and health sciences immune physiology pathology and laboratory medicine engineering and technology pathogens synthetic biology immunology microbiology synthetic bioengineering retroviruses organisms immunodeficiency viruses viruses rna viruses artificial intelligence sequence motif analysis macromolecular design molecular biology techniques linear programming antibodies research and analysis methods bioengineering sequence analysis immune system proteins computer and information sciences lentivirus mathematical functions bioinformatics proteins medical microbiology hiv mathematical and statistical techniques support vector machines microbial pathogens molecular biology biochemistry physiology database and informatics methods viral pathogens biology and life sciences macromolecular engineering protein sequencing machine learning

2018

Integrating linear optimization with structural modeling to increase HIV neutralization breadth

Neural progenitors produce neurons whose identities can vary as a function of the time that specification occurs. Here, we describe the heterochronic specification of two photoreceptor (PhR) subtypes in the zebrafish pineal gland. We find that accelerating PhR specification by impairing Notch signaling favors the early fate at the expense of the later fate. Using in vivo lineage tracing, we show that most pineal PhRs are born from a fate-restricted progenitor. Furthermore, sister cells derived from the division of PhR-restricted progenitors activate the bone morphogenetic protein (BMP) signaling pathway at different times after division, and this heterochrony requires Notch activity. Finally, we demonstrate that PhR identity is established as a function of when the BMP pathway is activated. We propose a novel model in which division of a progenitor with restricted potential generates sister cells with distinct identities via a temporal asymmetry in the activation of a signaling pathway. The development of a functional nervous system requires the production of an amazing diversity of cell types. The precise identity of each neuron is acquired through a complex process referred to as neuronal subtype specification. Although different molecular mechanisms have been reported to control the specification of neuronal subtype identity, the activity of signaling pathways is at the heart of this process. Bone morphogenetic proteins (BMPs) have been widely linked with neuronal specification, such as in the mouse retina, where they promote the expression of M-opsin at the expense of S-opsin in photoreceptors (PhRs) [1,2]. Cell–cell communication involving the Notch pathway has also been implicated in neural specification [3]. In numerous cases, neuronal subtype specification is influenced by the concomitant activity of several signaling pathways, but the mechanisms underlying how these signals collaborate to establish distinct neural subtypes are only now beginning to be uncovered [4–7]. For instance, progenitors in the p2 domain of the spinal cord have the choice between the v2a and the v2b interneuron fate. In this system, BMP and Notch cooperate to promote the v2b fate [8–11], with Notch acting to promote activation of the BMP pathway in the future v2b cell [5]. These two pathways are also involved in neural subtype specification in the zebrafish pineal gland but with the roles being reversed [6]; BMP operates first to promote responsiveness to Notch signaling during the choice between PhR and projection neuron (PN) fates. Neuronal subtype specification can also be temporally guided, with different neuronal fates being produced over time from a common pool of progenitors. Indeed, it has been shown that neuronal progenitors in the vertebrate retina and spinal cord, as well as the mammalian cortex and olfactory bulb, generate distinct subtypes of neurons depending on when they are produced. In one model, the sequential production of distinct neuronal subtypes is the result of the evolution in the competence of neuronal progenitors [12]. In invertebrates, feed-forward cascades of temporal transcription factors have been described that control the evolution of competence within neural progenitors (see [13] for a review). Although this mechanism has apparently been conserved in vertebrates, so far only three factors have been identified that promote early or late fates; whereas forkhead box G1 (Foxg1) suppresses and IKAROS family zinc finger 1 (Ikzf1) promotes early cortical fates [14,15], Ikzf1 promotes early fates and castor zinc finger 1 (Casz1) promotes late fates in the retina [16,17]. In addition to transcription factors, it is expected that signaling pathways also contribute to temporally guided mechanisms of fate specification. For instance, neurospheres generated from cortical progenitors undergo temporal transitions that do not occur when cell–cell contact between progenitors is prevented [18]. In a different lineage, the signaling molecule transforming growth factor β2 (TGFβ2) operates as a temporal switch that promotes a late-born identity at the expense of an earlier one [19]. Another mechanism that has been proposed to influence neuronal fates is asymmetric division, which allows for the production of two different fates in sister cells derived from a common progenitor and largely involves asymmetric segregation of fate determinants during division. Asymmetric segregation of Notch pathway components has been extensively described in Drosophila neuronal lineages [3,20]. In vertebrates, however, asymmetric segregation of Notch interactors has been implicated in the decision to remain a progenitor or become a neuron [21,22] but not between adopting distinct neuronal fates. The decision to become a v2a or a v2b interneuron is by far the best-described vertebrate case of a Notch-dependent binary fate decision, but whether asymmetric segregation of Notch interactors plays a role in this specific instance is unclear. Indeed, although the v2a and v2b fates are produced from a common progenitor cell, the division producing these two neurons does not seem to occur at a specific angle, suggesting that a process of asymmetric segregation of fate determinants is unlikely [23]. Finally, the question of how asymmetric division is integrated with the activity of signaling pathways other than Notch and the evolution of competence in progenitors over time has yet to be thoroughly addressed. The zebrafish pineal is a neuroendocrine organ containing two main populations of neurons: PhRs and PNs. We have previously shown that BMP and Notch cooperate during the acquisition of a generic PhR identity [6,24]. Here, we explore the mechanism underlying the specification of different PhR subtypes in the pineal gland. Using the expression of different opsin genes, we have identified three distinct subpopulations of pineal PhR. Two PhR subpopulations, which express exorhodopsin (exorh) or the parietopsin (PT), are specified sequentially, with exorh+ cells appearing earlier than PT+ cells. Reduction of Notch activity accelerates PhR production and concomitantly shifts the fate of the PhR produced from late PT+ to early exorh+ identity. Gain of BMP activity, on the other hand, promotes ectopic PhR whose subtype identity depends on the time when the activation of BMP is triggered. Using time-lapse confocal microscopy, we show that PhR-generating progenitors predominantly produce sister cells with a different timing of BMP activation. In contrast, in a context in which Notch activity is reduced, BMP activation occurs either more frequently before the final division or with more symmetric timing in sister cells. Our results suggest a model in which division of a PhR-restricted progenitor generates two sister cells that activate the BMP signaling pathway at different times, resulting in the acquisition of either an “early” or “late” PhR subtype identity. Opsins are G-protein-coupled receptors that enable cells to sense a specific spectrum of wavelengths and intensities of light [25]. The expression of several opsins has been reported in the zebrafish pineal gland. For instance, the expression of exorh and red cone opsin (red) has been described in the developing and adult pineal gland using in situ hybridization [24,26], and the expression of PT, a green-sensitive photopigment that belongs to the so-called non-visual opsins, has been described using reverse transcription PCR (RT-PCR) [27,28]. To address whether these opsins are expressed in overlapping or restricted PhR populations, we mapped their expression relative to each other using double in situ hybridization. We found that their expression is largely restricted to distinct subpopulations of cells (Fig 1A, 1B, 1C, 1D, 1E, 1F, 1G, 1H and 1I), with only a few embryos showing one or two cells coexpressing exorh and PT or exorh and red (see quantification in Table 1); no coexpression was observed between PT and red. In parallel, we performed in situ hybridization against the endogenous opsins coupled with immunostaining for green fluorescent protein (GFP) /cyan fluorescent protein (CFP) in a previously described Tg (exorh: EGFP) ja1 transgene [29] or a transgenic reporter for the PT gene, Tg (-2. 2parietopsin: CFP), that we established for this study. In the Tg (exorh: EGFP) ja1 background, 92. 3% of the GFP+ cells express the endogenous exorh gene (S1A, S1B and S1C Fig); in addition, extensive overlap was also observed between CFP and the endogenous PT gene (S1P, S1Q and S1R Fig) in Tg (-2. 2parietopsin: CFP) transgenic embryos, suggesting that the two transgenes recapitulate endogenous expression. Finally, no overlap was detected between the two transgenes (Fig 1J, 1K and 1L, Table 1). Cross comparisons between the expression of the endogenous opsins and transgenes revealed similar levels of coexpression of exorh: EGFP with PT or red (S1D, S1E, S1F, S1G, S1H and S1I Fig) and PT: CFP with exorh or red (S1J, S1K, S1L, S1M, S1N and S1O Fig) as between the endogenous genes (Table 1). Since cells coexpressing exorh and PT or exorh and red are rare (<2 cells) and not found in all embryos, our interpretation is that these cells represent inappropriate specification events rather than hybrid cell fates. Two PhR subtypes have already been described in the embryonic pineal gland, as defined by the expression of rhodopsin (rhod) and Arrestin 3a (Arr3a) [30]; these populations were called rod and cone PhRs in reference to the morphology of cells expressing these genes in the retina. We found that rhod mRNA expression is largely restricted to exorh: EGFP+ cells in the Tg (exorh: EGFP) ja1 transgene (S2A, S2B and S2C Fig). Similarly, we used a Tg2PAC (opn1lw1: GFP, cxxc1: RFP) transgenic line [31], in which the red+ population of the pineal gland is labeled with red fluorescent protein (RFP), to establish that the red+ fate we describe corresponds to the Arr3a+ population described previously (S2D, S2E and S2F Fig; [30]). Given that the combined number of exorh, PT, and red+ cells accounts for the total number of PhR and that at 48–54 hours post fertilization (hpf) the average pineal gland contains 20 exo+ cells, 15 red+ cells, and 6 PT+ cells, we conclude that there are three subpopulations of PhR in the pineal gland, which at these stages corresponds to a composition of 48% exorh+, 36. 6% red+, and 14. 6% PT+ cells. As a previous study has provided insights into the specification of the red+/arr3a+ PhR fate [30], we chose to look more closely at the exorh+ and PT+ populations. These subpopulations do not appear to occupy a specific region of the pineal gland except that the PT+ population seems to be more peripheral (Fig 1D, 1E, 1F, 1G, 1H, 1I, 1J, 1K and 1L); PT+ cells are found in the center of the pineal gland at early stages (Fig 2D), suggesting that this peripheral pattern is reached secondarily. We noted, however, that whereas exorh expression can already be detected in the pineal at 24 hpf (Fig 2A), PT expression was not detected at 24 hpf and was detected only in 2 out of 10 embryos at 26 hpf (Fig 2B and 2D). Further quantification between 24 and 30 hpf confirmed that the exorh+ and PT+ pineal PhR populations are specified in a temporal sequence (Fig 2F). Analysis of the expression dynamics of exorh and PT suggests that these two PhR subtypes are specified at different stages during development, leading us to test the hypothesis that timing might play a role in their specification. Using a transgenic line, Tg (hsp70l: dnXla. Rbpj-MYC) vu21, in which a dominant negative form of the Notch effector recombination signal binding protein for immunoglobulin kappa J region (Rbpj) is overexpressed upon heat shock, we found that higher numbers of pineal cells express the pan-PhR marker anaat2: GFP [24] at 26,36, and 42 hpf (but not at 48 hpf) relative to control siblings when heat shock was performed at 14 hpf (Fig 3A, 3B and 3G), suggesting that PhR production is accelerated upon reduction of Notch activity; this relatively late heat shock elicits only a limited effect on the PN population (S3A Fig) compared to situations in which Notch signaling has been impaired from earlier stages [24]. We next addressed whether premature PhR production modifies the subtype of PhR produced and found an increase in the number of exorh+ cells and a reciprocal decrease in the number of PT+ cells (Fig 3C, 3D, 3E, 3F and 3H). Similar results were obtained using N-[N- (3,5-difluorophenacetyl) -L-alanyl]-S-phenylglycine t-butyl ester (DAPT; the gamma-secretase inhibitor) (S3B and S3C Fig), a pharmacological inhibitor of Notch activity. Taken together, these data suggest that advancing PhR production favors early PhR fate at the expense of late PhR fate in the pineal gland. To begin to address how a reduction of Notch activity could promote early production of PhR and the concomitant shift to early PhR fates, we analyzed the lineage relationships between PhR and PN in wild-type and Notch-impaired contexts. For this, we performed time-lapse confocal analyses on Tg (aanat2: GFP) y8 embryos from 15 hpf. Embryos were injected with synthetic mRNA encoding a Histone2B: RFP fusion protein at the one-cell stage to label all nuclei and permit backtracking aanat2: GFP+ cells identified at the end of the time-lapse acquisitions, and embryos were labeled for HuC/D expression at the end of the time-lapse series to identify cells that had adopted a PN fate. In an initial set of experiments, embryos were imaged for a duration of 20 hours, and based on the number of anaat2: GFP+ and HuC/D+ cells at the end of the movie, we estimate that the embryos reached 33–35 hpf at the end of the acquisition. Our results show that PhRs were born either from divisions generating 2 PhRs or a PhR and a cell with no defined identity (which we will refer to as" ø cell" ) (Fig 4; S1 Movie, S4 Fig, S2 Movie); ø cells do not express the PN marker HuC/D and do not divide a second time during the time-lapse series (S4 Fig, S2 Movie). Similarly, we observed that PNs were born from divisions resulting in 2 PNs or a PN and an ø cell (S5 Fig). Given that PhR–PN divisions were never detected, we hypothesized that most pineal neurons originate from fate-restricted progenitors at their last division and that in the cases of the PhR–ø divisions, the ø cell represents a future PhR that has not yet differentiated. To confirm this, we imaged embryos until they reach a stage close to 46 hpf based on the number of anaat2: GFP+ and HuC/D+ cells. Here, we found an increase from 67. 8% to 92. 3% of PhR–PhR divisions compared to the shorter acquisitions, supporting the notion that ø sisters of PhRs that have not yet acquired their PhR identity at 33–35 hpf will ultimately become PhRs (Fig 4I and 4J); again, no PhR–PN divisions were detected. Our results thus suggest that the vast majority of PhRs are born from fate-restricted progenitors at their last division. To address whether impairing Notch signaling affects the lineage relationships described in wild-type embryos, we next performed similar lineage experiments in Tg (hsp70l: dnXla. Rbpj-MYC) vu21 embryos heat shocked at 14 hpf. We found that lineage outcomes were comparable to those in wild-type siblings, with PhR being generated from a PhR fate-restricted progenitor; the percentage of divisions generating two aanat2: GFP+ PhR was 75% in Notch-impaired embryos compared to 67. 8% in the wild-type siblings (Fig 4I). We conclude that reducing Notch activity from 14 hpf onwards does not modify the lineage relationships between PhR and PN. Activation of the BMP pathway is both necessary and sufficient to promote a PhR fate in the pineal [6]. A mechanism to explain how reducing Notch activity affects the timing of PhR determination could, therefore, involve the premature activation of the BMP pathway. To test this, we took advantage of the Tg (BMPRE-AAV. Mlp: EGFP) mw29 transgenic line, in which a Bmp-responsive element (BRE) from the id1 locus has been placed upstream of GFP to create a transgene that behaves as a faithful reporter of BMP pathway activation, including in the pineal gland [32]. We counted the number of BMPRE-AAV. Mlp: EGFP+ (Bre+) cells in wild-type or Tg (hsp70l: dnXla. Rbpj-MYC) vu21 embryos at 22 hpf after heat shock at 14 hpf. Under these conditions, we observed an increase in BMPRE-AAV. Mlp: EGFP+ cells when the Notch pathway was inhibited (Fig 5A, 5B and 5G). A similar result was obtained using a second independent transgenic line Tg (BMPRE-AAV. Mlp: d2EGFP) mw30 (S6 Fig; [32]). Activation of the BMP pathway results in the phosphorylation of Smad1/5/8, which can be detected in the pineal anlagen from around 15 hpf and peaks at 18–20 hpf [6]. We tested the possibility that Notch might act to prevent premature phosphorylation of Smad in PhR progenitors. For this, we quantified the number of phosphorylated Smad1/5/8+ (P-Smad1/5/8+) cells in the pineal of wild-type and Tg (hsp70l: dnXla. Rbpj-MYC) vu21 embryos using the Tg (-1. 6flh: GAP-EGFP) u711 transgene to delineate the pineal domain. We did not observe premature Smad phosphorylation in PhR progenitors in Tg (hsp70l: dnXla. Rbpj-MYC) vu21 embryos heat shocked at 14 hpf (Fig 5C, 5D, 5E, 5F and 5H). Altogether, these results suggest that Notch activity affects BMP signaling in the pineal downstream of Smad1/5/8 phosphorylation and upstream of the BMP reporter Tg (BMPRE-AAV. Mlp: EGFP) mw29. To explore the dynamic nature of the effect of impairing Notch, we performed time-lapse analysis using the Tg (BMPRE-AAV. Mlp: EGFP) mw29 transgene in a manner similar to that described above in the Tg (aanat2: GFP) y8 background. As before, we observed two types of divisions (Fig 6A): those generating two BMPRE-AAV. Mlp: EGFP+ cells (2 Bre+: 80. 4%; n = 46 divisions) and those generating one BMPRE-AAV. Mlp: EGFP+ cell and one BMPRE-AAV. Mlp: EGFP- cell at the end of the acquisitions (Bre+/ø; 19. 6%; n = 46 divisions). In contrast to the situation found for Tg (aanat2: GFP) y8, however, expression of the Tg (BMPRE-AAV. Mlp: EGFP) mw29 transgene was observed either before or after the final division. To analyze the dynamics of BMP pathway activation in pineal PhR lineages, we compared a set of temporal variables for divisions generating two BMPRE-AAV. Mlp: EGFP+ cells between wild-type and Notch-impaired conditions (Fig 6B). To generate the variables for comparison, we first set the time of final division as zero. Next, we defined the time at which the first daughter cell expresses the Tg (BMPRE-AAV. Mlp: EGFP) mw29 transgene as t1 and the time of activation in the second daughter cell as t2. Finally, the asynchrony in BMP pathway activation between the two daughter cells was calculated (Δt = t2 − t1). In a wild-type context, we observed two different types of divisions generating a pair of BMPRE-AAV. Mlp: EGFP+ cells: those in which the expression of the transgene occurs before division (t1 = t2 = 0) and asymmetric divisions (Δt > 0), with the latter being far more numerous (Fig 6C, 6D, 6E, 6F, 6G, 6H, 6I, 6J, 6K, 6L and 6M; S3 Movie). In Tg (hsp70l: dnXla. Rbpj-MYC) vu21 embryos heat shocked at 14 hpf, the final outcome of divisions was globally similar to that observed in a wild-type context, with 81% of divisions generating two BMPRE-AAV. Mlp: EGFP+ cells and 19% divisions generating one BMPRE-AAV. Mlp: EGFP+ and one BMPRE-AAV. Mlp: EGFP- cell at the end of the acquisition (n = 31 divisions, Fig 6A). Unlike the wild-type situation, however, impairing Notch activity led to more symmetric divisions. In particular, we observed divisions in which the expression of the transgene occurs synchronously after division (Δt = 0, t1 > 0), a case never observed in wild-type controls. Furthermore, more cases of expression of the Tg (BMPRE-AAV. Mlp: EGFP) mw29 transgene before division were detected in Notch-impaired than in wild-type embryos (Fig 6C, 6N, 6O, 6P, 6Q, 6R, 6S, 6T, 6U, 6V and 6W; S4 Movie). These results suggest that pineal progenitors respond earlier to BMP signaling when Notch signaling is compromised, and the effect of Notch in this context appears to occur downstream of Smad phosphorylation. Reducing Notch pathway activity simultaneously accelerates the response to BMP signaling and favors early PhR subtype identity. To address a potential causal link between these two observations, we tested whether the timing of BMP activity controls the subtype of PhRs formed. Forced activation of the BMP pathway in the Tg (hsp70: bmp2b) fr13 transgenic line induces ectopic PhR production [6]. We quantified the populations of each PhR subtype obtained after induction of the Tg (hsp70: bmp2b) fr13 transgene at stages between 10 and 21 hpf (Fig 7A). We detected an increase of the exorh+ fate when the heat shock was performed relatively early (up to 18 hpf, Fig 7B, 7C, 7D and 7H). In contrast, only upon relative late heat shock (16–21 hpf) were the numbers of PT+ PhR increased (Fig 7E, 7G and 7I); activating BMP signaling at 10 hpf reduced the size of the PT population (Fig 7E, 7F and 7I). Together, these results suggest that activating BMP signaling at different stages produces different fates. We propose that the inhibition of the PT+ fate upon early BMP activation reflects the premature determination of exorh+ PhRs, with the consequence that the PhR progenitor pool is depleted at the time when the PT fate should be specified. Our analysis of BMPRE-AAV. Mlp: EGFP+ cells in a wild-type context suggests that in nearly 90% of cases, the divisions that generate these cells are asymmetric in the sense that the activation of BMP activity does not occur simultaneously in sister cells. The timing of BMP pathway activation in these divisions becomes more symmetric in a context in which Notch activity is reduced. Indeed, we observe that reduction of Notch activity increases the percentage of cells that activate BMP prior to division, suggesting that there is an inhibition of BMP activity by Notch at the level of the PhR progenitor. In this regard, the approximately 11% of divisions generating sister cells that simultaneously begin expressing GFP after mitosis in Tg (hsp70l: dnXla. Rbpj-MYC) embryos could represent cells in which gfp is transcribed already in the progenitor but GFP protein is not yet detected. We propose that an important prerequisite for temporally asymmetric divisions is the prevention of a premature fate decision within the progenitor. In invertebrates, sister cells in neuronal lineages often communicate via Notch to acquire their fates [3,20]. At this point, it is unclear whether Notch communication is occurring preferentially between sister cells in the pineal gland rather than between neighbors of unrelated lineage. Results presented here suggest an intricate interplay between Notch and BMP activities while also expanding the roles for BMP signaling during development of the pineal gland. The pathway is first required to define the dorsoventral position of the pineal anlage [33]. In a second step, it is required for the proliferation of pineal progenitors [6]. These two BMP activities do not seem to be controlled by Notch. BMP and Notch are then required to specify a generic PhR identity and, subsequently, specific PhR subtypes. The pleiotropic effects of BMP signaling on pineal development considerably complicate the identification of BMP targets relevant for each of these processes. Indeed, although classical BMP targets such as id1 and msxb/c/e are expressed in the pineal gland [32,34], it is unclear whether they underlie specific roles for BMP signaling. Our results have led us to propose the following model (Fig 8). In a wild-type context, Smad1/5/8 is phosphorylated in fate-restricted PhR progenitors but at a level that does not translate into the activation of BMP target genes, because of inhibition exerted by Notch activity. Notch pathway activity thus restrains the progenitor from adopting a PhR fate. After division, this inhibition is progressively released, although not synchronously, and this allows for the production of exorh+ and PT+ PhR, whose fates are dependent on the timing at which they activate the BMP pathway. How PhR precursors evolve from generating only exorh+ cells early to also generating PT+ fates later remains an open question. Future studies concerning how different sets of BMP target genes evolve over time should shed light on this. We have previously shown that during the PN/PhR fate decision, BMP is required for proper activation of Notch pathway targets, which in turn inhibit the PN fate in the future PhR [6]. The results presented here suggest a previously unanticipated complexity in Notch/BMP cross talk. Indeed, how can we reconcile that BMP signaling is required to activate Notch targets during the PN/PhR fate choice but that Notch pathway activation is required to prevent premature BMP activation during PhR subtype specification? One hypothesis relies on the existence of complexes containing the intracellular domain of Notch (NICD) /Rbpj and P-Smad1/5/8. Our analysis of P-Smad1/5/8 in wild-type and Notch-impaired embryos suggests that Notch inhibits the response to BMP downstream of Smad phosphorylation. The published BMP response element used in this study does not contain Rbpj binding sites [32]. This suggests that the effect of Rbpj on the Tg (BMPRE-AAV. Mlp: EGFP) mw29 transgene is either indirect or involves trapping of P-Smad1/5/8 species in inactive complexes. It has been shown that NICD coprecipitates with Smad1 in the presence of the coactivators P300 and the p300/CBP associated factor (P/CAF) [35]. These interactions have been proposed to reinforce the activation of Notch target genes in neural cells. Similarly, complexes containing NICD and Smad1/5/8 have been detected in cerebrovascular endothelial cells, where they are proposed to activate transcription of N-cadherin via an Rbpj binding site [36]. We propose that NICD/Rbpj/P-Smad1/5/8 complexes in the pineal participate both in the transcription of Notch target genes, as previously suggested [6], but also prevent the activation of BMP targets through a squelching mechanism. In this model, simultaneous activation of Notch and BMP receptors would lead to the formation of NICD/Rbpj/P-Smad1/5/8 complexes, but these complexes would first go to Notch targets, perhaps because of the higher affinity of Rbpj for its target sequences in the genome. The presence of Smad1/5/8 in these complexes would help transactivation of Notch targets, as previously described [35,36]. In a second step, either owing to a reduction in the level of ligands available to bind the Notch receptor or to a Notch-inhibiting signal, Notch activation would progressively decrease while activation of the BMP receptor would be maintained, and this would permit the activation of BMP target genes. Finally, whereas, to our knowledge, our study provides the first evidence that temporal control of BMP activity is crucial for specification of postmitotic neurons, the importance of a proper timing of BMP activity for specification of progenitor pools has been previously demonstrated in the dorsal spinal cord. In this case, rather than the timing of onset of BMP activity, duration of the exposure to BMP ligands seems the most important variable [37]. It is at present unclear whether the duration of exposure to BMP activity also plays a role in the specification of pineal PhRs. Examples of vertebrate neuronal lineages in which loss of Notch activity promotes early fates at the expense of late ones are already known [38–43]. In contrast, the mechanisms behind these effects of Notch are unclear. Notch could simply act to slow down neurogenesis and thus indirectly prevent temporal transitions. Alternatively, the pathway could play a more active role in triggering such transitions by promoting switches in the expression of temporal transcription factors in a manner analogous to what has been proposed for TGFβ2 in the hindbrain [19] or by regulating the competence of progenitors to respond to specific signaling pathways whose activity is interpreted differently over time, as is the case for pineal PhR subtype specification. Neurons in the left and right habenular nuclei of zebrafish develop with a temporal asymmetry in identity [38]. In this system, Notch has been proposed to promote identity through a general effect on the timing of neurogenesis [38]. Wnt activity was also recently shown to be necessary for the acquisition of right-sided neuronal phenotypes in the zebrafish habenulae [44]. Thus, an alternative hypothesis would be that Notch acts more directly to limit the competence to respond to Wnt activity. Along the same line, as Notch has been shown to facilitate Sonic hedgehog signaling during the specification of neural fates in the ventral spinal cord [4], it would be interesting to address whether a similar cross talk operates during the temporal neural to glial fate switch occurring within some of these ventrally specified progenitors, as such a switch has been shown to depend on a late burst of Sonic hedgehog activity [45]. Numerous diseases that affect retinal PhRs and lead to blindness have been described. One promising area of research aimed at treating these conditions involves generating PhRs in vitro with the longer-term aim of developing cell replacement strategies [46]. A recent study on the induction of retina from human induced pluripotent stem cells (iPSCs) in culture suggests that inhibition of Notch activity accelerates the production of PhRs [47]. These observations led the author to envisage using pharmacological inhibitors of Notch activity to accelerate the production of these cells. A question that has not yet been answered in this system, however, is whether modifications in timing are concomitant with fate changes in the types of PhRs that are induced, as we describe in the present study. We propose that the zebrafish pineal gland provides a powerful model for understanding molecular mechanisms driving neuronal subtype specification and for addressing specific questions concerning the establishment of distinct PhR identities. All animals were handled in the CBI fish facility, which is certified by the French Ministry of Agriculture (approval number A3155510). The project was approved by the French Ministry of Teaching and Research (agreement number APAFIS#3653–2016011512005922), in accordance with the guidelines from the European directive on the protection of animals used for scientific purposes (2010/63/UE), French Decret 2013–118. Embryos were reared at 28. 5°C and staged according to standard protocols [48]. The Tg (exorh: EGFP) ja1 [29], Tg (hsp70l: dnXla. Rbpj-MYC) vu21 [49], Tg (aanat2: GFP) y8 [50], Tg (BMPRE-AAV. Mlp: EGFP) mw29 and Tg (BMPRE-AAV. Mlp: d2EGFP) mw30 [32], Tg (-1. 6flh: GAP-EGFP) u711 [51], and Tg (hsp70: bmp2b) fr13 [52] have been described previously. Conditions of heat shock were as follows: Tg (hsp70: bmp2b) fr13 30 minutes at 37°C and Tg (hsp70l: dnXla. Rbpj-MYC) vu21 30 minutes at 39. 5°C. Genotyping of Tg (hsp70l: dnXla. Rbpj-MYC) vu21 was performed either using immunohistochemistry against the Myc epitope tag or via a nested PCR with the following couples of oligos: 5′-GCCACTTTTGTCCCTGATGC-3′ 5′-CTTTTTACATGTGGACTGCC-3′ and then, 5′-CCTTCCAGGTTCAGCTGCTG-3′ 5′-CGGGCATTTACTTTATGTTGC-3′. Genotyping of Tg (hsp70: bmp2b) fr13 was performed as described in [6]. To generate an in vivo marker of PT-expressing PhRs in the pineal gland, we amplified a 2. 2-kb fragment of PT regulatory sequences immediately upstream of the ATG by PCR using the following oligos: 5′-CGACCTCGAGGTAGGCCTACATTAAGCGAT-3′ 5′-GCGCGGATCCGATGATTCGGAATGATCTTC-3′. The resulting fragment was subcloned into a pBS-I-SceI backbone upstream of the coding region of CFP. To generate the Tg (-2. 2parietopsin: CFP) transgenic line, this construct was coinjected with I-SceI meganuclease into freshly fertilized embryos following previously described protocols [53]. The presence of successfully inserted transgenes was assessed using PCR with the following oligonucleotides: 5′-GGACACGCTGAACTTGTGG-3′ 5′-GGTACTTGTTCAGATGGCTG-3′. For experiments in which numbers of cells were assessed in fixed material, we assessed statistical significance using either a t test, a Mann Whitney test, or a Kruskal-Wallis test with Dunn’s post hoc comparisons (Fig 7). Statistical tests and number of embryos used are stated in each figure and/or figure legend.

A major goal in the field of developmental neurobiology is to identify the mechanisms that underly the diversification of the subtypes of neurons that are needed for the function of the nervous system. When investigating these mechanisms, time is an often-overlooked variable. Here, we show that in the zebrafish pineal gland—a neuroendocrine organ containing mostly photoreceptors (PhRs) and projection neurons—different classes of PhRs appear in a temporal sequence. In this simple system, the decision to adopt a PhR fate is driven by the activation of the bone morphogenetic protein (BMP) signaling pathway. Following the final cell division of a PhR progenitor, the sister cells normally activate the BMP pathway at different times. When Notch signaling activity is abrogated, activation of the BMP pathway occurs earlier and synchronously, which in turn favors the development of early PhR fates at the expense of later fates. We propose a model in which preventing sister cells from activating a signaling pathway in a synchronous fashion after their final division allows diversification of cell fates.

Abstract Introduction Results Discussion Material and methods

medicine and health sciences pineal gland in situ hybridization molecular probe techniques cell cycle and cell division cell processes social sciences neuroscience notch signaling developmental biology molecular biology techniques embryos research and analysis methods embryology probe hybridization animal cells sensory receptors molecular biology signal transduction cellular neuroscience psychology cell biology anatomy neurons endocrine system photoreceptors biology and life sciences cellular types afferent neurons sensory perception cell signaling bmp signaling

2019

A Notch-mediated, temporal asymmetry in BMP pathway activation promotes photoreceptor subtype diversification

Homologous recombination (HR) is the principal mechanism of DNA repair acting during meiosis and is fundamental for the segregation of chromosomes and the increase of genetic diversity. Nevertheless, non-homologous end joining (NHEJ) mechanisms can also act during meiosis, mainly in response to exogenously-induced DNA damage in late stages of first meiotic prophase. In order to better understand the relationship between these two repair pathways, we studied the response to DNA damage during male mouse meiosis after gamma radiation. We clearly discerned two types of responses immediately after treatment. From leptotene to early pachytene, exogenous damage triggered the massive presence of γH2AX throughout the nucleus, which was associated with DNA repair mediated by HR components (DMC1 and RAD51). This early pathway finished with the sequential removal of DMC1 and RAD51 and was no longer inducible at mid pachytene. However, from mid-pachytene to diplotene, γH2AX appeared as large discrete foci. This late repair pattern was mediated initially by NHEJ, involving Ku70 and XRCC4, which were constitutively present, and 53BP1, which appeared at sites of damage soon after irradiation. Nevertheless, 24 hours after irradiation, a HR pathway involving RAD51 but not DMC1 mostly replaced NHEJ. Additionally, we observed the occurrence of synaptonemal complex bridges between bivalents, most likely representing chromosome translocation events that may involve DMC1, RAD51 or 53BP1. Our results reinforce the idea that the early “meiotic” repair pathway that acts by default at the beginning of meiosis is replaced from mid-pachytene onwards by a “somatic-like” repair pattern. This shift might be important to resolve DNA damage (either endogenous or exogenous) that could not be repaired by the early meiotic mechanisms, for instance those in the sex chromosomes, which lack a homologous chromosome to repair with. This transition represents another layer of functional changes that occur in meiotic cells during mid pachytene, in addition to epigenetic reprograming, reactivation of transcription, changes in the gene expression profile and acquisition of competence to proceed to metaphase. DNA damage response is one of the most critical processes for cell survival and proliferation. Of the different forms of DNA damage, double-strand breaks (DSBs) are by far the most harmful. DSBs can arise spontaneously as a consequence of exposure to physical and chemical agents or following replication errors. In somatic cells, two main mechanisms, non-homologous end joining (NHEJ) and homologous recombination (HR) operate to repair DSBs [1]. NHEJ is the most common mechanism, working in all phases of the cell cycle [2] and its action is apparently simple. Classical NHEJ relies on the recruitment of the Ku70/80 complex and other regulatory factors, such as 53BP1, to the site of breaks to prevent DNA resection. This is followed by the incorporation of DNA-PK and DNA ligase IV, which reseals the break with the help of accessory factors such as XRCC4 [3,4]. In recent years, in addition to the classical NHEJ, a variety of alternative end-joining pathways, which use additional biochemical components, have been uncovered [4,5]. Although NHEJ pathways are quite efficient, they are also error-prone as they do not discriminate whether the two rejoined ends were the correct ones and resection and/or exonuclease activity may have removed sequences from the broken ends. In contrast, HR uses an intact DNA molecule as a template for repair, ensuring high fidelity of repair. However, this mechanism only acts when a DNA copy, usually the sister chromatid, is available, which only happens during the S/G2 phases of the cell cycle. The critical difference of HR repair versus NHEJ is that DNA is resected around the break point [4], generating single-stranded DNA (ssDNA) fragments that are used to search for the template sequence. The ATM kinase and the MRN protein complex (comprised of MRE11, RAD50 and NBS1) function as damage sensors by recognizing DSBs [3,6, 7]. The MRN complex, together with other proteins (e. g. CtIP, BRCA1, BLM, EXO1, DNA2), then performs a 5' to 3' resection of DNA on either side of the break, which forms 3' -protruding ends of ssDNA [4]. The newly produced ssDNA is covered by RPA, which protects it from degradation [3]. Then, the ATR-ATRIP (Ataxia Telangiectasia and Rad3-related and ATR-Interacting Protein) complex binds directly to the RPA-coated ssDNA, thus localizing the kinase ATR to DSBs [6]. After DNA resection, the recombinase protein RAD51 replaces RPA and forms nucleoprotein filaments, allowing the ssDNA to invade the DNA double helix of the template DNA and further proceed with the repair of the DSB [4,8]. HR may operate in both somatic and meiotic cells. During meiosis, homologous chromosomes undergo a series of complex processes, including pairing and synapsis, recombination and segregation. Meiotic recombination is in essence a HR repair mechanism that ensures the proper segregation of chromosomes during the first meiotic division and increases genetic diversity [8,9]. Although the molecular mechanisms mediating HR in somatic and meiotic cells are similar, there are several differences. The first one is the way DSBs are produced. In somatic cells, DSBs are usually generated by spontaneous events while in meiosis hundreds of DSBs are endogenously induced by SPO11 endonuclease during the leptotene stage of the first meiotic prophase [9–11]. Template choice for DSB repair is another important difference between somatic and meiotic HR. Sister chromatid is the common choice for DSB repair in somatic cells. However, in meiosis HR is tightly regulated to favor recombination with the homologous chromosome, although the sister chromatic can still be used. Repair with the homologous chromosome promotes crossing-over formation, which ensures coordinated chromosomal disjunction at the first meiotic anaphase [11–14]. Finally, during meiosis a specific recombinase, DMC1, is expressed in addition to RAD51. The coordinated action of DMC1, RAD51 and other regulatory factors drives repair to favor non-sister chromatid donors during meiosis [15]. In meiosis, DNA contacts between homologous chromosomes can ultimately resolve as reciprocal or non-reciprocal recombination events, which lead to crossovers or gene conversion events, respectively. Although HR is the main DNA repair pathway acting during meiosis, NHEJ can also be used [16]. Components of the classical NHEJ pathway such as Ku70/80 and 53BP1 have been detected in mouse meiotic cells, both in the course of normal meiosis [17] and after the exogenous induction of DNA damage [16,18]. Not surprisingly, this mechanism seems to be triggered only in the late stages of first meiotic prophase. This may be a consequence of the upregulation of HR repair during the early stages of meiosis following the endogenous production of DSBs by SPO11 and the resection of DNA that is concomitant with SPO11 removal [19]. However, coexistence of HR and NHEJ is possible during the late stages of meiotic prophase to repair DNA damage that was induced by either endogenous or exogenous mechanisms. Radiation exposure experiments, for instance, have reported an increase of both 53BP1 and RAD51 levels in pachytene and diplotene spermatocytes [16,18,20,21]. As mentioned above, the key event for the choice between HR and NHEJ relies on the resection of DNA around the break [2]. The production of ssDNA overhangs hampers the action of NHEJ mechanisms, which require intact ends. Although the regulation of DNA resection at DSBs is not completely clear, a reciprocal regulation of factors promoting and inhibiting resection has been reported [22]. 53BP1, which plays several roles in the regulation of DNA repair [2], has been proposed to play a key function in inhibiting resection by hampering the loading of the CtIP-BRCA1 complex to the DNA, thus diverting repair to the NHEJ pathway [4,23,24]. CtIP-BRCA1, in turn, is thought to negatively regulate 53BP1 by inducing displacement of both 53BP1 and Ku70/80 from the break point and stimulating DNA resection by the MRN- (EXO1-DNA2-BLM) complex [2,25]. Interestingly, both 53BP1 and BRCA1 seem to rely on ATM kinase for phosphorylation, which is necessary for their function. Additional factors, such as the action of specific CDK-cyclin complexes and the epigenetic landscape around the break point also contribute to the regulation of DNA end resection [22]. In addition to the biochemical interactions described above, the morphological, temporal and epigenetic scenario in which DNA repair occurs during meiosis must be considered. Synapsis, the intimate association of homologues, is mediated by a highly specialized structure called the synaptonemal complex (SC) [26]. Assembly and disassembly of SC components during the first meiotic prophase is a tightly regulated process crucial for proper chromosome recombination and segregation [27], as evidenced by the number of synapsis mutants in which recombination is disturbed, and vice versa [28–32]. Furthermore, during first meiotic prophase, the complex regulation of transcription and chromatin modifications can influence the response to DNA damage [33–35]. Most conspicuously, histone H2AX is phosphorylated to give rise to γH2AX, which localizes throughout the chromatin during the leptotene stage in response to DSBs [36]. This contrasts with the pattern of γH2AX in somatic cells, where it usually forms small and discrete foci after DNA damage [37]. γH2AX is involved in recruiting many DNA repair factors [4,10,36,38,39] and in the transcriptional silencing that is characteristic of the beginning of meiosis and the sex chromosomes [34,36,40]. Notably, ATM, ATR and DNA-PK can all phosphorylate H2AX [4,41,42]; therefore, γH2AX is a marker of both the HR and NHEJ pathways. Upon DNA repair, γH2AX seems to be displaced from the chromatin and/or dephosphorylated by protein phosphatases [43–45]. To shed light on the complex relationships between HR and NHEJ repair mechanisms acting during meiosis, we assessed DNA repair responses during mammalian male meiosis after the exogenous production of DSBs. We irradiated mice with gamma rays and then analyzed the localization and dynamics of various markers of DNA repair response, including γH2AX, DMC1, RAD51,53BP1, Ku70 and XRCC4, at different times of recovery. We have uncovered two distinct epigenetic patterns in response to DNA damage in early and late prophase-I spermatocytes: a typical meiotic one and a somatic-like one acting at early and late stages, respectively. The transition to a somatic-like response during mid-pachytene coincides with the sequential cessation of the meiotic HR response at mid-pachytene and the consecutive activation of NHEJ and somatic-like HR repair mechanisms. In addition, we report the formation of chromosome bridges between non-homologous chromosomes associated with either HR or NHEJ markers. We firstly analyzed the distribution pattern of γH2AX (H2AX phosphorylated at serine 139) in response to DNA damage. Phosphorylation of this histone is one of the first cytological events detected after DNA damage and has been used extensively to localize DSBs in both somatic and meiotic cells [4,10,36,37,39]. Staging of spermatocytes during first meiotic prophase was made on the basis of the degree of chromosome synapsis between autosomes, the morphology of SC and the morphology of the sex chromosomes following SYCP3 immunolabeling, as previously characterized [34]. In control spermatocytes, γH2AX is first detectable at the beginning of leptotene, when short threads of SYCP3 mark the initial assembly of axial elements (AEs) along the chromosomes. At this early stage, only a few discrete γH2AX foci are observed scattered throughout the nucleus (Fig 1A). During mid to late leptotene, when AEs form longer filaments, γH2AX is broadly localized throughout most of the nucleus (Fig 1B). This broad nuclear distribution is maintained during early zygotene (Fig 1C), when AEs start to synapse. From mid-zygotene onwards, γH2AX signal decreases and, by the end of zygotene, is mainly associated with unsynapsed regions (Fig 1D). During pachytene, when homologous chromosomes are fully synapsed, γH2AX localizes almost exclusively on the sex chromosomes, which have extensive unsynapsed regions (Figs 1E, 2A and 2B). Nevertheless, large γH2AX foci are sometimes observed associated to the SCs of some autosomal bivalents (Fig 2B, arrowhead). These foci have been previously described [10,34] and interpreted as unrepaired DSBs that tend to disappear with pachytene progression or, alternatively, as regions of transcriptional silencing [46]. During diplotene, when homologues desynapse, γH2AX remains present only on the sex chromosomes (Fig 2C). In gamma-irradiated spermatocytes, visible changes in the pattern of γH2AX localization are observed one hour after irradiation. In early leptotene cells, γH2AX is seen throughout the nucleus, in contrast to the small scattered foci seen in controls, indicative of a massive broadly distributed DNA repair response (Fig 1F). This pattern is also observed in late leptotene, zygotene and early pachytene spermatocytes (Fig 1G–1J). Changes at late leptotene and zygotene stages are less evident as γH2AX is already broadly localized throughout the nucleus in control cells at these stages. This pattern indicates that cells at the beginning of meiosis up to early pachytene respond to the induction of DNA damage similarly. In contrast, the response of spermatocytes from mid-pachytene onwards is rather localized. Large γH2AX foci are observed emerging from the SCs (Fig 2D–2F). This kind of signals have been called large foci [10], flares [46] or eruptions [47] and their morphology resembles that found in control spermatocytes (Fig 2B) and somatic cells [37]. These results reveal the existence of morphological differences in the response to DNA damage between early and late meiotic prophase spermatocytes. Irradiated spermatocytes show a clear diminution of γH2AX in most stages 24 hours after treatment. Similar to control cells, early leptotene cells have a few scattered γH2AX foci (Fig 1K). Provided that meiotic progression is not greatly affected by irradiation, then cells should progress to further stages during the recovery time. In order to test this possibility, we followed the progression of meiosis in control and irradiated mice after incorporation of EdU. Comparison spermatocyte population progression at different time points (24 and 72 hours) show that the advance of meiosis is not affected after irradiation (S1 Fig and S1 Table), in agreement with previous reports [48]. Therefore, early leptotene cells at 24 hours post treatment could have been at pre-leptotene when irradiated (see S1 Fig for an estimation of the length of each meiotic stage). In late leptotene and early-mid zygotene spermatocytes, γH2AX is distributed throughout the nucleus, similar to control cells (Fig 1L and 1M). Likewise, the pattern of γH2AX at late zygotene and early pachytene is comparable to that of the controls (Fig 1N and 1O), in which γH2AX appears to label the unsynapsed chromosomal regions and some foci in a few chromosomes. These cells were likely irradiated at leptotene and zygotene stages, respectively, indicating that cells irradiated at early meiotic stages are able to achieve a control pattern corresponding to their stage 24 hours after irradiation. In contrast, cells from mid-pachytene to diplotene retain several foci associated with SCs (Fig 2G–2I). These differences are also observed 72 hours after irradiation (Figs 1P–1T and 2J–2L). In this case though, cells at mid-pachytene 72 hours after irradiation were at an earlier pachytene stage at the time of irradiation. These cells likely had widespread localization of γH2AX at an earlier stage in response to DNA damage but their γH2AX pattern changes as they progress, very much like under endogenous production of DSBs. The persistence of γH2AX foci, however, indicates incomplete DNA repair. Notably, leptotene cells were very scarce 72 hours after irradiation (see S1 Fig). Previous reports indicated that spermatogonia are particularly sensitive to radiation [18,20,49,50]. In order to confirm apoptosis of these cells, we performed a TUNEL assay on testicular sections (S2 Fig) and observed an increase of apoptosis in specific cell populations at different recovery times. Specifically, 24 hours post irradiation, a noticeable, but not massive, increase of apoptosis is observed in spermatogonia and prophase-I spermatocytes, while at 72 hours apoptosis is mainly observed in metaphase cells. This leads us to infer that irradiation may partially ablate spermatogonia population, but probably also interrupts the normal entrance of these cells in meiosis, which explains the scarcity of leptotene cells. The two patterns of response to DSBs, early and late, also appear to differ in terms of γH2AX removal. Spermatocytes irradiated at late pachytene or diplotene, or those that reach these stages during recovery, remove γH2AX more slowly than those irradiated at earlier stages. In order to ascertain the efficiency of DNA repair, we recorded the number of γH2AX foci from mid-pachytene to diplotene (S1 Table) and analyzed the progression of repair by recovery time (Fig 2M) and cell stage (Fig 2N). One hour after irradiation, the number of foci increases in the three stages. The ANOVA test showed no significant differences between stages at this time. However, at 24 hours, the number of foci returns to control levels in mid-pachytene spermatocytes. In contrast, late pachytene and diplotene spermatocytes still show an increased number of foci, which is maintained even 72 hours after treatment. These results support the idea that γH2AX removal is less efficient as cells progress to later stages of prophase-I and that the number of foci seems to reach a steady state with no significant reduction. One striking feature observed after irradiation is the formation of connections between non-homologous chromosomes, which can be visualized by SYCP3 immunostaining (Figs 1T, 2J and 3). Connections are observed at all post-treatment times (1,24 and 72 hours) and could be clearly identified in zygotene to diplotene spermatocytes. On the basis of their morphological appearance, we classified connections in three categories (Fig 3): 1) distal contacts, in which chromosomes interact end-to-end (Fig 3A and 3B); 2) interstitial contacts, in which a filament emerges from one bivalent and contacts one or more bivalents laterally (Fig 3C–3F) and 3) intrachromosomal contacts, in which the connection is observed within the same bivalent (Fig 3G–3I). In some cases, the SYCP3-positive filament of a bivalent seems to split into two with a thin filament, probably involving a single chromatid, providing the connection (Fig 3C). In other cases, the filament appears thicker (Fig 3D). Chromosome connections can be observed between autosomal bivalents, between autosomes and sex chromosomes or between sex chromosomes. The presence of these bridges is likely not an artifact of the spreading technique as they are also observed in squashed spermatocytes (Fig 3J). Furthermore, chromosome fragments and bridges are observed during anaphase- and telophase-I (Fig 3K and 3L), indicating that these connections may represent chromosomal translocations. While connections between bivalents can result in a non-homologous chromosomal translocation, bridges within bivalents can potentially link the two homologous chromosomes or different parts of the same chromosome. The presence of these chromosomal aberrations at metaphase-I, which are rarely detected in the control cells, might account for the increased apoptosis observed at this stage 24 and 72 hours after treatment (S2 Fig). In order to understand the dynamics of chromosome bridge formation, we quantified the number of cells showing at least one of these chromosomal connections during pachytene and diplotene (Fig 3M) (connections were more difficult to discern from chromosome tangles in earlier stages). No bridges were found in 316 control cells analyzed. However, after irradiation, the frequency of spermatocytes bearing bridges increases from 6. 48% at one hour to 9. 15% at 24 hours and 24. 13% at 72 hours, indicating a clear rise in the number of bridges with time. Regarding the distribution of bridges by stage and time, at 24 hours, most of the cells with bridges are at early pachytene; however, by 72 hours, the distribution is more uniform among stages. In order to investigate the action of HR mechanisms, we first examined the spatial and temporal localization pattern of DMC1, which is exclusively present in meiosis and acts together with RAD51 [51,52]. To compare DMC1 distribution with the γH2AX pattern just described, we performed triple immunostaining of SYCP3, DMC1 and γH2AX. In control spermatocytes, a few DMC1 foci are seen scattered throughout the nucleus at early leptotene (Fig 4A). These foci are not specifically associated with either the short SYCP3 fragments or the small γH2AX foci already present. The presence of DMC1 foci at the beginning of leptotene suggested that they might be responding to DSBs produced by a SPO11-independent mechanism. However, their absence in SPO11 null mutants (S3 Fig) rules out this possibility. During late leptotene (Fig 4B) and early zygotene (Fig 4C), many more DMC1 foci are observed. At late zygotene, the number of DMC1 foci decreases (Fig 4D). Some foci remain associated with autosomes but they are clearly more abundant on the unsynapsed AE of the X chromosome. During early pachytene (Fig 4E), fewer foci are visible. Although DMC1 and γH2AX are co-localized on some autosomes, in many cases, DMC1 and γH2AX foci are not associated with one another (see detail in Fig 4E). At mid-pachytene, the number of autosomal DMC1 foci still decreases, though the sex chromosomes still have a high number of foci (Fig 4F). DMC1 is no longer detectable at a cytological level after mid-pachytene. After irradiation, a notable increase in DMC1 foci is observed (Fig 4G–4Z). As in control mice, these foci appear associated with unsynapsed AEs during leptotene, with synapsed and unsynapsed regions during zygotene and with synapsed autosomes and the AE of the X chromosome from pachytene onwards. Interestingly, DMC1 is not detected beyond mid-pachytene, indicating that radiation exposure is not able to trigger the appearance of DMC1 foci after this stage. Similar to control cells, some co-localization of DMC1 and γH2AX is observed in irradiated pachytene cells (see details in Fig 4K and 4L). We also observed DMC1-positive filaments connecting different chromosomes (Fig 4N, 4W and 4Z). These filaments are mainly present at 24 and 72 hours after irradiation and likely represent the nucleoprotein filaments formed during the ssDNA invasion of the intact DNA copy. Although it is unclear whether these filaments join homologous or heterologous chromosomes at earlier stages (Fig 4N), by pachytene, heterologous associations are clearly observed. Indeed, some of these filaments appear to be associated with SYCP3 threads that bridge different bivalents (Fig 4W), suggesting a role for DMC1 in DNA repair between heterologous chromosomes under these experimental conditions. In order to analyze the dynamics of DNA repair associated with DMC1, we scored the number of foci in control and irradiated cells at different stages. On the basis of the morphological features of SC formation and the γH2AX localization pattern described above, we considered six different substages: early leptotene, mid-late leptotene, early-mid zygotene, late zygotene, early pachytene and mid-pachytene. We did not record the number of DMC1 foci in leptotene cells 72 hours post irradiation given the scarcity of this cell population and the occurrence of morphological abnormalities, as mentioned above. Our quantitative analysis revealed some interesting features (Fig 4A’–4F’, S4 Fig and S1 Table). First, the early leptotene cell population of control spermatocytes has a low number of DMC1 foci and very low standard deviation. As described above, this population is also characterized by a few small γH2AX foci. In contrast, mid-late leptotene cells show an increase in the number and standard deviation of DMC1 foci, in agreement with a previous report [53]. This stage is also associated with broad γH2AX labeling, as pointed above. Peak abundance of DMC1 foci occurs during early-mid zygotene and decreases thereafter. According to the ANOVA and Tukey' s multiple comparisons tests, differences between each stage and the next one are significant (S4 Fig), indicating that DMC1 distribution can be used to distinguish the cell populations of the six substages. Second, as expected, the number of foci increases one hour after irradiation in most phases. As in the control, peak abundance of DMC1 foci is observed in early-mid zygotene spermatocytes, and each stage differs significantly from the following one, excepting mid-late leptotene and early-mid zygotene. However, the number of DMC1 foci induced by irradiation differs greatly among the different meiotic stages. The increase of foci compared to control is on average 79,125,73,44,14 and 1 for each of the six substages, respectively (see S1 Table). This striking result indicates that the cell stages are not equally sensitive to irradiation or that DMC1 localization to DSBs may be differentially regulated at the different stages due to the availability of this protein or other DNA repair factors. Furthermore, in irradiated mid-pachytene spermatocytes, the number of DMC1 foci did not increase significantly regardless of recovery time, indicating that DMC1 is no longer inducible at this or later stages. These results can be easily discerned when data are grouped by cell stage (Fig 4A’–4F’). Third, after the increase of DMC1 foci immediately after irradiation, a slow diminution is observed with recovery time for most stages; however, most did not reach control levels even after 72 hours of recovery time (Fig 4A’–4F’). Nevertheless, we observed two main stage-specific features: 1) early leptotene cells show almost control levels 24 hours later. Moreover, while the number of DMC1 foci is quite variable one hour after treatment, 24 hours later, the range of foci narrows, similar to the control. This finding may reflect the presence of newly formed leptotene cells that had just entered meiosis. Unfortunately, we could not record the number of DMC1 foci in early leptotene spermatocytes 72 hours after irradiation owing to the scarcity of this stage; and 2) in early pachytene, the number of DMC1 foci does not decrease but maintains constant in time. Since irradiation does not greatly disrupt meiotic progression, cells irradiated at a particular stage would continue to advance through meiosis and be at later stages when observed 24 or 72 hours later. Therefore, we arranged the quantitative data following a putative duration of 24 hours for leptotene, zygotene and early pachytene [35,50,54] (S1 and S4 Figs). This means that a cell irradiated at early leptotene would be at early zygotene 24 hours later and at mid-pachytene 72 hours later, and so on (S4 Fig). Considering four initial cell populations (early leptotene, late leptotene, early zygotene and late zygotene), we observed that, one hour after irradiation, the number of DMC1 foci increases in all cases and, in most cases, decreases 24 and 72 hours later, indicating efficient DNA repair in all cell populations. Control levels of DMC1 foci are reached after 72 hours of recovery, since all these cell populations are expected to reach mid-pachytene within the 72-hour period following irradiation. This suggests that DNA repair has been successfully completed in all cells. Alternatively, DMC1 might have been released from chromosomes at mid-pachytene, regardless of whether repair had been completed or not. We then analyzed the distribution of RAD51, which acts with DMC1 in the HR pathway, in control and irradiated spermatocytes. In agreement with previous reports [51,52], we found that RAD51 has a similar, albeit not identical, distribution pattern as DMC1 during first meiotic prophase (Fig 5). During zygotene stage (Fig 5A, S5 and S6 Figs), RAD51 localizes to the AEs of chromosomes, with a peak number of foci observed mainly in early-mid zygotene, decreasing continuously thereafter. At early pachytene and later stages, RAD51 foci remain associated with some autosomal SCs but are mainly found on the unsynapsed AE of the X chromosome (Fig 5B and 5C). Most of these foci are not associated with γH2AX, which at this stage is restricted to the sex chromosomes and a few foci over the autosomes. RAD51 disappears during late pachytene (Fig 5D) and is absent at diplotene (Fig 5E). This pattern is very similar to that of DMC1; however, we observed that RAD51 remains associated with chromosomes up to later pachytene stages. In order to observe this more clearly, we performed double immunostaining for both proteins (S5 and S6 Figs). During early stages of prophase-I, the localization of both proteins is almost, but not completely, identical. We noticed that not all DMC1 foci are associated with RAD51 foci and vice versa. More importantly, these two proteins are removed from chromosomes sequentially since the number of RAD51 foci on both autosomes and sex chromosomes clearly exceeds that of DMC1 at the mid to late pachytene transition (S5E–S5E” Fig). Thus, while the recruitment of RAD51 and DMC1 can be simultaneous upon the production of DSBs at the beginning of meiosis, persistent DSBs at the last stages of repair may lose DMC1 but maintain RAD51, which may reflect a role in promoting inter-sister versus inter-homolog interactions for the repair of DSBs [15]. Additionally, this persistence could be explained by the late production of DNA damage, which would not incorporate DMC1. Similar to the results with DMC1, we observed an increase in the number of RAD51 foci after irradiation (Fig 5F–5T). Although we did not quantify the distribution of RAD51 at early meiotic stages, the broad co-localization of RAD51 and DMC1 up to mid-pachytene (S5 and S6 Figs) suggests that both proteins follow a very similar pattern, i. e. , peaking one hour after treatment then decreasing with recovery time. This pattern agrees with that reported in previous studies [18,20]. However, the localization patterns of these proteins are not identical. For instance, though RAD51-positive filaments bridging chromosomes are also observed (S6 Fig), they are thinner and scarcer than DMC1 ones. Strikingly, after irradiation, RAD51 is observed in late pachytene and diplotene spermatocytes (Fig 5I, 5J, 5N, 5O, 5S and 5T). Given that RAD51 is not observed at these stages in control spermatocytes, these foci must represent newly localized protein induced after irradiation. Indeed, this RAD51 population differs with the one observed at earlier stages. First, the signal strength of RAD51 on the sex chromosomes is similar to that of autosomes. Second, foci tend to be larger and sometimes irregularly shaped. Finally, while virtually all RAD51 foci observed at early stages (up to mid-pachytene) are associated with the AEs or SCs, during late prophase, a significant proportion of RAD51 is detached from the SCs (Fig 5J, 5N, 5O). Most of these foci co-localize with γH2AX one hour after irradiation, indicating they correspond to regions of DNA damage (Fig 5J and 5O). To examine the dynamics of this late-appearing population of RAD51, we scored the number of foci present in mid-pachytene to late diplotene spermatocytes (Fig 6A and 6B; S1 Table). This analysis uncovered some interesting features. First, one hour after irradiation, RAD51 increases significantly in all stages analyzed except mid-pachytene. This striking result parallels the behavior of DMC1, which is also not inducible at mid-pachytene, suggesting that HR repair can be compromised at this stage immediately after irradiation. The increase of RAD51 observed in late pachytene, early and late diplotene stages is quite similar. Second, after 24 hours of recovery, the number of RAD51 foci is significantly higher in all cell populations, though the increase is more pronounced in late pachytene cells and conspicuously lower in diplotene cells. After 72 hours of recovery, RAD51 levels remain high, though a slight decrease is observed in all stages, except mid-pachytene. The unexpected behavior of RAD51 during mid-late pachytene and diplotene stages suggests that the HR response to induction of exogenous DSBs initially may be absent or weak but increases with time, at least until 24 hours after irradiation. We were intrigued by the presence of RAD51 foci that were not associated with SCs. We analyzed the dynamics of RAD51 foci during diplotene (Fig 6C and S1 Table) and found that both SC-associated and non-associated RAD51 foci follow the same pattern, increasing one hour after irradiation, peaking 24 hours later and decreasing thereafter. We found that the number of foci associated with SCs is clearly higher in cells at early and late diplotene but that the proportion of non-associated RAD51 foci increases in late diplotene cells 24 and 72 hours after irradiation. In order to ascertain the action of the NHEJ repair mechanism, we studied the temporal and spatial localization of XRCC4 and Ku70 components of this pathway. We first examined the localization XRCC4 (Fig 7), which is a ligase-IV co-factor. No protein is observed at a cytological level during early meiotic stages in control spermatocytes (Fig 7A). At late pachytene, however, a weak signal appears throughout the nucleus (Fig 7B) and becomes more intense at diplotene (Fig 7C). At this stage, the signal appears slightly more intense over the sex chromosomes. In order to rule out the absence of XRCC4 labeling in early spermatocytes as an artifact of the spreading technique, we also immunostained testicular sections (S7 Fig). XRCC4 is absent in spermatogonia and basal spermatocytes of the seminiferous tubules and is only detectable in spermatocytes located in the middle of the epithelium, corresponding to late pachytene-diplotene cells. We also studied the localization of Ku70, which is involved in the protection of broken DNA ends. Immunostaining of this protein yielded nearly identical results to XRCC4 (S8 Fig). These results indicate that these components are present by default during the normal course of meiosis, in agreement with previous reports [16,17]. We observed a very similar pattern in irradiated spermatocytes: no signal is detected prior to late pachytene and, from this stage onwards, the proteins are distributed homogeneously throughout the nucleus (Fig 7D–7L; S8 Fig). No marked differences in the signal strengths of these proteins were observed between control and irradiated cells. Moreover, neither XRCC4 nor Ku70 accumulates at putative DSB sites after irradiation (e. g. in a pattern resembling that of γH2AX). Therefore, induction of DNA damage has little to no effect on the spatial and temporal localization of XRCC4 and Ku70, consistent with these proteins being present by default at these stages. We also analyzed the localization of 53BP1, which has a main role in protecting broken DNA ends from resection during NHEJ repair. In control cells, 53BP1 is absent during leptotene, zygotene and early pachytene (Fig 8A) but present by mid-pachytene (Fig 8B), accumulating over the chromatin of the sex chromosomes, in a similar way to γH2AX. 53BP1 signal is maintained during late pachytene (Fig 8C) and early diplotene (Fig 8D) but becomes weak by late diplotene. After irradiation, in addition to sex chromosomes, 53BP1 localizes to the autosomes from mid-pachytene up to the end of diplotene. One hour after treatment (Fig 8E–8H), a large number of irregularly shaped foci are observed on the autosomes, very similar to the γH2AX eruptions. Indeed, most 53BP1 foci on the autosomes co-localize with γH2AX, although unassociated foci of both proteins are also observed. The same pattern is found at both 24 (Fig 8I–8L) and 72 hours after treatment (Fig 8M–8P). We observed that some chromosomal bridges, which are frequent in cells after treatment, are associated with 53BP1 (Fig 8N), indicating the involvement of NHEJ repair pathway proteins in this type of chromosome interactions. The quantitative analysis of 53BP1 after treatment (Fig 8Q and 8R, S1 Table) shows the dramatic increase in the number of foci one hour after irradiation in mid- and late pachytene and early diplotene spermatocytes and its sharp decline 24 and 72 hours later. This pattern clearly contrasts with and seems antagonistic to that of RAD51, with NHEJ proteins acting as a fast response and HR proteins acting in two phases, weakly immediately after DNA damage and strongly 24 hours later. We also observed that late stages tend to have more 53BP1 foci, indicating stage-specific differences in the response. Nevertheless, by 72 hours after irradiation, all stages show control levels of 53BP1. Phosphorylation of histone H2AX is one of the first key events to occur in response to DNA induced damage [21]. Two types of responses can be clearly distinguished in prophase-I according to the cellular phase: a massive response, characterizing the early stages, and a more focused response from mid-pachytene to diplotene. This focused response is typically found in somatic cells [37] even though, under irradiation overexposure, both somatic and meiotic cells can show a pan-nuclear response [21,55]. However, in our case, all cells were exposed to the same dose of irradiation; therefore, the differences in response are not due to dosage-dependent effects. The origin of these differences could be related, in part, to changes in chromatin configuration and transcriptional activity, as previously suggested for somatic cells [56]. Highly dynamic replacement and modification of histones and proteins associated with chromatin are known to occur during prophase-I [34,35,57–59]. Mouse spermatocytes in early prophase-I are characterized by a widespread distribution of histone H3 monomethylated at lysine 4 and trimethylated at lysine 9, which are both related to chromatin compaction and transcriptional repression [34,35,60]. These modifications are lost or re-localized between early and mid-pachytene, concomitant with other relevant epigenetic changes, such as the incorporation of histone H1t, which is related to the competency of cells to proceed to chromatin condensation stages [57], and a general reactivation of transcriptional activity, which is accompanied by the acetylation of histone H3 and other associated factors [34,35,61,62]. Therefore, the epigenetic changes occurring in meiotic cells at this stage likely act as regulatory factors modulating the DNA damage response. Changes in chromosome organization may also play a role in the shift in the DNA damage response. In C. elegans, changes in both chromatin conformation and organization of the SC central element are proposed to be involved in the change in the DNA damage response during the mid to late pachytene transition [63,64]. Indeed, exogenous damage can lead to desynapsis of homologous chromosomes [64]. Although no dramatic remodeling of the SC occurs in mouse spermatocytes during this transition, the gradual lengthening of the SC during pachytene [65] is a feature that potentially resembles such reorganization and thus may change the framework in which DNA repair proteins function. An additional cause of this change may be related to the different kinases that promote H2AX phosphorylation. At least two rounds of H2AX phosphorylation dependent on two different kinases have been proposed to occur in mouse meiosis: the first during leptotene involving ATM and the second at the end of zygotene involving ATR [40,47]. Our efforts to corroborate this hypothesis by immunostaining for kinases, including ATM, ATR and DNA-PK, were unsuccessful. However, indirect proof can be inferred. In this sense, ATM kinase activity seems to produce an amplification loop in the phosphorylation of H2AX that extends up to several megabases beyond the DSBs [42], whereas the phosphorylation produced by ATR and DNA-PK entails a more focused response in which the signal is limited to areas close to DSBs [66]. Therefore, the two responses we observed with γH2AX may reflect a main role of ATM at the beginning of prophase-I and a higher activity of ATR and DNA-PK at later stages [47]. Interestingly, the response of somatic cells to irradiation, in which most DSBs are repaired by NHEJ [67,68], usually produces discrete foci of γH2AX in the nucleus [37,42], similar to those found in pachytene and diplotene spermatocytes. Therefore, it seems that early stages have a meiotic-specific γH2AX response, while late stages have a repair response more similar to somatic cells (Fig 9). Based on γH2AX removal, the early response seems to be more efficient as cells irradiated at early stages return to control levels 24 after treatment. In contrast, cells irradiated at later meiotic stages retain a number of γH2AX foci for the duration of recovery. This contradicts findings that, on the basis of the removal dynamics of several repair proteins, suggested repair of DSBs induced at early stages of meiosis is slower than those occurring at later stages [20]. Therefore, it is important to be cautious with these interpretations. We found that many of the γH2AX foci observed at the different stages and recovery times are not associated with DMC1, RAD51 or 53BP1. The persistence of γH2AX in late stages may not be completely related to a delay in the completion of DNA repair but instead to delayed dephosphorylation or turnover of the histone [37]. A similar persistence of γH2AX foci has been also reported after etoposide-induced damage [21]. On the other hand, a substantial number of DMC1, RAD51 or 53BP1 foci are associated with chromosomes long after γH2AX has been displaced, indicating the persistence of unrepaired events. Therefore, it seems that the production of exogenous DSBs challenges both early and late repair pathways during meiosis, resulting in an overall lower efficiency of meiotic repair of exogenous damage compared to somatic cells, as has been previously suggested [18,20]. Our analysis of DSB repair pathways clearly indicates that HR is preeminent or exclusive at early meiotic stages, up to mid-pachytene, and reveals interesting clues about the pattern of HR response under both normal and experimental situations. We identified a population of early leptotene cells that in control mice is characterized by a low number of γH2AX and DMC1 foci. Then, a burst of these proteins is detected in late leptotene. Although we cannot rule out the possibility that these two patterns are just the two extremes of a linear rise of DSBs during leptotene [53], it is also possible that they represent two different physiological stages. While early DSBs are clearly SPO11-dependent, the rate in which they arise is limited, probably owing to the action of a limited number of SPO11 complexes [53,69] or to restrictions imposed by associated factors. Some proteins that stimulate SPO11 activity, like IHO1, are associated with the AEs [70]. Therefore, DSB production could be limited in a chromosomal context in which AEs have not yet formed. Given this context, these DSBs would be able to only trigger a focus-limited (somatic-like?) γH2AX response. The extensive γH2AX labeling of this early leptotene population after irradiation indicates that these cells are competent to display a broad (meiotic) γH2AX reaction. Nonetheless, this bona fide meiotic response is only detected later in leptotene, once AEs have become more extended. This interpretation poses interesting questions about the transition from spermatogonia to meiosis, which includes other puzzling features like the premeiotic pairing of homologous chromosomes [71]. An intriguing issue arose when we compared the distribution of DMC1 and RAD51. We noticed that these two proteins tend to form mixed foci but that their localization patterns are not identical. Several studies reporting similar findings in budding yeast, plants and female mouse meiosis have suggested that these proteins occupy different positions along the nucleoprotein filaments [15,72,73], perhaps performing complementary functions. More strikingly, we found that the temporal pattern of DMC1 and RAD51 do not coincide, particularly in the transition from mid to late pachytene. Temporal displacement between DMC1 and RAD51 loading and unloading has been also observed in plant meiosis [74]. This result may provide insight on the last stages of meiotic DNA repair pattern, particularly on the sex chromosomes. Several studies have hypothesized that DSBs on the X and Y chromosomes do not have homologous templates which can be used for repair, except obviously the pseudoautosomal region, and that repair can only be accomplished with the sister chromatid [33,36,75,76]. DMC1 may play a key role in interhomolog bias [15], such that its persistence on the sex chromosomes may explain why unresolved DSBs remain on these chromosomes long after most breaks have been repaired on the autosomes. The removal of DMC1 from sex chromosomes, and autosomes, at the mid-late pachytene stage may be related to a relaxation of the interhomolog bias, allowing the repair with the sister chromatid. Irradiation clearly stimulates the increase in the number of DMC1 and RAD51 foci from leptotene up to early pachytene. We did not find any markers for NHEJ at these stages. Although we cannot rule out that alternative NHEJ pathways might be present, it seems that the early response is mostly mediated by HR mechanisms, consistent with the findings of other studies [17,18,20]. It is reasonable to assume that induction of additional breaks cells will simply use the repair machinery that is already present. Therefore, endogenous and exogenous DSBs can enter the same repair pathway. Consequences of this include, for instance, an increased number of chiasmata, as previously reported [20,77]. Nevertheless, this response is limited as prophase-I proceeds. The net increase of DMC1/RAD51 foci at each stage is lower as spermatocytes progress to more advanced stages. Most strikingly, a meiotic HR response to exogenous DNA damage is very weak or not detected at mid-pachytene, thus providing additional evidence of the functional shift of spermatocytes at this stage (Fig 9). This transition likely involves the cessation of expression of some meiotic-specific genes (like DMC1) and the initiation of a new gene expression profile [78–80]. This is interesting not only in terms of DNA repair but also in relation to the regulation of meiotic progression. Several studies have provided evidence of a pachytene checkpoint that monitors DNA repair, chromosome synapsis and other physiological processes such as sex chromosome inactivation [40,54,81–83]. Although the mechanisms that drive this checkpoint in mouse have not been completely elucidated, defective spermatocytes appear to be largely eliminated at a specific point of meiotic progression, identified as stage IV of the seminiferous epithelium in mouse, which most likely corresponds to the mid-pachytene stage [54,81,84]. Once cells have cleared this checkpoint, inactivation of these surveillance mechanisms would be necessary to allow the progression of spermatocytes to later stages, allowing for instance desynapsis of chromosomes during diplotene without triggering meiotic arrest or inactivation of desynapsing regions. Likewise, inactivation of the early meiotic DNA damage response would be necessary after passing the checkpoint at mid-pachytene, with new events of DNA damage that occur from this stage onwards being subject to new control mechanisms, as previously suggested [21]. These new mechanisms would rely on different checkpoints, as exemplified by the elimination of spermatocytes in mutant mice for late recombination proteins such as MLH1 or MLH3 at metaphase-I [85,86]. One may expect that, in the absence of a DMC1 response at mid-pachytene, RAD51 takes the role of driving HR repair at this stage. As discussed above, RAD51 remains associated with chromosomes after DMC1 has detached, and previous reports have indicated that RAD51 is inducible in late prophase-I after irradiation treatments [18,20,21]. Consistent with this, one hour after irradiation, we found a modest but significant increase in RAD51 from late pachytene onwards; however, the increase was not significant at mid-pachytene. Instead, we observed increased levels of 53BP1 at all stages from mid-pachytene to late diplotene, indicating a faster response of the NHEJ pathway at these stages. This contrasts with a previous study that reported the presence of 53BP1 only after longer periods of recovery [18]. Differences in methodological approaches used to determine 53BP1 localization might account for these discrepancies. The quick trigger of NHEJ in late prophase-I may be due to a change in the choice of the default mechanism for DSB repair (Fig 9). Somatic cells first attempt to use NHEJ to repair DSBs, even in the G2 phase of the cell cycle when a sister chromatid is available to carry out more reliable repair by HR [2,5, 87]. The choice of NHEJ as the default mechanism from late pachytene onwards is illustrated by the constitutive presence of Ku70 and XRCC4 in the nucleus and the location of 53BP1 on the sex chromosomes. Therefore, as soon as new exogenous or even endogenous DNA breaks appear, this mechanism would quickly respond. This proposal is plausible in terms of the biochemistry of DNA repair. Given that the choice between NHEJ or HR relies on the regulation of DNA resection around the break point [4], for which 53BP1 has an inhibitory role [23,24], it is clear that NHEJ must be a first option. Otherwise, once resection has been performed, repair by this mechanism would be no longer possible. Nevertheless, it is also clear that both mechanisms are acting at the same time, raising the possibility that NHEJ and HR proteins are competing for DNA repair, especially from late pachytene onwards. In any case, after 24 hours of recovery, few 53BP1 loci remain, which may be due to rapid repair by NHEJ [2], but also to competition with HR. In somatic cells, HR seems to be preeminent in regions of high transcriptional activity [56], which is indeed the case of late pachytene and diplotene cells during meiosis. The increased presence of RAD51 foci 24 hours after irradiation indicates that HR repair mechanisms prevail again at that time. The late HR response has many differences with the early one. The most relevant is that it only involves RAD51. In the transition to a somatic-like DNA damage response, DMC1 is clearly no longer inducible, likely related to the change in gene expression pattern during pachytene [78–80]. The activity of RAD51 alone means that some of the constraints introduced by DMC1 in relation to DNA repair, such as homologous bias [11–13,15], would be relaxed. Therefore, this late repair could favor repair with the sister chromatid, which would be advantageous at the diplotene stage as desynapsis of chromosomes potentially hinders repair with the homologous chromosome. The RAD51 foci present during the late HR response are larger. Although we do not have a clear explanation for the morphological change of RAD51 foci, differential organization of the repair machinery around the break point involving, for instance, the accumulation of several DSBs in each foci, or comprising the resection of longer DNA stretches, may account for this change, as previously suggested in C. elegans [63]. These foci are also correlated with the formation of smaller discrete γH2AX foci, which, as mentioned above, may be due to the action of ATR or DNA-PK over ATM in the phosphorylation of H2AX, leading to a different architecture of the repair foci [42,66]. Finally, the finding of RAD51 foci not associated with the AEs/SCs of chromosomes is an intriguing feature. Whether RAD51 foci are always associated with AEs/SCs has been a matter of long debate [88]. Current models propose that endogenous DSBs in early stages are produced either in the context of the AE or rapidly taken there by the action of regulatory factors, including MEI4, IHO1 and HORMAD1, among others [70,89]. This is probably provided by their ability to interact with SPO11 before or at the time of DSB production. We found that most DSBs induced after radiation also localize at the AEs/SCs, as revealed by the pattern of DMC1/RAD51 foci and as reported in a previous study [50]. Thus, at early meiotic stages, both endogenous and exogenously induced DSBs likely rely on similar mechanisms to be taken to the chromosomal axis. However, at late stages, the situation might be different. The presence of IHO1 and HORMAD1 has been reported in diplotene cells [70,90]; however, it remains unclear whether these proteins, or others required for DSB localization at the axes, are completely functional at these stages. Partial failure of this process might explain the fraction of DSBs located far from AEs/SCs. Likewise, the progressive loss of these proteins as prophase-I proceeds may account for the increased frequency of non-associated foci in late diplotene spermatocytes compared with previous stages. The appearance of chromosomal bridges involving SYCP3-positive filaments is an intriguing feature that poses a number of questions about the nature of DNA repair. The presence of chromosomal connections and fragments is commonly found in irradiation experiments [91,92]; however, they are usually observed in metaphase and do not involve SC connections. To our knowledge, our study is the first to report that interactions between non-homologous chromosomes may involve not only the DNA contacts but also the axial structures they are attached to. This contrasts with the normal interactions of endogenous DSBs, which do not involve the formation of connections between the AEs of homologous chromosomes. A more in-depth characterization of these connections is needed to better understand the organization of the SC around break points and their role in promoting, facilitating or stabilizing chromosomal links. Bridges appear soon after irradiation, indicating they are part of a very fast response, and at increased frequencies with recovery times. Several possible interpretations can be drawn from these results. Given that bridges are observed between non-homologous chromosomes, or between non-homologous sequences of the same chromosome (intrachromosomal junctions), one would expect that they correspond to DSBs repaired by NHEJ. However, the appearance of bridges at early stages such as zygotene and early pachytene, in which the main repair mechanism is HR, challenges this interpretation. Our results and previous reports [16,17] suggests that NHEJ does not operate at early stages; however, alternative NHEJ mechanisms may potentially be present. Involvement of the NHEJ repair pathway in the formation of chromosomal bridges is doubtless only from mid-pachytene onwards. The presence of chromosomal bridges at early stages may be a consequence of the action of the HR pathway over homologous regions. This idea may seem completely counterintuitive but would be supported by the observation of DMC1/RAD51 filaments bridging non-homologous chromosomes. For HR to efficiently start, a minimum length of perfect homology is needed, which in mammals is 200–250 base pairs [4]. For this reason, repair templates are usually the sister chromatid or the homologous chromosome. However, the multitude of repeated sequences in the genome could provide sufficient homology to induce repair by HR. In the normal course of meiosis, endogenous DSBs are prevented over repeated DNA sequences. Moreover, mismatch repair mechanisms are in place to avoid recombination between highly homologous sequences of non-homologous chromosomes [93]. In addition, the number of DSBs is tightly regulated to not exceed a certain number, thus allowing mismatch repair mechanisms to function effectively [94]. Proteins like MEI4 and IHO1 seem to be involved in this limitation [70,89]. However, the excess of DSBs produced by radiation may have a deregulatory effect on the control of repair mechanisms such that homology requirements may be bypassed, thus allowing repair between non-homologous chromosomes. Regardless of the mechanisms used to form bridges, the final output is the production of chromosome connections that likely lead to the occurrence of translocations and fragmentation. This may cause cells to be compromised in the faithful distribution of chromosomes during the first meiotic division. Indeed, a noticeable increase of apoptosis is observed in metaphase/anaphase cells 24 and 72 hours after irradiation. In conclusion, the results presented here provide new insights on the transition between different programs of DNA repair during meiosis that act in a stage-dependent manner. A switch in DNA damage repair responses during meiosis has been also reported in other animal models such as C. elegans [63,64] and Drosophila [95]. Strikingly, these transitions also occur at the mid or late pachytene stages, indicating they may represent a conserved feature of meiosis. However, since both Drosophila and C. elegans control synapsis and DNA repair differently than mammals, particularly as they lack a DMC1 orthologue, the regulation of this transition might be different. The evidence presented also offers new clues about the location and dynamics of DNA repair mechanisms during meiosis and raises new questions about the differential functions performed by DMC1 and RAD51. The late HR pathway very much resembles the somatic response, presenting focalized γH2AX and involving only RAD51. This somatic-like response likely acts to repair DSBs that were not properly repaired by the meiotic default pathway (e. g. those on the sex chromosomes) or the occasional DNA damage that occurs after the primary phase of meiotic repair has concluded until cells start to condense chromatin and prepare for cell division. At this point, re-triggering a complex repair mechanism leading to the production of crossovers (provided chiasmata formation has been properly accomplished) would not be necessary. A simpler response using NHEJ or somatic HR would be sufficient. As previously suggested [63], this shift could simply be contributing to the maintenance of genome integrity before spermatocytes are engaged in segregating chromosomes to daughter cells. For spermatocyte spreads, we used the procedure described by Peters and coworkers [96]. Seminiferous tubules were disaggregated with forceps in a petri dish and a cell suspension was collected in phosphate buffered saline (PBS: 137 mM NaCl, 2. 7 mM KCl, 10. 1 mM Na2HPO4,1. 7 mM KH2PO4, pH 7. 4). After tubule fragments settled to the bottom of the dish, the cell suspension was transferred to a tube and centrifuged. The pellet was then resuspended in 400 μl of 100 mM sucrose. Cells were spread onto a slide submerged in 1% formaldehyde in distilled water containing 50 mM Na2B4O7 and 0. 15% Triton X-100 and then left to dry for two hours. Slides were subsequently washed with 0. 04% Photo-Flo (Kodak) in distilled water and air-dried before being used for immunofluorescence or stored at -80°C. Spermatocyte squashes were prepared as previously described [97]. Seminiferous tubules were fixed for 10 minutes in 2% formaldehyde in PBS containing 0. 1% Triton X-100. Fragments of tubules were placed on a slide coated with 1 mg/ml poly-L-lysine (Sigma) with two drops of fixative. A coverslip was put on top of the tubules and the cells were released by gently pressing the coverslip with a pencil. Finally, tubules were squashed, the slide was frozen in liquid nitrogen and the coverslip removed with a blade. Slides were immediately placed in PBS for further use. Spread and squashed slides were rinsed three times for 5 min each in PBS and incubated overnight at room temperature with primary antibodies diluted in PBS. The following primary antibodies and dilutions were used: mouse monoclonal anti-SYCP3 (Abcam, 97672) at 1: 200; rabbit anti-SYCP3 (Abcam, 15093) at 1: 100; mouse monoclonal against histone H2AX phosphorylated at serine 139 (γ-H2AX) (Upstate, 05–636) at 1: 1000; rabbit anti-DMC1 (Santa Cruz, SC-22768) at 1: 50; rabbit anti-RAD51 (Santa Cruz SC-8349) at 1: 50; rabbit anti-53BP1 (Abcam 36823) at 1: 100; goat anti-XRCC4 (Santa Cruz, SC-8285) at 1: 100; goat anti-Ku70 (Santa Cruz, SC-1486) at 1: 50. After incubation, slides were rinsed in PBS three times for 5 minutes each and subsequently incubated with the appropriate secondary antibodies in a moist chamber at room temperature for 1 h. We used anti-rabbit, anti-mouse and anti-goat secondary antibodies raised in donkey and conjugated with either Alexa 350, Alexa 488, Alexa 594 (Invitrogen), DyLight 549 or DyLight 649 (Jackson ImmunoResearch). Slides were subsequently rinsed in PBS three times for 5 min each and mounted with Vectashield (Vector). For double detection of two antibodies raised in the same species, we used Fab secondary antibodies as previously described [98]. Observations were made on an Olympus BX61 microscope equipped with a motorized Z axis. Images were captured with an Olympus DP72 digital camera using the Cell-F software (Olympus, Hamburg, Germany) and processed using the public domain software ImageJ (National Institutes of Health, USA; http: //rsb. info. nih. gov/ij) and Adobe Photoshop 7. 0. Testicles were fixed in cold 1% formaldehyde in PBS for 6 hours and then dehydrated and embedded in paraffin. Transverse sections (7 μm) were cut and mounted onto slides. Slides were then deparaffinized and treated with 0. 1% sodium citrate buffer containing 0. 1% Triton-X 100 for 10 min at 37°C. Sections were subsequently processed for immunofluorescence as described above or for TUNEL (Roche) following manufacturer instructions. Slides were counterstained with DAPI and mounted with Vectashield. Mice were injected intraperitoneal with 100 μl 10 mM 5-ethynyl-2-deoxyuridine (EdU) (Thermo Fisher Scientific) diluted in PBS. Since first meiotic prophase time length is longer than 72 hours (the longest period of recovery we analyzed), it was necessary to irradiate mice at two different times after EdU injection in order to evaluate the interference of meiotic progression at different sub-stages. Thus, mice were irradiated with 5Gy of gamma radiation one or six days after EdU administration and sacrificed 0,24 and 72 hours after irradiation (see S1 Fig for a detailed description of the EdU administration, mice irradiation and sample collection). This allowed us to cover the most relevant periods under analysis: the transition from leptotene to mid-pachytene and from mid-pachytene to diplotene. Testicular samples were processed as described above and spermatocytes were labelled with anti-SYCP3. EdU was revealed using the Click-iT as recommended by the supplier. γ-H2AX, DMC1, RAD51 and 53BP1 foci were scored manually in three individuals per each recovery time and the results were compared between different cell stages and times after irradiation. Foci were counted in at least 15 cells for each protein and cell stage per individual. Data were analyzed using ANOVA and Tukey' s multiple comparison tests for individual comparisons between different stages. Chromosomal bridges, apoptotic cells and EdU labelled spermatocytes were scored in two individuals for each recovery time after irradiation. For chromosomal bridges, spermatocytes containing at least one bridge were considered as positive. At least 150 cell were analyzed per individual. For the TUNEL assay, 300 tubules (150 per individual) were analyzed for each treatment, recording the proportion of tubules with apoptotic cells and the total number of positive cells, which were classified as spermatogonia, prophase-I and division spermatocytes owing to their size, position on the seminiferous epithelium and chromosome condensation. For EdU labeling, at least 200 cells were scored per individual and the proportion of labeled cells at different meiotic stages was analyzed using Chi squared test. Statistics and graphics were made using GraphPad Prism 6 or Excel software.

DNA repair is critical for both somatic and meiotic cells. During meiosis, hundreds of DNA double strand breaks (DSBs) are introduced endogenously. To repair this damage, meiotic cells use a specialized version of the homologous recombination (HR) pathway that uses specific meiotic recombinases, such as DMC1, to promote repair with the homologous chromosome instead of the sister chromatid. This process is important to ensure chromosome segregation during meiosis and, as a side consequence, increases the genetic diversity of offspring. Nevertheless, under specific circumstances, meiotic cells can use other DNA repair mechanisms such as non-homologous end joining (NHEJ), which is error-prone. We investigated the response of mouse spermatocytes to increased DNA damage caused by gamma radiation, which is commonly used in cancer therapy. We found that the excess of DSBs produced by irradiation is processed by the meiotic HR recombination pathway in spermatocytes at the early stages of first meiotic prophase. However, this response is not inducible from the mid-pachytene stage onwards. From this point on, spermatocytes rely on a response that shares many features with that of somatic cells. In this response, the NHEJ pathway is first used to repair DNA damage but is subsequently replaced by a HR mechanism that does not use DMC1. Instead, it relies only on RAD51, which is known to function in both somatic and meiosis cells and, contrary to DMC1, has a preference for the sister chromatid. This switch from a meiotic to a somatic-like response is accompanied by a conspicuous change in the epigenetic response to DNA damage, reinforcing the idea that a functional transition occurs in meiotic cells during the mid-pachytene stage.

Abstract Introduction Results Discussion Materials and methods

meiosis homologous chromosomes spermatocytes cell cycle and cell division cell processes germ cells dna damage dna sperm sex chromosomes animal cells chromosome biology autosomes biochemistry cell biology nucleic acids genetics biology and life sciences cellular types dna repair non-homologous end joining chromosomes

2019

Transition from a meiotic to a somatic-like DNA damage response during the pachytene stage in mouse meiosis

Plasmodium knowlesi is now recognised as a leading cause of malaria in Malaysia. As humans come into increasing contact with the reservoir host (long-tailed macaques) as a consequence of deforestation, assessing the potential for a shift from zoonotic to sustained P. knowlesi transmission between humans is critical. A multi-host, multi-site transmission model was developed, taking into account the three areas (forest, farm, and village) where transmission is thought to occur. Latin hypercube sampling of model parameters was used to identify parameter sets consistent with possible prevalence in macaques and humans inferred from observed data. We then explore the consequences of increasing human-macaque contact in the farm, the likely impact of rapid treatment, and the use of long-lasting insecticide-treated nets (LLINs) in preventing wider spread of this emerging infection. Identified model parameters were consistent with transmission being sustained by the macaques with spill over infections into the human population and with high overall basic reproduction numbers (up to 2267). The extent to which macaques forage in the farms had a non-linear relationship with human infection prevalence, the highest prevalence occurring when macaques forage in the farms but return frequently to the forest where they experience higher contact with vectors and hence sustain transmission. Only one of 1,046 parameter sets was consistent with sustained human-to-human transmission in the absence of macaques, although with a low human reproduction number (R0H = 1. 04). Simulations showed LLINs and rapid treatment provide personal protection to humans with maximal estimated reductions in human prevalence of 42% and 95%, respectively. This model simulates conditions where P. knowlesi transmission may occur and the potential impact of control measures. Predictions suggest that conventional control measures are sufficient at reducing the risk of infection in humans, but they must be actively implemented if P. knowlesi is to be controlled. Despite advances in the control and treatment of malaria more than half the world' s population remain at risk of infection and disease. Of the estimated 216 million episodes of disease occurring worldwide in 2010,13% were estimated to occur in South-East Asia [1]. In 2004, large numbers of malaria cases previously diagnosed as Plasmodium malariae in the Malaysian Borneo were discovered to be due to the simian Plasmodium knowlesi malaria [2]. P. knowlesi is a zoonotic malaria of macaques transmitted by the Anopheles leucosphyrus group of mosquitoes in South East Asia, and is increasingly recognised as a human malaria as incidence among humans continues to increase [3], [4]. It is difficult to determine whether the increase in reported P. knowlesi cases is genuine or a product of previous misdiagnosis as P. malariae. However the significant increase in the total number and proportion of malaria patients aged 50 years and above, an age group over-represented among genuine P. knowlesi patients suggests that there has been a true increase in P. knowlesi cases, especially as this coincides with reduced transmission of P. falciparum and P. vivax [4]. Although only recently confirmed in Malaysian Borneo, there is further evidence that P. knowlesi is much more widespread than previously thought with sporadic cases reported in China, Thailand, Myanmar and other neighbouring countries [5]–[9]. As an emerging infection that could become a major public health threat, it is critical to understand the extent to which transmission is maintained by the simian host population and whether wider spread of infections outside the traditional forested areas is likely. To date, the identified human P. knowlesi cases are mostly reported from individuals who have a history of exposure through proximity or travel to forest environments [10], supporting the premise that P. knowlesi is primarily zoonotic, with incidental human infections when humans encroach on non-human primate habitats at the forest-fringe. In addition, although Anopheles latens (a member of the An. leucosphyrus group and one of the main vectors in Malaysia) will feed on both non-human primates (NHPs) and humans, it is primarily a forest-feeder, and macaques found in human settlements have a lower prevalence of infection compared to their wild counterparts [5]. P. knowlesi was not considered an important cause of human malaria in the 1960s when vectors were typically found in the primary forest which covered much of Malaysia. However, with the extensive population growth of the last decades, humans encroach on large expanses of natural P. knowlesi transmission causing further habitat disruption and destruction. In response, NHPs have moved towards the forest fringes and mosquito vectors are increasingly found in human habitats [11]. Therefore, the increasing overlap between macaque, human, and vector habitats may in part explain the recent rise in P. knowlesi cases as humans are increasingly exposed to the vector and host, together increasing the probability of successful cross-species transmission. As an increasing number of P. knowlesi cases are reported from traditionally malaria-free areas, and with the push to eliminate malaria by the end of 2015, it is crucial to be aware of zoonotic malarias which may undermine such efforts. As such there is an urgent need to investigate appropriate treatment and prevention strategies [4], [11]. Here we develop a mathematical model for the transmission cycle of P. knowlesi incorporating the human, macaque and vector hosts allowing for human-vector-human transmission which has been demonstrated under laboratory conditions [12]. Although human-vector-human transmission has yet to be definitively documented in the wild, autochthonous cases reported in the Philippines [13], and the familial clustering of cases reported from a wide age distribution in Sabah, Malaysia are suggestive of peri-domestic transmission and point toward potential human-mosquito-human transmission [14]. Using parameters derived from the literature we estimate the extent to which infection is sustained by the different host populations and hence the potential for a shift from zoonotic to sustained transmission in the human population if human-macaque contact increases as a consequence of deforestation. We then use the model to assess the likely impact of rapid treatment and the use of insecticide-treated bed nets in preventing wider spread of this emerging infection. We extended a previous multi-host model for P. knowlesi transmission, which incorporated transmission between macaques, mosquitoes and humans [15], and then extended this further by accounting for three characteristic geographical sites (forest (J), farm (F) and village (V) ) in which exposure to infection and transmission can occur. Here we define the forest as dense rainforest where macaques primarily reside, the farm area as an area on the forest-fringe that has been cleared for agricultural use where workers are present during the day, and the village as very small rural communities where humans live. A schematic of the model is shown in Figure 1 and full mathematical details are provided in the Supporting Information. Humans and macaques are assumed to move between locations (village-farm-forest and farm-forest respectively) whilst the vector population is stratified by each location. Each host (human, macaque, and vector) can be in one of two states – susceptible or infected – and hence we track the proportion of infected humans (IH), infected macaques (IM), and infected vectors in the forest (IVJ), farm (IVF) and village (IVV). A recovered compartment was excluded for macaques since infection is chronic, and for humans since data on immunity against P. knowlesi are not available. We assume that humans and macaques are immediately infected and infectious following a mosquito bite [16], [17]. Parasitaemia peaks at day 8 after infection in humans, and falls rapidly to low levels by day 13 after infection. Thus we assumed a recovery rate of 1/14 days for humans [17]. We allow a delay in the vector transition from susceptible to infected state of 10 days to represent the extrinsic incubation period of the parasite [18]. Humans become infected at a rate λH which depends on (i) the rate at which vectors blood feed (accounting for human and macaque population sizes, the biting preference of the vector, and a reduction in biting rates on the farm to account for the absence of both humans and macaques in the evening when mosquitoes are most active [19]); (ii) the probability of transmission from mosquito to human/macaque per infectious bite; and (iii) the proportion of vectors infected in each location. Similarly vectors become infected in each of the three locations (forest, farm and village) at rates λVJ, λVF and λVV respectively, which depend on their frequency of biting, the proportion of bites in each location taken on humans versus macaques and the prevalence of infection in humans or macaques. Since our model is not temporal, we have used the median vector biting rates from the literature [19]–[23]. Table 1 contains details of the parameters and values used and additional parameter values are given in the Supplementary Material (Table S1). To assess the potential for sustained transmission in the absence of the macaque population we also calculate the component of the basic reproduction number for a human-vector system, R0H. Full details are given in the Supporting Information. The total human population was fixed and distributed with 5%, 30% and 65% for the forest, farm, and village respectively. These percentages were chosen to reflect the proportion of time that an individual might spend in each location since our model does not explicitly include human movement. These values were chosen based on the Malaysian National Census and the average population density in corresponding areas [24]. Despite spending most of their time in the forest, macaques have been observed to encroach on farm land whilst foraging for food. The proportion of the total macaque population in daily contact with the farm was unknown. Furthermore there were very limited data available to inform the transmission probabilities between vectors and macaques, and vectors and humans, in addition to the infectious period among macaques. In order to account for uncertainty in the parameters describing transmission of P. knowlesi between humans, macaques, and mosquitoes, and the unknown duration of infection in macaques, we undertook a model validation step using Latin hypercube sampling to obtain sets of these unknown parameters that were consistent with the possible prevalence of infection in humans and that in macaques. Due to its zoonotic nature, infection prevalence of Plasmodium knowlesi in humans in South East Asia is very low with an estimated annual incidence of 1% (95% CI: 0. 4–1. 7%) in southern Vietnam [25], 0. 3% in Cambodia [26], and 0. 65% in Thailand. In contrast, macaque P. knowlesi infection prevalence in the wild is extremely high at over 90%. Vythilingam et al. , compared P. knowlesi infection among urban and forest macaques in Malaysia and found that, while urban macaques were infection free, forest macaques had a prevalence of 97% [27]. Tan et al. , also found a prevalence of 87% in Sarawak among long-tailed macaques [28]. Since there are still very limited data on the true burden of P. knowlesi infection in humans and given that P. knowlesi is now the leading cause of malaria in Malayisan Borneo accounting for 87% of malaria admissions in Sabah [29], we allowed the upper limit of human infections to be high to reflect the possible range of prevalence using molecular detection tools [30], [31]. We therefore chose target ranges of 0–5% prevalence in humans and 80–100% prevalence in macaques for model validation. 50,000 parameter sets were selected using Latin hypercube sampling, and only those sets that resulted in infection prevalence within the target ranges were retained. We considered the potential impact of two interventions on transmission – the provision of long lasting insecticidal nets and hammocks (LLINs, LLIHs) and more rapid treatment of human infections. To incorporate the former we adapted an approach previously described for models of P. falciparum transmission [32], [33]. Full mathematical details are given in the Supporting Information. In brief, the presence of a net reduces the biting rate on humans by providing direct protection to the individuals using a net; has a repellency effect which acts to increase the proportion of bites taken on other hosts and to increase the gonotrophic cycle length due to additional time spent searching for blood meals; and finally increases mosquito mortality due to the killing effect of the insecticide. Importantly, under this model, LLINs/LLIHs will affect the vector populations in the forest, farm and village differently, as there will be different numbers of humans sleeping under nets in each setting. We considered the impact that different levels of LLINs/LLIHs coverage and usage may have in reducing P. knowlesi infection in humans, the human reproductive number, and the basic reproduction number. LLIN/LLIH usage in the farm was set to 0 since there is no evidence of net usage in this area. We do however assume that insecticide-treated hammocks (LLIHs) can be used in the forest [34], [35]. As a baseline scenario coverage (defined as the proportion of individuals in the population who always sleep under a net) was set at 80% in both the village and the forest based on a study in peninsular Malaysia [19]. To explore the impact of rapid treatment (a recovery rate of 1/5 days as opposed to 1/14 days) we varied the coverage from 0–100% and assumed that this would result in clearance of the parasites and hence reduce the duration of infection in the human host. This would then prevent onward transmission among humans and have a knock-on effect on the basic reproduction number for human infections. Both human infection prevalence and the overall R0 depend on the proportion of macaques that spend time in the farm, with both these quantities reaching a peak when just over half of the macaque population are in the farm (Figure 2). When the majority of macaques remain in the forest, there is minimal overlap between areas where humans and macaques are active, and human infection prevalence stays low. As a greater proportion of the total macaque population are present in the farm, humans are increasingly infected. However when the ratio of macaques in the forest to farm reaches a certain threshold there is a switch to a situation of low infection prevalence in macaques as the high infection rates macaques experience in the forest are not maintained, and hence we observe a corresponding decrease in human prevalence. An increased LLIN/LLIH coverage in the village and forest is predicted to decrease human infection prevalence. At 100% coverage prevalence drops by approximately 40% due to the combined direct impact of personal protection and the indirect impacts of vector killing, repellency, and a longer gonotrophic cycle (Figure 3a). The low infection prevalence still observed with some plausible parameter sets at 100% coverage is due to infection in the farm, where LLINs/LLIHs are not assumed to be used, and also the small proportion of individuals not using a net even when they are available. The human component of the reproduction number under control (R0H_C) decreases as expected with increasing LLIN/LLIH coverage (Figure 3b), with coverage greater than 5% required to reduce R0H_C to less than 1 in the single scenario in which R0H was greater than 1 (represented by the pale pink area in Figure 3b and as described in Table 2). As expected, human infection prevalence decreases rapidly with increasing coverage of access to rapid treatment where the infectious period in humans is three times shorter than when rapid treatment is not available (Figure 4). Since rapid treatment decreases the infectious period in humans, clearance of the parasite brings overall human infection prevalence down. Additionally, if all infected individuals were treated promptly, there would be minimal onward transmission from humans to vectors and hence a lower risk of infection from vectors to humans in the villages and farms. Our model examines how P. knowlesi transmission may depend on different mixing patterns between humans and the primary host, long-tailed macaques, in different ecological settings. The model suggests that macaques sustain P. knowlesi transmission with minimal or no self-sustaining transmission between humans and vectors in the absence of macaques. However there is potential for this to change if macaque mixing patterns change in the farm (at the forest-fringe) with the highest infection prevalence among humans occurring when macaques forage in the farms but return sufficiently frequently to the forest where they experience higher contact with vectors and hence sustain transmission. The model suggests that the majority of transmission is sustained and driven by the macaque population. This result is supported by data that show that patients presenting at hospitals infected with P. knowlesi were mostly subsistence farmers whose work took them into the forest or plantations associated with forest on a regular basis, or individuals who travelled through at risk areas [36]. As such, among the population at risk, the majority of infections occur in men aged 20–29 years [10]. However, P. knowlesi is now the most common cause of malaria in Sabah, Malaysia and true numbers of human infections may be missed by passive case detection at facilities. Recent studies have shown that all ages and both sexes are susceptible to infection with cases also reported in Malaysian children [29], and Vietnamese children [37], [38]. Additionally, familial clustering of cases has been demonstrated indicating transmission is probably now occurring peri-domestically contrary to previous reports, and that this may be linked to deforestation and/or land-use change in these environments [14]. The vector species that have been implicated in the transmission of P. knowlesi are numerous and the dynamics of many of these are poorly understood [23], [39]. Therefore additional data on the exact vector species present in the different ecological zones; forest canopy, forest ground level, farm, and village and their respective bionomic data including extrinsic incubation periods could be used to improve the model. P. knowlesi has not yet been reported beyond the range of the An. leucosphyrus group which are predominantly forest mosquitoes, occasionally found at forest fringes and open areas where presumably incidental human infection occurs. Experimentally, however, the entire Leucosphyrus group, comprised of 20 species, can transmit P. knowlesi under favourable conditions [16]. Thus it is probable that the current restriction of P. knowlesi to a vector which prefers the forest fringe habitat rather than a completely anthropophilic one has limited the emergence of P. knowlesi as a fully human malaria parasite and public health threat [5]. The likelihood is that where multiple vectors exist, such as in the Malaysian Borneo, they occupy distinct environmental niches with mosquito trapping likely to be logistically demanding [40]. The extent to which variations in species-specific host blood meal choice and susceptibility to plasmodial infection influence transmission dynamics is not known. Even if infection becomes more prevalent in the human population and the domestic environment, it is the individuals who spend time in proximity to areas where macaques are also active, the farm or forest, who will remain at most risk of zoonotic P. knowlesi infection. Thus control measures directed to these at-risk areas and populations would be beneficial as a whole. Our simulations showed that with 100% LLIN/LLIH coverage in the village and the forest, human infection prevalence can be reduced by up to 42%. Studies looking at the effectiveness of bed nets on P. falciparum have reported overall protective effectiveness of 17%–54% [41], [42]. We have assumed that insecticide-treated hammocks (LLIH) can be used in the forest and that they are as effective as LLINs [35]. Magris et al. , found that LLIHs could reduce parasitaemia by 83% among the Yamomami people in Southern Venezuela [43]. Other studies have found reduction in malaria prevalence was 1. 6 times greater when LLIHs were included in the intervention, with a 46% (95% CI: 35–55%) reduction in biting rates against Anopheles minimus in forested villages in Cambodia [35], [44]. Since the majority of infection is maintained in the forest by macaques, individuals who frequent these at risk areas should be made aware of the risks and encouraged to use LLIHs and other preventative options such as repellents as an easy and effective method of protection. However we did not find any impact on macaque infection prevalence with the use of LLIHs in the forest. The use of bed nets in the village will also become increasingly beneficial if human-human transmission becomes more frequent. We found that rapid treatment of infected individuals to be the most effective in reducing infection prevalence among humans with a 95% reduction if every case is treated quickly (or within 5 days in our model). P. knowlesi has a rapid 24 hour erythrocytic cycle, and can result in severe and fatal infections if diagnosis and treatment are not prompt [3], [45]. Current observations show that P. knowlesi patients with uncomplicated malaria respond well to standard schizonticidal drugs with good prognosis and recovery after administration, with no relapse as P. knowlesi does not form dormant liver stages [16], [46]. There is no evidence for chloroquine-resistant P. knowlesi and as such chloroquine represents an inexpensive and highly effective therapy for uncomplicated P. knowlesi infections [47]. Additionally, since the majority of transmission is sustained by macaques, treatment of humans would not exert any substantial drug pressure. As demonstrated in the model validation step, there are wide ranges of parameter values that are consistent with our current understanding of P. knowlesi from the limited data available. The upper bound of the overall R0 of 2267 was due to the extreme values of macaque to vector (CMV) and vector to macaque (CVM) transmission probabilities of 0. 97 and 0. 98 respectively in combination with a 10 year infectious period in macaques. Without detailed bionomic studies and empiric quantification of the natural history of P. knowlesi, it is impossible to reduce the uncertainties surrounding these values. Furthermore the role of super-infection has not yet been documented but it is plausible to assume that infection in macaques is more dynamic than the chronic infection assumed here. Additionally there are several other limitations to the model structure. In conventional malaria models individuals will move from susceptible to a pre-infectious compartment to take into account the latent period, around 9 −12 days from experimental studies in humans [12], rather than straight to an ‘infected’ compartment as set up here. Although the vector populations have been set up to incorporate the extrinsic incubation period crucial to any malaria model, it is assumed that humans and macaques are infectious immediately upon infection. Experimental observations suggest that although P. knowlesi produces gametocytes in mammalian hosts more rapidly than Plasmodium falciparum, they still take approximately 48 hours to develop and mature [16], [17]. Thus the addition of a pre-infectious period for humans and macaques would make the model more robust. This model has assumed a constant seasonality in terms of vector and macaque densities. In many settings seasonality is a key factor in malaria transmission intensity where rainfall influences vector breeding and density. Seasonality is evident in the peak of P. knowlesi notifications in June in Sabah, Malaysia [4]. Seasonal fluctuations in the abundance and availability of different food types in the forest and on farms will also affect the behavior and density of macaques that move between these areas to forage and roost, and therefore any cross-species transmission that occurs may depend on the time of year. Finally this model has been constructed based on the conditions observed in Malaysia and particularly in Sabah; thus predictions derived here may not be applicable to P. knowlesi infections elsewhere. With P. knowlesi cases being reported from several countries throughout South East Asia including Thailand [48], Singapore [49], [50], Indonesia [6], Vietnam [37], Myanmar [51], Cambodia [26], and the Philippines [13], both environmental conditions, demographics, and vector species involved are likely to be considerably different. In summary, our results show that sustained human-vector-human transmission is unlikely to be occurring at present. However, as environmental change continues, there is the potential for the prevalence of P. knowlesi to increase and to become a significant public health problem. Our results highlight the need for sustained control and awareness of this zoonotic malaria particularly as Malaysia enters the pre-elimination stage for other malaria species.

Plasmodium knowlesi is a malaria of macaques which is now recognised as a leading cause of human malaria in Malaysia. Although current human infections are a result of human-macaque contact, there is a potential for P. knowlesi to be transmitted solely among humans. The authors developed a multi-host, multi-site transmission model to assess the likelihood of this happening due to increased human-macaque contact as a consequence of deforestation, population growth, and land-use change. How effective currently available malaria control measures were against P. knowlesi was also an important issue that was explored using the model. Although the model predicts that conventional control measures will be sufficient against P. knowlesi, with the push to eliminate malaria by the end of 2015, it is crucial to be aware of zoonotic malarias which may undermine such efforts.

Abstract Introduction Methods Results Discussion

infectious diseases veterinary diseases zoonoses medicine and health sciences population modeling epidemiology infectious disease epidemiology biology and life sciences population biology infectious disease modeling computational biology infectious disease control veterinary science

2014

Transmission and Control of Plasmodium knowlesi: A Mathematical Modelling Study

Neither genetic nor environmental factors fully account for variability in individual longevity: genetically identical invertebrates in homogenous environments often experience no less variability in lifespan than outbred human populations. Such variability is often assumed to result from stochasticity in damage accumulation over time; however, the identification of early-life gene expression states that predict future longevity would suggest that lifespan is least in part epigenetically determined. Such “biomarkers of aging, ” genetic or otherwise, nevertheless remain rare. In this work, we sought early-life differences in organismal robustness in unperturbed individuals and examined the utility of microRNAs, known regulators of lifespan, development, and robustness, as aging biomarkers. We quantitatively examined Caenorhabditis elegans reared individually in a novel apparatus and observed throughout their lives. Early-to-mid–adulthood measures of homeostatic ability jointly predict 62% of longevity variability. Though correlated, markers of growth/muscle maintenance and of metabolic by-products (“age pigments”) report independently on lifespan, suggesting that graceful aging is not a single process. We further identified three microRNAs in which early-adulthood expression patterns individually predict up to 47% of lifespan differences. Though expression of each increases throughout this time, mir-71 and mir-246 correlate with lifespan, while mir-239 anti-correlates. Two of these three microRNA “biomarkers of aging” act upstream in insulin/IGF-1–like signaling (IIS) and other known longevity pathways, thus we infer that these microRNAs not only report on but also likely determine longevity. Thus, fluctuations in early-life IIS, due to variation in these microRNAs and from other causes, may determine individual lifespan. Inter-individual variation in human longevity has not been found to be under substantial genetic control, with heritability generally between 15% and 30% [1], [2]. At the same time, shared environmental factors contribute little in these human studies, and can be completely controlled in large-scale experiments on inbred invertebrates without abrogating lifespan variability [3]–[5]. Indeed, rearing Caenorhabditis elegans in a homogenous, chemically defined liquid medium more than doubles the coefficient of variability in lifespan compared to feeding the animals live bacteria on solid agar (a less-homogenous environment) [6]. As the external environment of C. elegans can be easily controlled, the genetics of its lifespan are well understood [7], and its developmental plan is famously invariant, this nematode is an ideal organism in which to investigate how and when individuality arises, and how these differences produce a phenotype as variable as lifespan [8]. The identification of “biomarkers of longevity” – measurable parameters that predict individual longevity better than chronological age [9] – will help pinpoint genetic and physiological processes that promote or defer senescent decline. Further, such biomarkers may help clarify whether lifespan differences are simply the result of variable accumulation of damage over time, or whether they may also result from gene-regulatory states, potentially set early in life, that determine individual robustness [10]. To date, most identified and proposed biomarkers in C. elegans have largely been phenomenological, downstream indicators of homeostatic maintenance. One important class of such markers is locomotory function. Herndon and colleagues showed that a qualitative evaluation of individual locomotory ability correlates with remaining lifespan of same-aged animals, and, moreover, that these movement classes correlate with the degree of sarcopenia (decline in muscle mass and function) in those individuals [11]. Later work showed quantitative correlations between the rate of decrease in body movement and lifespan [12], as well as between the span of functional pharyngeal pumping or body movement and lifespan [13], [14]. In addition to muscle decline, general decreases in macromolecular homeostasis have long been observed in aging C. elegans via increases in non-hydrolysable, autofluorescent “age pigments” such as lipofuscin [15] in intestinal lysosomes [16]–[18]. Lipofuscin accumulation correlates with the qualitative movement classes defined in Herndon et al. [19], though such accumulation has not been directly shown to predict an individual' s future longevity, and in one recent work was specifically not found be predictive of longevity [20]. (This last observation was made of green-wavelength autofluorescence, which is more specific for flavin compounds, while lipofuscin per se fluoresces most strongly in blue wavelengths [19], [21].) Lastly, animals that reach their final adult size more rapidly have shorter lifespans [14]. As part of this work, we systematically validated adult growth and movement rates, tissue homeostasis, and age pigment accumulation as phenomenological biomarkers of longevity in nematodes, and further, by measuring multiple biomarkers per individual, deduced the relationships among these markers. While they may suggest clinically relevant markers of human aging, such measurements do little to elucidate the genetic mechanisms underlying lifespan variability. Transcriptional profiling of aging C. elegans has suggested sets of genes that change expression during aging and may thus report an animal' s “physiological age” [22]–[24]. The definitive test remains demonstrating that a particular gene' s expression level predicts future longevity on an individual basis. The first genetic predictor of individual lifespan was identified by Rea and colleagues, who demonstrated that the ability to upregulate a reporter for expression of the heat-shock protein hsp-16. 2 after a mild heat stress correlates with post-stress longevity [25]. However, such stress also induces a protective effect [26]. Thus it is not clear whether the measured effect reflects innate differences in “heat-shock response capacity”, which in un-stressed animals might also correlate with future longevity, or whether the degree of heat-shock response is determined stochastically at the time of the stress. Recently, mid-life expression variation in several additional genes in C. elegans, including daf-16 and its well-known target sod-3, have been shown to predict future longevity in un-perturbed individuals [20]. Therefore, we sought regulatory factors further upstream that might have constitutive activities that determine robustness to damage and/or longevity in unperturbed animals. MicroRNAs (miRNAs) – short non-coding RNAs that bind to and regulate the expression of target mRNAs – have been proposed as determinants of organismal robustness to environmental variation [27], a prediction that has been borne out experimentally [28], [29]. Similarly, miRNAs may regulate longevity by determining individual capacities to respond to damage [30]. lin-4 was the first miRNA to be shown to regulate lifespan and stress-resistance, through its action on the insulin/IGF-1-like signaling (IIS) pathway [31], which is well known for its role in longevity determination [32]–[34]. Many miRNAs change expression levels during aging in C. elegans [35], and recently mir-71, mir-239, and mir-246, all of which increase in expression over time, have been shown to promote (mir-71, mir-246) and antagonize (mir-239) longevity and stress-resistance, through IIS (mir-71, mir-239) and the DNA damage response pathway (mir-71) [36]. Further, miRNAs in other contexts have proven to be able biomarkers of various human pathologies [37]–[40] and perhaps also aging [41]. Here we report that mir-71, mir-239, and mir-246 expression profiles, measured by promoter: : GFP reporter constructs, predict individual longevity in C. elegans. To determine early-life correlates of eventual longevity, we developed a minimally invasive individual-nematode culture system (Figure 1A) that allows in situ imaging of freely moving, unanesthetized animals. Briefly, single eggs at the pre-hatch “pretzel” stage and a bacterial food source are deposited atop PEG-1000-methacrylate hydrogel pads embedded in and crosslinked to a glass slide (see Materials and Methods). The top of the slide is covered with liquid polydimethylsiloxane (PDMS), which polymerizes in approximately 12 hours at 23°C to yield a thin, transparent, and gas-permeable membrane that reduces desiccation and prevents contamination. (All ages reported in this work refer to time after slide preparation; as approximately 98% of viable eggs hatched within 5 hours, we simply report this as “age post-hatch. ”) All strains were crossed into the temperature-sensitive fertility-defective strain spe-9 (hc88) and all assays were conducted at 23°C to prevent reproduction [42]. We obtained good developmental synchrony with this method; after 40 hours, most animals are near the middle of the 4th-larval stage, based on vulval morphology (not shown). The mean lifespan of 10. 7 days at 23°C in this apparatus is similar to that in standard culture conditions, according to previous reports and our own controls (see Materials and Methods). At each timepoint, brightfield/fluorescence image pairs were acquired for each animal, and movement rates and health status evaluated by examining motion after stimulation with 0. 25 seconds of green light; animals that did not respond were deemed to have perished (Figure S1). Figure 1B illustrates the distribution of lifespans of 463 individuals cultured in this apparatus and the corresponding survival curve. In this fashion, measurements can be made on individual animals throughout their lives and correlated with eventual longevity. In particular, the position of each animal was identified in brightfield images via custom semi-automated software, allowing quantification of various morphological and image-based features. As an example, Figure 1C illustrates the length of one particular animal measured at daily intervals from hatching until death. In attempting to determine correlates of future longevity, we focused on measurements made during days 3–7 post-hatch, stretching from the attainment of adulthood (beginning of the reproductive period) to the onset of mortality. Less than 3% of the animals die before day 7; while measurements made later often correlate better with remaining longevity, they are of more limited utility as more of the study population has died before the measurements could be made. The day 3–7 range is illustrated in Figure 1C along with the two measures we employed to summarize data in this range: the mean level of a particular measurement over that span, and the slope of a least-squares linear fit of the data in that span. As noted previously [14], we observed that adult C. elegans tend to shrink over time (Figure 1D). This shortening is not observed in the length distributions of heat-killed animals of different ages [43]; we also anecdotally observed that animals often “relax” and lengthen after death. Thus, the shortening appears to reflect a physiological process and not an actual change in the size of the cuticle. As such, we wished, as an illustrative test case, to examine whether size and/or size maintenance over time had any relevance for eventual lifespan. Figure 1E illustrates a retrospective analysis: the average length-versus-time profile is shown for animals grouped according to the number of days lived. Quite clearly, longer-lived animals are both markedly larger than their short-lived siblings and better able to maintain their length over time. This analysis can be made prospectively as well: the slope of the length vs. time curve between days 3 and 7 (as per Figure 1C) correlates well with each animal' s future longevity (Figure 1F). Specifically, 27% of overall lifespan variability is accounted for by the days 3–7 length slope alone. (This correlation, and all others shown, is the aggregate of several trials, described in Table S1; per-trial results are given in Figure S3.) The mean length over that time range also correlates positively with lifespan; including both in the regression analysis increases the R2 measure of lifespan-predictive ability to 32%. We found similar correlations with volume and surface area; however, length is the most robust. Note that the R2 value is often an over-optimistic estimate of how well a model will predict values from future data, due to “over-fitting” of particular features of the original dataset, particularly with least-squares models, small or outlier-prone datasets, and/or multiple independent parameters. We therefore also estimated future predictive ability via leave-one-out (l. o. o.) cross-validation, in which the prediction for each data point is generated from a model constructed using all other data points. For the length measurements the l. o. o. R2 is 31%. Finally, regression models predict lifespan quantitatively; we can simplify this to a categorical measure to ask how well above- or below-average predicted longevity translates to actual longevity. Figure S2A shows the distributions of observed lifespans for animals with above-average and below-average predicted longevity based the two length measurements (days 3–7 slope and mean); Figure S2B illustrates the corresponding survival curves. We find that above-average length-predicted lifespan is 71% sensitive and specific for above-average longevity. (Defining the test about the average predicted and measured values yields balanced sensitivity and specificity; other thresholds trade off between the two.) The above-average-predicted-lifespan cohort has a 17% increase in mean lifespan compared to the below-average cohort. It had been previously speculated that that age pigments, known to correlate with the current health state, will be predictive of future longevity [19], though this was not borne out in a recent study [20]. We therefore tested accumulation of autofluorescent age pigments (imaged through a red filter set and apparent in gut granules and in aged gonads; see Materials and Methods and Figure S4). Figure 2A shows the patterns in autofluorescent age pigment accumulation in two individuals, computationally straightened and fit to the average day-5 shape and size for visualization, between days 3 and 7 post-hatch; Figure 2B shows the pigment accumulation trends for cohorts with different lifespans. (These measurements are of the 95th percentile of pixel intensity within the defined “worm region” of the original images; here and in all subsequent cases, other measures such as mean or median yield similar results.) Clearly, the longer-lived animals have lower absolute levels of pigmentation, even early in life, and also lower rates of increase in pigment levels. Prospectively, animals with higher levels of autofluorescence, and also those with higher rates of autofluorescence accumulation, in days 3–7 post hatch are likely to experience shorter lifespans (Figure 2C). Various measures of movement rates have been shown to predict future longevity [11]–[14]; we attempted to replicate this finding by calculating a daily movement score from pixel intensity differences in sequential images (see Materials and Methods); higher scores indicate more movement (Figure 2D). We found that the mean and slope of the motion score, days 3–7, positively correlate with eventual longevity (Figure 2E). That is, high movement rates and maintenance of these rates through mid-adulthood are markers of longer lifespan, strengthening the conclusions from previous studies. Finally, aged C. elegans have a very typical “decrepit” appearance in brightfield images [11], [18]. As quantitative measurements of image texture have previously been used as proxies of age-related tissue deterioration (in particular, sarcopenia) in nematodes [44], [45], we defined a daily measure of whole-animal textural decrepitude (see Materials and Methods and Figure 2F) which, examined between days 3 and 7, predicts a sizable fraction of longevity variation (Figure 2G): more deteriorated-appearing animals (high mean decrepitude days 3–7) and those that more rapidly become so (positive slope) are shorter-lived. Overall, each of the individual measurements shown in Figure 1 and Figure 2 consists of a phenomenological evaluation of one or more aspects of nematode health states, encompassing sarcopenia (motion, image texture, size), tissue maintenance (image texture, size), and autophagocytic ability (autofluorescence accumulation). Because these measurements of tissue and cellular homeostasis are integrative and relatively “downstream”, they provide powerful mid-life predictors of eventual longevity; however, they yield few clues regarding the origin of individual differences in longevity. The miRNA mir-71 increases rapidly in expression during larval development, peaks at early to mid-adulthood, and then gradually declines (Figure 3A, 3B and [36]). Beyond differences in lifespan and stress-resistance [36], mir-71 mutant animals appear phenotypically wild-type [46]. Further, mir-71 genetically interacts with IIS downstream of the insulin receptor homolog daf-2 but upstream of daf-16, the FOXO transcription factor that is a major IIS effector [36]. miR-71 is predicted to target several genes in the IIS pathway; of these, pdk-1 levels are greatly increased in aged animals lacking mir-71 [36]. Additionally, mir-71 appears to be both a downstream target of and a regulator of DNA damage responses via CDC-25. 1 [36]. A transgenic reporter, mir-71: : GFP, containing the promoter of mir-71 driving GFP expression, was previously characterized [36], [47]. Though ubiquitously expressed during adulthood, mir-71: : GFP expression is most prominent in the hypodermis, pharynx, vulva, intestinal, and tail cells [36], [47], [48]. Longer-lived cohorts have distinctly different temporal patterns of mir-71: : GFP expression compared to shorter-lived siblings (Figure 3B). Expression levels are here quantified as the 95th percentile of pixel intensities in the worm' s head region; other measurements (mean, median, etc.), and/or aggregating across the whole body, produce similar results though we find this to be the most robust. Specifically, retention of “youthful” mir-71 states, both in terms of high levels of mir-71: : GFP expression and of maintenance of these levels through mid-adulthood, correlates with longevity. Prospectively, both the mean mir-71: : GFP expression levels and the change in these levels between days 3 and 7 both correlate with future lifespan variability (Figure 3C and 3D); together these two parameters predict 35% of lifespan variation (the l. o. o. value is 32%). Animals with higher or longer-lasting mir-71: : GFP expression tend to live longer, consistent with the known role of miR-71 in promoting lifespan [36]. To further examine and quantify trends in mir-71: : GFP expression patterns, we used principal components analysis (PCA). This procedure is conceptually similar to one described recently [49], which employed hierarchical clustering instead of PCA. We used 979 fluorescent images from 146 individual animals, controlled for individual differences in size, shape, internal compression due to locomotion, and overall mir-71: : GFP expression level (see Materials and Methods) so that the analysis captured trends only in the spatial pattern of expression. The first principal component, which by definition explains the single largest correlated trend in the dataset (18% of total expression-pattern variability in this case), captures a transition from a highly specific head/vulva/tail expression pattern to more diffuse whole-body expression (Figure 3E). As the individuals shown in Figure 3A illustrate, there is both low-level whole-body background expression of mir-71: : GFP that remains relatively constant over time, and strong head/vulva/tail-specific expression that peaks and declines. The position of the mir-71: : GFP expression pattern along this principal component (“PC score”), in terms of standard deviations above or below the mean expression pattern (Figure 3E), therefore quantifies the degree of strong, tissue-specific expression at a given day. Figure 3F clearly shows that longer-lived cohorts have more positive and increasing scores, corresponding to high (and increasing) degrees tissue-specificity of expression, while short-lived cohorts have more negative and decreasing scores. Quantitatively, maintenance of head/vulva/tail expression (measured here by the slope of days 3–7 PC scores, though in general other approaches could be employed), and the average overall degree of head/vulva/tail expression (mean PC score day 3–7), are highly correlated with longevity (R2 = 29% for slope and 39% for mean); jointly they predict 47% of individual longevity variation (Figure 3G; l. o. o. 45%). Animals with above-average predicted longevities based on these two measurements have substantially different observed lifespans than those with below-average predictions, illustrating the utility of these measures as a diagnostic test of longevity (Figure S2D; the difference in mean lifespan between these two cohorts is 20%). We confirmed that GFP expression alone does not predict lifespan by examining animals bearing the mIs10 transgene, which contains a myo-2 promoter driving GFP expression in the larval and adult head, a gut enhancer driving intestinal GFP expression in the adult, and pes-10: : GFP, which is expressed embryonically. We found no whole-body or head-only summary of GFP (mean/median/95th percentile/etc.) that, measured in terms of mean or slope over days 3–7 (or various other ranges), predicts longevity to any significant or substantial degree in this dataset (not shown). In addition, other miRNA promoter: : GFP fusions correlate (and anti-correlate) with longevity to different degrees (below). If the primary lifespan-determining target of miR-71 regulation is IIS, then mir-71: : GFP expression patterns should no longer correlate with lifespan absent IIS. We tested this by examining mir-71: : GFP fluorescence in a daf-16 (mu86) background, which lacks this primary IIS effector. Because daf-16 lies extremely close to the spe-9 genomic locus, it was impractical to construct a mir-71: : GFP; daf-16; spe-9 strain. Thus, we modified our experimental protocol to use and to allow for the placement of synchronized young-adult animals onto gel pads treated with the drug 5-fluoro-2′-deoxyuridine (FUDR) to prevent reproduction in the culture apparatus (see Materials and Methods). In this regime, lifespan was somewhat extended, along with a concomitant “stretching out” of the rise-peak-fall temporal expression pattern of mir-71: : GFP intensity (Figure S5). This may be at least partially due to the FUDR, which has been shown to extend lifespan in an environment-dependent fashion [50]. Nevertheless, mir-71: : GFP levels remain predictive of longevity after the day 3–7 measurement window is adjusted to account for the lifespan extension (Figure S5A and Figure 4B). However, in the daf-16 (mu86) background, mir-71: : GFP expression differences were no longer apparent between cohorts with different lifespans (Figure 4A). Because daf-16 (mu86) animals are short-lived, we adjusted the “early adulthood” window in which measurements of GFP expression were made to match (Figure S5A); in this window, neither the increase in (Figure 4B), nor the mean level of (not shown), mir-71: : GFP positively correlated with lifespan. If no accounting for shortened lifespan is made (and, indeed, we observe no compression of the rise-peak-fall temporal expression pattern of mir-71: : GFP in the daf-16 background; Figure S5B), we found that GFP expression changes anti-correlate with lifespan (Figure 4B). This anti-correlation appears to be driven by a small subpopulation of the most short-lived animals which have very high mir-71: : GFP expression levels. Thus, in a daf-16 null background, the predictive power of mir-71: : GFP expression is either suppressed, or, potentially, reversed. Finally, we note that compared to matched mir-71: : GFP controls, mir-71: : GFP; daf-16 (mu86) animals had approximately double the peak GFP expression levels (Figure S5B), though the shape of the temporal pattern (Figure S5B) and spatial distribution (Figure S5C) of expression remained extremely similar. This may suggest negative-feedback regulation of miR-71 by DAF-16 or one of its targets. Much like mir-71, mir-246 mutants appear phenotypically wild-type except for decreased longevity and stress-resistance [36], [46]. The expression of miR-246 increases over time, and a mir-246: : GFP construct shows that the gene is expressed in the gonadal sheath [36], [47]. Our detailed analysis of mir-246: : GFP in individual animals shows a gradual plateauing of mir-246 expression in late adulthood (Figure 5A and 5B), but, unlike mir-71, no concomitant loss of tissue specificity. We find that animals in which mir-246: : GFP levels plateau more slowly (measured by the slope of mir-246: : GFP 95th-percentile fluorescence intensity between days 3 and 7) are relatively longer-lived: change in mir-246 expression in this time range predicts 20% of total longevity variation (l. o. o. 18%; Figure 5B and 5C). The mean level of mir-246: : GFP between days 3 and 7, however, does not clearly predict longevity. Additionally, we observed that while the distributions of lifespans for animals with slow- vs. fast-increasing mir-246: : GFP expression are significantly different (Figure S2E), they have nearly identical modal values; however, a subset with particularly early mortality appears to be associated with low mir-246: : GFP slopes. None of the principal components of spatial expression variability correlate with longevity: mir-246: : GFP expression is highly tissue-specific, and so the principal components predominantly capture uninteresting variations in the internal position of the gonad sheath. Unlike mir-71 and mir-246, mir-239 antagonizes longevity: mutants lacking the identical mir-239a and mir-239b sequences have increased lifespan and stress-resistance (though, again, no other clear phenotypes) [36], [46]. Genetic experiments suggest that mir-239 is downstream of daf-2 and upstream of daf-16; miR-239 may promote IIS (which is anti-longevity) via indirectly increasing levels of the cytoplasmic IIS transduction components AGE-1 (the catalytic subunit of phosphatidylinositol 3-kinase) and PDK-1. mir-239: : GFP expression is predominantly in several head and tail neurons, with lower levels in pharyngeal and gut tissues [36]. While mir-71 and mir-246 levels peak or plateau, respectively, at mid-life (Figure 3A, 3B and Figure 5A, 5B), we observe that mir-239: : GFP expression levels drift upward over time (Figure 5D and 5E). Consistent with its role as a lifespan antagonist, higher mir-239: : GFP levels correlate with shorter lifespans. As there is relatively little variability in mir-239: : GFP expression at post-hatch day 3, the days 3–7 increase and the day-7 magnitude of expression capture similar information, and both predict approximately 10% of longevity variability. For the slope measure, the l. o. o. R2 is 7% (Figure 5F). Though this value is somewhat low compared to the others reported, the lifespans of high and low mir-239-expressing animals remain significantly and substantially different, with a difference in means of approximately one day (Figure S2F), or ∼10% of the average lifespan; moreover, as a diagnostic test for above-average longevity, above-average mir-239: : GFP slope performs nearly as well as mir-246: : GFP or the motion- or texture-based measures (Figure S2). As with mir-246: : GFP, PCA applied to mir-239: : GFP images did not yield any expression-pattern trends predictive of future longevity. We measured many of the reported biomarkers in the same animals, enabling us to construct a multivariate predictor of nematode longevity, which we term the “survival prediction index”. This model, incorporating the length, motion, texture, and autofluorescence-accumulation measurements described in Figure 1 and Figure 2, predicts 62% of all lifespan variability across all datasets for which these parameters were measured (l. o. o. 57%; Figure 6A), and above-average “survival prediction indices” divide the animals into well-delineated long-lived and short-lived subgroups (Figure S2G; mean lifespan is 22% increased in the high-survival-index cohort compared to the low-index cohort). The relative importance of each biomarker can be inferred by examination of the relative regression weights: autofluorescence (slope): −0. 302; length (slope): 0. 266; length (mean): 0. 174; motion (mean): 0. 149; motion (slope): 0. 137; texture (mean): 0. 121; texture (slope): 0. 077. (To render regression weights directly comparable, all input parameter values were expressed in unit-free terms of standard deviations from their mean; including mean autofluorescence values did not improve lifespan-predictive ability.) Overall, the measures of size and age pigments dominate (for these measures alone, lifespan-prediction R2 = 55%, l. o. o. 53%). Further, we find that adding mir-71 or mir-239 measurements to those above do not improve the lifespan predictions of the model (not shown), indicating that these measurements provide information that is also captured by the downstream, phenomenological markers (see also below). However, adding mir-246 slope measurements adds at least 5% to the R2 value attained from the phenomenological markers, which suggests that miR-246 promotes longevity via more than just the mechanisms reported on by length, texture, motion, and autofluorescence. We formalized this analysis by inferring the conditional independencies of various parameters from partial correlations. That is, while all of the biomarkers correlate with one another, it is possible to statistically infer whether these relationships are direct or indirect. For example: mir-71 levels correlate with longevity, but this correlation is largely abrogated when length maintenance is controlled for; conversely, controlling for mir-71 reduces the correlation between length and lifespan to a much lesser degree (see Table S2, which lists the correlation of each marker with longevity after various controls). Therefore either mir-71 influences length, which then influences longevity, or length variability is an upstream cause of both mir-71 variability and lifespan variability. To systematically evaluate these interactions, we constructed a partial correlation network (also known as a graphical Gaussian model) from our data (Figure 6B; see Materials and Methods) [51]. We have not yet directly evaluated the relationship between the different miRNAs, which requires measuring the promoter activity of these genes in the same animals. We can, however, indirectly infer the relationship between these miRNAs via intervening factors measured in all datasets: the network in Figure 6B reflects the consensus of networks calculated from each dataset (Figure S6). While the relatively weak relationship between mir-239 and lifespan prevents its accurate placement into the network, other trends are clear. Age-pigment accumulation, motion, and texture/mir-71/length provide relatively independent information about lifespan from one another, though there may be some relation between the motion and texture scores (Figure S6). Including mir-246 decreases the link between length and lifespan (Figure S6 and Table S2), suggesting that, while its precise position is indeterminate, mir-246 is either an upstream cause of both length and lifespan variation, or mir-246 levels are determined by physiological processes that regulate length maintenance, and these levels more directly control lifespan. Finally, we examined the timing of each measure' s lifespan-predictive ability: do some measurements provide earlier hints at longevity than others? The accuracy of lifespan prediction for each measure (measured as the R2 value) is plotted versus time in Figure 6C. For this analysis, we did not manually determine one or two summaries of the time-course of biomarker measurements (e. g. “mean length between days 3 and 7”), but instead predicted longevity using all of the raw time-course data up to a given age (e. g. to evaluate the predictive ability of length up to day 4, we regressed lifespan against the lengths measured at days 2,3, and 4). We used ridge-regression with an automatically-determined penalty in order to prevent over-fitting due to the increased number of parameters at later time-points (see Materials and Methods); note however that the R2 values are in some cases more optimistic compared to the simpler measurements defined earlier. Based on this analysis, we observe that the various predictors of longevity differ markedly in their timing: the mir-71 PCA scores provide the earliest, and best, predictors of lifespan, while mir-246 and motion provide the latest-onset information about future lifespan. We used a novel culture system that allowed us to quantitatively examine individual nematodes throughout time and directly correlate inter-individual variability in early life with variability in eventual lifespan. Through this observational study, we identified phenotypic markers that, measured while >97% of the population remains alive, predict over 60% of future individual longevity variability. In addition to confirming markers that previously had been shown to correlate with lifespan (movement rates [12]–[14]) and widely suspected to do so (visual decrepitude [11], [18], [45], age pigment accumulation [17]–[19], [43]) we have also found a novel biomarker of longevity in maintenance of adult size through mid-life (Figure 1E and 1F). This dovetails with the recent finding that juvenile “growth span” (time to reach adult size) also predicts lifespan [14]. Our results regarding age pigmentation are inconsistent with others recently reported [20], an observation which may be explained by the fact that we measured age pigmentation in the red spectral range (see Materials and Methods), while Sánchez-Blanco and Kim examined green autofluorescence. Note that lipofuscin, per se (as opposed to other autofluorescent “age pigments”), fluoresces most strongly in blue ranges [21]. This indicates the potential heterogeneity of age pigmentation across the color spectrum. Seeking a more detailed understanding of the genetics of individual longevity determination, we found that fluctuations in the levels of miR-71,239, and 246 (imputed via promoter: : GFP reporters) predict a substantial portion of longevity. To date, only one other group has reported any genes, in any species, in which endogenous, un-perturbed fluctuations in expression level have been reported to be predictive of future longevity [20]. The ability of expression states to predict later longevity quite early in life further suggests that some fraction of longevity variance is indeed the result of developmentally determined epigenetic states of “robustness” or “frailty”. This hypothesis has been raised theoretically [52], and the degree of post-heat-shock longevity-extension (correlated with HSP-16. 2 expression [25]) has been shown to fit this model well [53], [54]. In particular, the degree of strong, tissue-specific mir-71 expression in the head, vulva, and tail is the single most powerful biomarker identified, explaining 47% of future longevity (Figure 3F and 3G), and also the earliest indicator of future longevity (Figure 6C). Combining these findings, which are based on observation of essentially unperturbed individuals, with earlier knockout studies that demonstrate that abrogation of these miRNAs alters mean lifespan [36], we infer that individual, wild-type variability in the expression of these regulators not only reports on longevity but may likely determine it as well. (In all cases, the correlation or anti-correlation of miRNA levels with longevity is in the same direction as suggested by previous knockout and overexpression studies [36].) Note that it is not necessarily the case that genes that alter lifespan when removed or artificially overexpressed will determine individual lifespan in wild-type conditions: core health-determining genes may be tightly regulated to have low inter-individual variation, or that variability may be buffered downstream and not translate into altered longevity. In the case of mir-71 and mir-239, the mechanism of individual lifespan determination may be via the well-known insulin/IGF-1-like signaling pathway (IIS) [36], the absence of which promotes longevity via stress-response and macromolecular homeostatic mechanisms, among others [55]–[57]. In particular, mir-71 knockout led to increased expression of components of the IIS transduction machinery [36], and we here found that daf-16 null animals have increased levels of mir-71: : GFP expression. These observations suggest that DAF-16 and miR-71 levels are held in homeostatic balance via a mutually regulatory feedback loop. Thus, it is possible that inter-individual fluctuations in miR-71 levels may directly determine inter-individual levels of tonic IIS activity, and hence, individual lifespan. This conclusion is strengthened by our finding that, absent DAF-16, mir-71: : GFP is no longer predictive of longevity. One final piece of evidence for the role of IIS in lifespan determination comes in the work of Sánchez-Blanco and Kim [20], which identifies daf-16 and sod-3, a canonical DAF-16 target, as among the best predictors of individual longevity. (In fact, sod-3 reporter expression is their strongest predictor reported, with an R2-value of 0. 32.) Further, mir-71 may also act by downregulating cell-cycle checkpoint proteins [36], the absence of which also promotes stress-response factors, even in postmitotic cells [58]. The early predictive ability of mir-71: : GFP suggests that mir-71, and thus either or both of the signaling pathways it regulates, may play an early-life role in determining organismal robustness and longevity. According to our network analysis, image texture, mir-71: : GFP expression, and size are all closely related, with size as the most downstream measure of what we take to be the age-regulated tissue disorganization and sarcopenia reported previously [11], [18]. Overall, and especially given mir-71' s role in regulating pathways that determine levels of stress responses, we suspect that this nexus of predictors reflects processes of somatic growth and maintenance. However, while both mir-71 and mir-246 levels are correlated with length maintenance, mir-246, unlike mir-71, provides additional information about longevity not captured by length or other “somatic maintenance” features. Given this, the relatively late (i. e. post-reproductive) timing of its predictive ability, the sharp upregulation of mir-246: : GFP expression at reproductive maturity (Figure 5A, 5B and [47]), and its localization literally at the interface between the gonad and other somatic tissues, it is tempting to speculate that mir-246 is involved in balancing reproduction and somatic maintenance [32], [59]. Further, we find that age pigment accumulation also provides a degree of information about lifespan not captured by the “somatic maintenance” nexus, suggesting that it, too, reflects an inter-related but parallel lifespan-determining process. It is also worth examining these results in light of those of our previous work, which compared miRNA expression over time in wild-type and long-lived daf-2 IIS mutant animals [36]. Often, daf-2 animals are taken as a paradigm case of “long-lived” animals, and the presence or absence of a physiological feature in these mutants assumed indicative of a role as a biomarker of successful aging. While in many cases this is undoubtedly so (for example, decreases in rates of lipofuscin accumulation in these animals [19]), our results yield the unsurprising finding that the relationship between physiological events in daf-2 mutants and long-lived wild-type animals are inexact. Specifically, we here find clear evidence that elevated levels of mir-71 and mir-246 expression are associated with extended longevity, yet in daf-2 animals these miRNAs are not upregulated, likely reflecting their upstream, negative-regulatory roles in insulin signaling [36]. Similarly, we find that daf-16 animals are physiologically quite dissimilar from short-lived wild-type individuals in terms of mir-71: : GFP expression. Intact individuals which are short-lived typically have low mir-71: : GFP levels, while daf-16 animals show a dramatic elevation in mir-71: : GFP. Mechanistically, this may be due to disrupted negative feedback from DAF-16 (or a target) on mir-71; pragmatically it suggests the limitations of inference about the physiology of intact animals based on findings in particular mutants. In this case, based only on the average difference in mir-71: : GFP in daf-16 vs. wild-type, one might incorrectly conclude that high miR-71 levels are a marker of short lifespan. We thus believe there is great utility to the quantitative observation of individual wild-type (or nearly so) animals. Lastly, in all cases observed, it is the retention of young-adult-like trends (high and/or increasing length, mir-71 and 246 expression; low or slowly-increasing autofluorescence and mir-239) into mid- and late-adulthood that predicts longer lifespan. Classic antagonistic pleiotropy theories of aging posit that some age-related degeneration may be due to alleles which are beneficial in early life but become damaging over time (“live fast, die young” effects) [60]; however, in the examples here presented we find that the loss of youthful biometrics, not their continuation, proves most harmful. Thus, from this and previous studies [36], these miRNAs appear to be relatively upstream regulators of lifespan-determining pathways that are relevant to the determination of inter-individual variation in nematode lifespans. The miRNAs we have identified are not well conserved in higher animals so these particular mechanisms of lifespan determination are likely nematode-specific. However, given the conserved nature of aging pathways across phylogeny, our work does suggest that fluctuations in other regulators of these pathways (including other miRNAs) may predict or determine individual aging rates in more complex organisms, perhaps foreshadowing or even controlling the timing of age-related decline in humans. Waterjet-cut borosilicate glass slides were obtained from Advanced Waterjet and Engraving (Anaheim, CA). Before and after use, slides were soaked in a base bath (2: 5 isopropanol: H2O, 1. 5M KOH) to remove all organic material. Prior to use, the slides were treated to functionalize the glass surface with reactive methacryl groups as follows: slides were rinsed in distilled H2O (dH2O) and submerged in 5% HCl (aq.) for 10 min to protonate surface hydroxyls, rinsed again in dH2O and then submerged with agitation for 2 min. in methacryl silane solution (2% v/v 3-methacryloxypropyltrimethoxysilane [Z-6030, Dow-Corning; Midland MI] in 95% ethanol with 0. 02% v/v glacial acetic acid, made fresh and stirred vigorously for 10 min. immediately prior to use). The slides were then rinsed in 95% ethanol, heated at 110°C to effect the condensation of the silane reagent to the glass surface, and stored with desiccant. Prior to use, one side of each slide was sealed with Scotch Premium Performance packing tape (3M; St. Paul, MN). We use a methacryl-difunctional polyethylene glycol to create a crosslinked hydrogel [61] that, when polymerized in the methacryl-derivatized glass wells, crosslinks also to the sides of the wells. This prevents the high rate of escape down the sides of the wells that we observe when using agar gels. Agar-free but otherwise standard nematode growth media [62] was supplemented with 4% w/v dimethacryl PEG-1000 (Polysciences; Warrington, PA), 4% w/v monomethacryl PEG-1100 (Sigma-Aldrich; St. Louis, MO) as a plasticizer, and 0. 1% w/v 1-[4- (2-hydroxyethoxy) -phenyl]-2-hydroxy-2-methyl-1-propane-1-one (Irgacure 2959, BASF; Ludwigshafen, Germany), a water-soluble, photo-activatable crosslinking initiator. This PEG-NGM was filter-sterilized and pipetted to fill the wells in the glass slides level to the top. The slides were then placed in a sealed chamber with a UV-transparent borosilicate glass lid, which was purged with nitrogen and exposed to 1. 5 J of shortwave UV (λmax = 350nm) radiation to initiate crosslinking. 1 µL 12. 5% w/v E. coli OP50 (resuspended in M9) was pipetted onto each NGM-PEG pad, and individual eggs at the pre-hatch “pretzel” stage of development were transferred with an eyelash pick. Liquid polydimethylsiloxane (PDMS; Sylgard 184, Dow-Corning; Midland MI) was mixed 1∶10 with its crosslinking agent, de-gassed for 20 min. under vacuum, and pipetted atop the slide assemblies, which were then placed in 10 cm diameter polystyrene Petri dishes alongside small dH2O-saturated cotton strips (to prevent desiccation), sealed with parafilm, and stored at 23°C. PDMS polymerizes after approximately 12 hours in these conditions. Most eggs hatch within 5 hours of slide preparation and reach their full adult size approximately 50 hours later. Ages reported are hours and days post slide preparation. These values are within the described range for this temperature [63], suggesting that our culture apparatus is substantially similar to standard conditions. Further, our observed mean lifespan of 10. 7 days at 23°C (Figure 1B) is similar to our own measurements of spe-9 (hc88) animals picked as pretzel-stage eggs onto standard NGM plates seeded with OP50 (mean lifespan = 9. 5 days at 24°C, n = 350), and measurements of wild-type animals on solid media reported by others [6], [17]. However, we observe a somewhat smaller standard deviation in lifespan of ≈1. 9 days vs. the 3–4 in those previous studies, suggesting that this culture apparatus provides an extremely uniform environment. For the daf-16 epistasis analysis, we modified the above protocol to allow for chemical sterilization of young adult animals by 5-fluoro-2′-deoxyuridine (FUDR; Sigma-Aldrich; St. Louis, MO). Specifically, FUDR from a 10 mg/mL aqueous stock was added at 1∶100 to PEG-NGM prior to filter-sterilization, which was polymerized in the glass slides as above. FUDR causes growth arrest of animals prior to the 4th larval stage, so synchronized young adult animals were produced by hypochlorite treatment of gravid adults to isolate eggs [42] followed by overnight starvation in M9 buffer to synchronize animals as L1s, which were then plated on standard NGM-agar plates with OP50 food and allowed to grow to young adulthood at 23°C. These animals were transferred individually to PEG-NGM-FUDR slides supplemented with concentrated OP50 as above. Moving animals crawl into polymerizing PDMS, so the slides were sealed with 0. 5mm-thick strips of PDMS that had been pre-cured on a glass plate at 100°C for one hour and cut to size. We observed an increase in longevity under these conditions relative to non-FUDR treated animals plated as embryos (Figure S5A). The following C. elegans strains provided by Caenorhabditis Genetics Center (CGC) were used in our studies: VT2084 (mir-71: : GFP), VT1607 (mir-246: : GFP), and PD4793 (mIs10: myo-2: : GFP; pes-10: : GFP; F22B7. 9: : GFP). mir-239: : GFP was generated previously [36]. All strains were crossed into BA671, a spe-9 (hc88) temperature-sensitive fertilization-deficient mutant, and assays were conducted at the restrictive temperature of 23°C. This strain has normal longevity at this temperature [42]. The small fraction of animals that reproduced in the culture apparatus and could not be clearly distinguished from their offspring was excluded from further analysis. We determined that VT2084 actually contains the complete precursor miRNA sequence for miR-71 inside the “promoter” region driving transgenic GFP expression; therefore we wished to determine whether this strain overexpresses miR-71, which was previously shown to increase longevity [36]. Overexpression of miR-71 was not likely because the transgene in strain VT2084 was integrated via low-copy bombardment; however we confirmed via quantitative RT-PCR analysis (as performed previously [36]) that mature miR-71 levels in mir-71: : GFP; spe-9 (hc88) animals reared at 23°C (as similar as possible to our own culture conditions) showed minimal overexpression. Specifically, synchronized populations of spe-9 (hc88) animals and mir-71: : GFP; spe-9 (hc88) animals were prepared by hypochlorite treatment and overnight starvation in M9, reared on NGM plates seeded with OP50 at 23°C, and harvested at 5 days post-plating, corresponding to the time of peak mir-71: : GFP expression at young adulthood. Levels of miR-71 in mir-71: : GFP; spe-9 (hc88) animals were 102% or 115% that of spe-9 (hc88) animals, depending on whether the U18 RNA or miR-66 (which are not temporally regulated), respectively, was used as a loading control. Mature miR-71 levels were also measured in homozygotic mir-71 null (n4115); mir-71: : GFP animals prepared similarly; here, miR-71 expression was approximately 58% or 25% that of wild-type (using the U18 or miR-66 control, respectively). (The n4115 strain without the mir-71: : GFP transgene had undetectable miR-71 expression.) Further, we see no phenotypic consequence of the extra copies of the miR-71 sequence in VT2084: the lifespan of mir-71: : GFP; spe-9 animals in our apparatus is approximately 1. 05 times that of the mean lifespan of the other strains analyzed, which is well within the range of inter-replicate variability. Thus, we conclude that for the purposes of this work, mir-71: : GFP acts as a phenotypically wild-type reporter of miR-71 expression. We calibrated our microscope daily to control for spatial and temporal variation in light-source intensity, as described previously [36]. At the desired sample interval (typically daily), each slide was briefly removed from its humid chamber and placed in an upright microscope (Axioplan 2i; Carl Zeiss; Oberkochen, Germany), driven by custom software, for acquisition of brightfield and fluorescent images at 10× magnification. Per-slide acquisition time was typically under 20 minutes. Each animal was manually located and brought into focus, and a series of 10ms-exposure brightfield images were acquired, interleaved with fluorescence exposures of 1,10, and 100 ms, in order to ensure that a properly exposed image was obtained. This sequence was performed for each filter-set of interest; in this case a GFP-bandpass filter to measure transgene expression (41017; Chroma; Bellows Falls, VT) and a TRITC filter to measure autofluorescence (41002c; Chroma). While the peak autofluorescence of lipofuscin, a chief age pigment, is in the blue range [19], blue light evokes a strong escape response [64], which is problematic as the animals typically leave the field of view rapidly thereafter. As the green range was used for GFP measurements, we compromised and measured age pigment species that autofluoresce in the red range. Like lipofuscin, we observed these species in gut granules [16] and also in larger gonadal inclusions (Figure S4). Further, we established that our measurements of GFP intensity were not biased by light emitted from the autofluorescent species that also fluoresce in the same wavelengths. We found that the relative degree of autofluorescence was quite low (<10% relative intensity) compared to the GFP signals measured, even in aged animals; correcting for this bleed-through did not alter any findings. Further, the 95th-percentile measurements of image intensity we used are less sensitive to low-intensity autofluorescence signals as compared to measures of mean image intensity (for example). After fluorescence image acquisition, each animal was further stimulated with an 0. 25-second pulse of green light, which stimulates a robust escape response in healthy animals and causes head and/or tail retraction in more decrepit individuals. After a 1-second delay, three images were subsequently recorded at 1-second intervals to measure post-stimulation movement. Post-stimulation motion image sequences were visually scrutinized to determine if voluntary “twitching” occurred. Animals with no detectable motion were determined to have died in the interval between the current and previous image acquisition. (Assuming a Bayesian null hypothesis of a uniform prior distribution over time-of-death within the sample interval, the expectation value for the actual, unknown, time of death is halfway between the current and previous acquisition time. We took the number of hours between slide preparation and this time-of-death estimate as the lifespan.) For each timepoint, the non-overexposed fluorescent image with the longest exposure time was selected, subject to manual review to ensure that the animal' s locomotion did not cause unacceptable blur. As each fluorescent image was acquired with flanking brightfield images, the brightfield image in which the animal' s position is closest to that in the fluorescent image was selected automatically as that with maximum mutual information with the fluorescent image [65], again subject to manual review. Once the best brightfield/fluorescence image pairs were defined, the outline of the animal in each brightfield image was determined using custom semi-automated software; the position was assumed to be the same in the fluorescent image. The nematode-finding procedure was as follows: based on a training set of labeled nematode/non-nematode regions of brightfield images, a logistic regression classifier was trained to estimate the probability that a given patch of pixels is inside of an animal [66]. The classifier was applied to each brightfield image to create a rough mask, which was distance-transformed to produce “valleys” of low values along the midlines of masked regions. Based on user-input head and tail points, a least-cost path through this distance-transformed mask was calculated using Dijkstra' s algorithm. This centerline was manually modified as necessary, and the left and right flanks of each animal determined based on average size for its age, with manual modifications. Given each animal' s centerline and outline, it is trivial to straighten the image and to warp the size and shape of any given animal to a standardized “unit worm” defined by the average size and shape of all animals of a given age. The PCA fluorescence measurements described below were made on standardized, warped images; all other measurements were made from the original images, within the animal' s boundaries as defined. Further, given the head-to-tail centerline, we defined the “head” region as the initial 20% of the animal. Fluorescence measurements were made on images corrected for background camera noise (dark field), spatial illumination inhomogeneities (flat field) and temporal variation in illumination via reference fluorescent beads [36]; after correction, intensity values were divided by the exposure time to render all images comparable. Pixel values within the defined whole-animal or head regions were extracted and summary statistics (such as 95th percentile of intensity) made. These raw measurements are available as Table S3, while per-animal summary statistics (days 3–7 slope and mean, etc.) are provided in Table S4. Four measures of motion were made within the defined animal region: the fraction of pixels changing relative intensity by more than 18% between the different brightfield images acquired, and the average pixel-wise coefficient of variation across these images, both before and after green-light stimulation. ν-support-vector regression (SVR) [67] using an RBF kernel was then used to map these four parameters to the number of days of life remaining, using LIBSVM [68]. Parameters were selected using 10-fold cross-validation on a subset of the input data: C = 10, ν = 0. 8, γ = 0. 3, though performance was roughly equivalent across several decades of C and γ values, 0. 2<ν<0. 9. Thereafter we used 100-fold cross-validation on our dataset of 4318 motion-statistics/days-remaining data points and made predictions for each data point without “peeking” by training the SVR on the days-remaining figure for that data point. The predicted “days of life remaining” based on the four measures of motion was used as our aggregate motion score. For simplicity, image texture features were calculated directly from pixel intensity patterns [69], though “filter-bank” methods have also been employed on nematode images [44], [45]. First, age-specific texture patterns (“textons”) were determined. Brightfield images (acquired through the GFP filterset only) were grouped by age: 3- and 4-day-old, 5- and 6-day-old, up to 15- and 16-day-old. For each group of images, 500000 17×17-pixel patches (within the defined animal outlines) were randomly sampled, after which k-means classification was performed to yield 30 representative textons for each age group (210 overall). Next, the texture of each animal was characterized as follows: for each 17×17-pixel patch falling within the defined brightfield image region, the closest texton (in terms of Euclidian distance) of the 210 overall was determined. The “texture signature” of a given image was defined as the 210-element histogram containing the number of closest-matching 17×17-pixel patches for each texton, divided by the total number of patches in that image. These signatures were then used as input to a support vector regression procedure precisely as described above (best parameters: C = 5, ν = 0. 6, γ = 0. 004; again performance was relatively insensitive to parameter setting). Texture-based predictions of “days of life remaining” were used as texture-decrepitude scores. Principal components analysis was performed on fluorescent images, from day 3 onward, warped to unit size and shape. However, as C. elegans stretch and compress as they move, and due to inter-individual anatomical variation as well as variation in animal-outline-finding, warping images based on the outline alone does not cause anatomical features to come into precise register across every animal. We therefore manually defined the position of the vulva on each brightfield image and used that position to initialize a mild nonlinear warping procedure, which longitudinally stretches and compresses the image using five evenly spaced control points. Given an image and a reference, hill-climbing optimization was then used to find the position of the control points that maximized the correlation coefficient between the image and reference pixels, with a penalty for large deformations. The mir-71: : GFP images were mutually aligned using the expectation-maximization algorithm as follows: the mean image across the population was calculated (expectation step), then each image was warped to match the mean (maximization step). These steps were alternated until convergence; typically three iterations sufficed. Results were then manually inspected to ensure face validity. Finally, the mean pixel intensity of each image was subtracted away so that inter-image variability was due only to the distribution of pixel intensities, and not to overall changes in mean brightness. After this procedure, the principal components analysis was performed on the images, and PCA scores along each component were calculated for each animal at each timepoint. Estimates of the underlying distribution of sampled data (length, lifespans) were performed with Gaussian kernel density estimation, using the Scott' s rule-of-thumb to choose the kernel variance (i. e. bandwidth): σ2n−0. 2 where σ2 is the sample variance and n is the sample size [70]. Pairs of lifespan distributions were tested for equality using the two-tailed Kolmogorov–Smirnov test. Single and multivariate regression of biomarkers versus lifespan was conducted with ordinary least-squares regression, with the coefficient of determination (R2) calculated according to the standard formula. Note that in the univariate case, this is equivalent to the squared Pearson product-moment correlation coefficient (r) between the biomarker and lifespan. Significance of correlations was measured with an F-test of the R2 value: the statistic σ2model/σ2error has an F distribution with (dfmodel, dferror) degrees of freedom, where σ2model and σ2error are the model and error components of the overall variance, respectively, dfmodel = p, dferror = n-p-1, p is the number of fit parameters, and n is the number of observations. As σ2model is the sum of squared distances between the predicted values and the mean value, divided by dfmodel, and σ2error is the sum of squared residuals divided by dferror, simple algebra on the definition of the R2 value yields F = R2 dferror/ ([1-R2] dfmodel). Leave-one-out R2 values were calculated as follows: given one or more biomarker values for a set of individuals, an ordinary least-squares regression model to predict lifespan from these values was estimated based on the data from each individual save one; then the lifespan of that individual was predicted using that model. This was repeated for each individual. The R2 value was calculated from the residuals of the leave-one-out predictions according to the standard formula. Partial correlation networks (Figure 6B and Figure S6) were computed using TETRAD IV [71], using the “PC” search algorithm with the multiple-regression independence test and the α threshold for the test set to 0. 001. Arrows of direction of the influences were discarded, as these inferences were not robust across independence tests or α values; however the basic network structure was robust. Lifespan-predictive values versus age (Figure 6C) were calculated as follows. For each age n, from 2–7 days, all levels measured for a particular marker from day 2 to n were considered. (That is, for n = 2, only the day-2 value was used; for n = 4, the values at days 2,3, and 4 were used.) The values under consideration, and all of their pairwise multiples (“interaction terms”, so that rates of change can be incorporated) were then used to construct a multivariate regression model to measure how well eventual longevity can be predicted using only data up to age n. In order to prevent the changing number of parameters over time from affecting the R2 values, and in particular to avoid over-fitting later timepoints due to large number of parameters and interaction terms, we used ridge regression, with a penalty term automatically chosen to minimize the generalized cross-validation error [72].

Why do some individuals live longer than others? Unexpectedly, genetic differences contribute surprisingly little to lifespan variation in humans. The situation is thrown into relief in studies of C. elegans, in which genetically identical siblings reared in identical environments usually experience different lifespans. In this work, we show that physiological differences between identical animals begin to appear relatively early in life and that markers of ill health in young adulthood presage shorter lifespans. Using fluorescent markers to examine the level of activation of several genes, we found three regulatory microRNA genes that vary in activity between individuals in a manner that predicts future lifespan. Moreover, two of these regulate insulin signaling, which is well known to alter lifespan in diverse species when experimentally manipulated. This indicates that different levels of insulin signaling in otherwise identical individuals may determine differences in lifespan.

Abstract Introduction Results Discussion Materials and Methods

aging physiological processes developmental biology organism development physiology genetics epigenetics biology anatomy and physiology genetics and genomics

2011

MicroRNA Predictors of Longevity in Caenorhabditis elegans

Filamentous fungi are of great importance in ecology, agriculture, medicine, and biotechnology. Thus, it is not surprising that genomes for more than 100 filamentous fungi have been sequenced, most of them by Sanger sequencing. While next-generation sequencing techniques have revolutionized genome resequencing, e. g. for strain comparisons, genetic mapping, or transcriptome and ChIP analyses, de novo assembly of eukaryotic genomes still presents significant hurdles, because of their large size and stretches of repetitive sequences. Filamentous fungi contain few repetitive regions in their 30–90 Mb genomes and thus are suitable candidates to test de novo genome assembly from short sequence reads. Here, we present a high-quality draft sequence of the Sordaria macrospora genome that was obtained by a combination of Illumina/Solexa and Roche/454 sequencing. Paired-end Solexa sequencing of genomic DNA to 85-fold coverage and an additional 10-fold coverage by single-end 454 sequencing resulted in ∼4 Gb of DNA sequence. Reads were assembled to a 40 Mb draft version (N50 of 117 kb) with the Velvet assembler. Comparative analysis with Neurospora genomes increased the N50 to 498 kb. The S. macrospora genome contains even fewer repeat regions than its closest sequenced relative, Neurospora crassa. Comparison with genomes of other fungi showed that S. macrospora, a model organism for morphogenesis and meiosis, harbors duplications of several genes involved in self/nonself-recognition. Furthermore, S. macrospora contains more polyketide biosynthesis genes than N. crassa. Phylogenetic analyses suggest that some of these genes may have been acquired by horizontal gene transfer from a distantly related ascomycete group. Our study shows that, for typical filamentous fungi, de novo assembly of genomes from short sequence reads alone is feasible, that a mixture of Solexa and 454 sequencing substantially improves the assembly, and that the resulting data can be used for comparative studies to address basic questions of fungal biology. Fungi are heterotrophic eukaryotes found in nearly all ecosystems. About 100,000 fungi have been described to date, but conservative estimates predict at least 1. 5 million different species [1], [2]. Fungi exhibit a wide range of different lifestyles, particularly as saprobes, pathogens or symbionts. As saprobes, fungi acquire nutrients from dead organic matter and are among the main recyclers on the planet. They play important roles in the degradation of cellulose and lignin, contributing greatly to the global carbon cycle. However, their saprotrophic activities also cause severe problems with the degradation of man-made products and in causing food spoilage. Mortality from human fungal pathogens has increased in recent years, especially in immunocompromised patients. In plants, ∼90% of diseases are caused by fungi, and these result in massive losses in crop yield worldwide, with often profound socio-economic effects, sometimes resulting in severe famines [3]. Nevertheless, fungi also have beneficial effects in symbioses, such as mycorrhiza (fungus/plant root) and lichen (fungus/algae) associations. Greater than 80% of terrestrial plants have mycorrhizal relationships with fungi that allow the plants to access key nutrients such as nitrogen and phosphorus from the soil [4]. Fungi are also of great importance in biotechnology, e. g. in the production of drugs and enzymes [5], [6]. In addition, many fungi can be easily cultured and are amenable to microbiological, genetic, and molecular techniques. Therefore, fungi were some of the earliest model organisms for the study of genetics, biochemistry, cell and developmental biology. It is thus not surprising that the first eukaryotic organism for which a complete genome sequence was obtained is a fungus, the budding yeast Saccharomyces cerevisiae [7]. Today, fungi are the eukaryotic group with the greatest number of completely, or nearly completely, sequenced genomes (http: //www. ncbi. nlm. nih. gov/genomes/leuks. cgi, [2]). This is not only owing to their ecological, medical, agricultural, biotechnological and economic significance, but also due to the fact that with a size of 10–90 Mb and 4,700–17,000 predicted genes, fungal genomes are among the smallest and most compact eukaryotic genomes known. The sequences for almost all sequenced eukaryotic genomes have been obtained by conventional Sanger sequencing technology. Over the past five years “next-generation sequencing” techniques have revolutionized large-scale sequencing projects because of massively increased throughput, resulting in much reduced costs per base [8]. One major disadvantage of the current techniques is that none of them delivers read lengths that approach conventional Sanger technology: whereas Sanger sequencing routinely yields 900 nt, the longest next-generation reads obtained are in the range of ∼450 nt for Roche/454 pyrosequencing (from now on abbreviated as 454 sequencing), and the techniques with the highest throughput are with 36–80 nt still well below this. Short reads, e. g. as obtained by Illumina/Solexa sequencing (from now on abbreviated as Solexa sequencing) cause severe difficulties for the assembly of genome sequences that contain repetitive sequences, as is the case for many higher eukaryotes. Thus, next-generation sequencing techniques have so far mostly been used for the de novo sequencing of prokaryotic genomes or the re-sequencing of eukaryotic species with reference genomes, where the next-generation reads can be mapped on an existing genome sequence [8]–[11]. Recent improvements, e. g. paired-end sequencing (reads from matched ends of longer DNA fragments) and a steady increase in read length should make the de novo assembly of high-quality eukaryotic genomes possible. For example, the genome of the filamentous fungus Grosmannia clavigera was assembled from a combination of Sanger, 454, and Solexa sequence data [12] and a first draft of the 2. 4 Gb Giant Panda genome has been assembled from Solexa sequence reads alone [13]. Because of their small size, fungal genomes are perfectly suited for the task of optimizing de novo assembly approaches to generate high-quality or even finished larger eukaryotic genomes. Here, we present the de novo assembly and annotation of the genome sequence of the filamentous fungus Sordaria macrospora. The genome was sequenced solely by next-generation techniques (Solexa sequencing by synthesis and 454 pyrosequencing). S. macrospora is an ascomycete with a long-standing history as a model organism for fungal sexual development and meiosis [14]–[18] (Figure 1). Development of a large set of genetic tools for this fungus [19]–[24] resulted in the discovery of novel proteins involved in central events of meiosis and organogenesis [25]–[32]. Similar to its close relative Neurospora crassa, S. macrospora is haploid with a nuclear genome of seven chromosomes and an estimated 39. 5 Mb of DNA sequence [24], [33]–[35]. Previous studies found ∼90% nucleic acid identity within coding regions of orthologous genes from S. macrospora and N. crassa as well as a high degree of synteny over large genomic regions [36], [37]. Despite their close phylogenetic relationship, S. macrospora is homothallic (self-fertile) in contrast to the heterothallic (self-sterile) N. crassa. The natural habitat of S. macrospora is herbivore dung in temperate climates, whereas N. crassa is usually found on burned vegetation and the soil throughout the world [14], [38]–[41]. Thus, these two closely related fungi have evolved different life styles and inhabit different ecological niches. These differences may be at least partially reflected in their genomes. The S. macrospora genome sequencing project had two aims: (1) to assemble a first, high-quality draft of the genome sequence after next-generation sequencing to show that this approach is feasible for filamentous fungi in general, and (2) to annotate the genome sequence by a community effort, with the goal of a better understanding of S. macrospora biology and the idea of improving its value as a model organism for fungal development. The genome of the S. macrospora strain k-hell was sequenced by a combination of Solexa and 454 sequencing. First, a total of 3. 4 Gb of DNA sequence in 95,153,034 Solexa 36-nt reads were obtained from one single-read lane (9,688,226 reads), four lanes of paired-end reads (55,337,284 reads) from a 300-bp insert library, and three lanes (30,172,524 reads) of paired-end reads from a 500-bp insert library (Table 1, Figure S1). This represents 85-fold coverage of the S. macrospora genome. Assembly of the Illumina/Solexa data with the Velvet assembler [42] resulted in 38. 7 Mb of sequence data in 3,344 contigs with an N50 size of 51 kb (Table 2). As expected, these contigs contained a substantial number of internal gaps (17,956 gaps, Table 2), because paired-end data allows contigs to be scaffolded by inferred physical linkage of the matched pairs in the absence of contiguous coverage of intervening segments. Despite the internal gaps in some of the contigs, we decided not to call them scaffolds to differentiate between the Velvet output (referred to as contigs even when containing gaps) and a subsequent scaffolding step (see below). When compared with the N. crassa genome, we were able to map 8,350 of ∼10,000 predicted proteins to the 10,066 predicted N. crassa genes (e-value ≤10−20) which is only slightly lower than the number obtained with the final high-quality draft (8,519 proteins, see below). Thus, even this preliminary assembly covered most of the protein-coding genome. To close most gaps, we obtained additional sequence data by 454 sequencing. Because of longer reads, a relatively low coverage with 454 reads in combination with the previously obtained Solexa reads was expected to allow assembly with a higher N50 value and close internal gaps in the contigs. We obtained 415 Mb (∼10-fold coverage) of single-end 454 reads with an average read length of 367 bp (Table 1, Figure S1). Assembly of 454 reads only (with the Celera Assembler 5. 3; Eurofins MWG GmbH, Ebersberg, Germany) yielded 14,123 contigs (N50 size 11 kb; 1,681 internal gaps; Table 2). Gaps in this assembly were primarily caused by sequencing ambiguities. The combined raw data (Solexa and 454 reads) and the pre-assembled 454 data were used for constructing an assembly with the Velvet assembler version 0. 7. 31 [42] (Figure S1). This resulted in an assembly of 39. 9 Mb of sequence data (5,097 contigs with an N50 size of 117 kb) and only 624 internal gaps within the contigs (Table 2). Thus, the combination of Solexa paired-end reads with 454 reads resulted in an increase of the N50 value and a drastic reduction in the number of gaps compared to assemblies where each data set was used alone. With a size of 39. 9 Mb, this combined assembly corresponds well to previous analyses of the S. macrospora genome by pulsed-field gel electrophoresis that estimated the genome size at 39. 5 Mb [24]. To determine whether similar results might be obtained with fewer sequence reads, thereby further decreasing sequencing costs, we generated test assemblies with different combinations of coverage levels (Figure S2, Table S1). The addition of 454 reads had the most drastic effect on the number and length of gaps whereas addition of paired-end reads improved mostly the N50 value. The inclusion of fewer sequence reads resulted in suboptimal assemblies; however, at the number of reads used for our assembly, bench mark values were no longer changing dramatically, suggesting that a plateau had been reached where addition of this type of sequence reads did not significantly improve assemblies. Further improvement might be achieved by sequencing paired-end libraries with longer inserts. The genome sequence of the filamentous ascomycete Grosmannia clavigera was assembled from a combination of Sanger paired-end reads (0. 3-fold coverage), 454 single reads (7. 7-fold coverage), and Solexa paired-end reads (100-fold coverage) [12], resulting in a high-quality draft genome sequence of 32. 5 Mb with an N50 size of 164 kb. Our data show that similar values can be obtained even without including Sanger sequencing data thereby drastically decreasing sequencing costs. It has been previously demonstrated that several regions of up to 50 kb of the S. macrospora genome are syntenic to N. crassa [36], [37]. To extend this analysis to the newly assembled S. macrospora contigs, the five largest contigs from the Velvet assembly (519–991 kb) were compared to contigs of the N. crassa finished genome that have been assigned to specific linkage groups by mapped genetic markers (Assembly 9; http: //www. broadinstitute. org/annotation/genome/neurospora/Regions. html). The results were visualized as dot plot (Figure 2A), and show that each contig maps to one or two linkage groups with only one to three breaks of synteny. Thus, the high degree of synteny between S. macrospora and N. crassa that was expected from previous studies was reflected in the Velvet assembly. To make use of this high degree of synteny and further improve the S. macrospora assembly, we generated a comparative assembly with Mercator by using the scaffolded chromosomes of the draft N. crassa genome (assembly 7, [43]) and the draft-sequences of the Neurospora discreta (http: //genome. jgi-psf. org/Neudi1/Neudi1. home. html) and Neurospora tetrasperma (http: //genome. jgi-psf. org/Neute1/Neute1. home. html) genomes to order and scaffold the S. macrospora contigs [44]. This resulted in a total of 152 scaffolds and 4,629 contigs with an N50 size of 498 kb (Table 2). Syntenic regions between the S. macrospora and N. crassa genomes were analyzed by dot plot analysis (Figure 2B). To verify that the scaffolded contigs represent the correct order within the S. macrospora genome, three regions spanning gaps between contigs on scaffolds 17,58, and 98, respectively, were amplified by PCR and sequenced. In all cases, sequences between 0. 8 and 1. 2 kb were retrieved that close the gap between adjacent contigs thereby validating the scaffolding results (data not shown). This assembly represents the first high-quality draft version of the S. macrospora genome (“S. macrospora assembly 1”, acc. no. CABT01000001-CABT01004783, http: //gb2. fungalgenomes. org/gb2/gbrowse/sordaria_macrospora). Neither the rDNA repeat units nor the mitochondrial genome was represented in the Velvet assembly. We therefore searched the raw data as well as preassembled 454 and Solexa contigs for sequences with significant identity to rDNA or mitochondrial DNA from other fungi (Text S1). These reads were used to assemble both one rDNA unit as well as the mitochondrial DNA using CodonCode Aligner version 3. 0. 3 (http: //www. codoncode. com/aligner/). The rDNA unit shows ∼98% DNA sequence identity to that of N. crassa. Unlike in N. crassa, no additional smaller rDNA regions with point mutations were found by this method. Four shorter contigs had SNPs in various locations when compared to the full-length rDNA region. These SNPs all occurred as part of a homonucleotide run (4–6 nt), suggesting either sequencing errors or true polymorphisms in the rDNA repeats, which are considered to be rare in filamentous fungi but do exist in N. crassa because of the occurrence of RIP (see below; K. M. Smith and M. Freitag, unpublished data). The mitochondrial genome encompasses 88. 4 kb, and thus is larger than the 64. 8 kb mitochondrial genome of N. crassa and smaller than the 94. 2 kb mitochondrial genome of Podospora anserina. With 33. 6%, the GC content of the mitochondrial genome is in the same range as that of N. crassa (36. 1%) and P. anserina (29. 9%) (Text S1, Figure S3). Our data show that not only the single copy regions of the nuclear genome can be assembled from the next-generation sequencing data, but also multi-copy regions like the rDNA unit and the mitochondrial genome, even if they are not initially recovered in typical Velvet runs. Gene models for the first draft of the S. macrospora genome were predicted with four independent ab initio gene prediction programs trained on N. crassa and evidence-based predictions with N. crassa proteins (see Materials and Methods). The results were integrated with Evigan [45] to yield ∼12,000 gene models. Additionally, 455 tRNA genes were predicted, similar to the 424 tRNA genes predicted for N. crassa [43]. The initially predicted ∼12,000 protein coding genes were screened for ORFs with internal stops, lack of initiation or termination codons, unusually long introns and insufficient support by sequence similarity. Such ORFs were corrected or removed resulting in a refined gene set of 10,789 genes with an average length of 1,432 bp for all predicted coding sequences (CDS, Table 3, Table S2). The overall GC content of the genome is 52. 4%. This is changed to 56. 5% in coding regions, which represent 38. 4% of the genome, and 49. 8% in non-coding regions, which make up 61. 6% of the genome. To address the question of sequencing errors, we PCR-amplified and resequenced coding regions for six predicted genes (SMAC_01188, SMAC_01198, SMAC_6009, SMAC_07685, SMAC_07776, SMAC_09680) with frameshifts or internal stops. These were confirmed by resequencing in four cases, whereas in two cases, insertions or deletions of 1 nt were found in the assembled sequence which when corrected led to the prediction of functional open reading frames. In total, we tested 21 kb of coding sequence by resequencing and found four insertion/deletion errors (0. 02%). Although it is difficult to compare errors and error rates, this rate is similar to the 0. 1–0. 001% error rates achieved in microbial draft genomes sequenced by Sanger technology [46], [47]. With 10,789 predicted and partially curated genes, the gene count in S. macrospora is similar to that of N. crassa (10,066 community-annotated and centrally curated genes). To determine how many predicted proteins in these two closely related species are orthologs, reciprocal BLASTP analysis was performed: At an e-value of ≤10−20,8, 519 S. macrospora proteins have at least one homolog among the N. crassa proteins; vice versa, 8,179 N. crassa proteins have at least one homolog among the S. macrospora proteins. In total, 7,855 proteins (72. 7% of all S. macrospora proteins) have reciprocal best hits in both searches identifying them as likely orthologs (Table S3). Sequencing of the first few eukaryotic genomes revealed relatively high frequencies of “orphan genes” (i. e. genes without apparent homologs in any of the already known sequence databases and proteomes). As more genomes become available, this number has been rapidly decreasing, e. g. for N. crassa from ∼41% [43] to currently 22% (2,219/10,066 [48]). Because S. macrospora is more closely related to N. crassa than any other previously sequenced filamentous fungus, we compared the N. crassa orphan genes with the S. macrospora genome using TBLASTN and BLASTP to assess how many proteins are lineage-specific (Table S4). Of 2,112 N. crassa orphan genes that were retrieved from the current N. crassa MIPS protein list (http: //mips. helmholtz-muenchen. de/genre/proj/ncrassa/), 870 do not have significant hits in the S. macrospora genome at an e-value of ≤10−20. Orphan genes might comprise more quickly evolving genes [48], and we therefore repeated our analysis at an e-value ≤10−5. This analysis still left 471 (4. 7%) genes without significant hits, suggesting that these genes may constitute the remaining true orphan genes that separate the genus Sordaria from Neurospora (Table S4). The recent sequencing of additional Neurospora species is expected to further reduce the number of genus-specific genes. In addition to assessing the conservation of protein-coding gene regions, we sought to investigate the conservation of non-coding regions between S. macrospora and its closest relatives. Therefore, we performed comparisons of 5′ upstream regions in 1 kb blocks from 1 kb to 4 kb as well as comparisons of introns and coding regions for S. macrospora, N. crassa, N. discreta and N. tetrasperma (Figure 3, Figure S4 and Table S5). We observed that introns are more conserved than upstream regions. Among the upstream regions, pairwise identity is slightly but significantly higher in the 1 kb upstream regions than in any of the other tested upstream regions (Table S5). This suggests that most regulatory (and therefore putatively conserved) elements in 5′ UTRs and promoters reside within the 1 kb upstream regions. We also compared the predicted S. macrospora proteins to the non-redundant GenBank and Swissprot databases (Table S2). Approximately 6% (631/10,789) of all predicted proteins did not have a significant hit against the non-redundant database at an e-value ≤10−5. This number is only slightly higher than that for N. crassa (4. 7%, 471 genes, see above). Taking into account that no other Sordaria species have been sequenced yet, we suggest that the number of true orphan genes in ascomycetes might be less than 5% or 500 genes per genome. A search for conserved protein domains in the predicted S. macrospora proteins was performed with the HMMER program hmmpfam [49], [50] and with the InterProScan function from Blast2GO [51], [52]. With HMMER, one or more conserved domains were found in 5,471 predicted proteins (50. 7%, Tables S2 and S6), the InterProScan found domains in 7,099 predicted proteins (65. 7%, Table S2). These values might seem rather low when compared to the more than 10,000 proteins that have a hit in the non-redundant database, but it reflects the fact that many (predicted, hypothetical or conserved hypothetical) proteins have not yet been functionally characterized; therefore many domains remain to be identified. In addition to a comparison to N. crassa, an analysis of the predicted proteins from S. macrospora, N. crassa, N. discreta, P. anserina, and Chaetomium globosum was performed with OrthoMCL, a software that clusters orthologs and “recent” paralogs [53]. We identified 9,971 orthogroups, and among these 5,428 (54. 4%) comprise single genes from each of the five species, i. e. single-copy genes that are conserved among all species investigated (Tables S7 and S8). 31 orthogroups contain genes with three or more paralogs in S. macrospora, but fewer or no paralogs in other fungi, and these were investigated further. Some of these orthogroups contained proteins suggestive of transposon activity (see below), whereas others have no homology to transposons or pseudogenes. Phylogenetic analysis of two orthogroups (99 and 79) indicates evolutionary histories of ancient gene family expansion and subsequent differential gene loss (Figure 4). Orthogoup 99 comprises three genes from S. macrospora and two genes from P. anserina, whereas in the Neurospora species and C. globosum, only one gene is present. The genes from this orthogroup encode putative P450 oxygenases, and one might speculate that these proteins are beneficial for a coprophilic lifestyle, because only the coprophilic fungi S. macrospora and P. anserina have retained more than one copy. A similar case of duplication and subsequent loss can be postulated for orthogroup 79, which contains genes encoding chitin binding and glycosyl hydrolase domains. In contrast, orthogroups 49 and 180 contain one or no gene for the Neurospora species, P. anserina, and C. globosum, but six and four members, respectively, in S. macrospora; and the S. macrospora genes cluster together in a phylogenetic tree (Figure 4). Thus, these genes seem to represent recent duplication events in S. macrospora. Both orthogroups are part of larger gene families, and to verify that placement of these subfamilies in different orthogroups was correct, an independent phylogenetic analysis was performed (Figure S5). This analysis supports the grouping by OrthoMCL. To test whether these genes are expressed genes and not simply annotation errors, quantitative real time PCR experiments were performed for ten S. macrospora genes from orthogroups 49 and 180 (Figure 4). For eight of the ten genes, transcripts were found under conditions of sexual development and/or vegetative growth, and all eight genes are upregulated during sexual development. In contrast, the homologous N. crassa genes are downregulated or not differentially regulated (Figure 4). Thus, the S. macrospora genes that are expressed under the conditions investigated might have gained developmental regulation after the split of the Neurospora and Sordaria lineages, probably as a result of gene family diversification after gene duplications. Whether these genes have a function during sexual morphogenesis in S. macrospora remains to be determined. Transposons and repeat elements have been identified in all eukaryotic groups investigated so far, and they can comprise large portions of a genome, e. g. 85% of the recently published maize genome [54], [55]. In fungi they usually make up only a comparatively small part of the genome (usually ≤10%), because effective defense mechanisms against repeated sequences are in place and because smaller genomes are more streamlined [56]. Eukaryotic transposons can be divided into two classes, class I elements that transpose via an RNA intermediate, and class II elements that transpose at the DNA level by excision and reintegration [57]. To analyze the transposon content of the S. macrospora genome, several approaches were used. First, amino acid sequences of known transposon open reading frames were used for comparison with the predicted S. macrospora peptides as described previously [58]. Second, DNA sequences of randomly selected scaffolds were compared to Repbase data [59]. These two approaches will identify only those repeated sequences or transposons that are similar to previously described elements. Third, DNA sequences of randomly selected scaffolds were compared to the complete genome sequence in order to identify new repeated sequences without similar entities in the databases. Most interesting is the presence of five ORFs with amino acid sequence similarity to the N. crassa Tad LINE-like transposon [60] (Table 4). In addition, there are ∼20 ORFs with sequence similarity to gypsy-type retrotransposons. However, these ORFs exhibit rather diverse sequences and do not form element families. In contrast to these class I elements, there are only three ORFs with similarities to class II eukaryotic transposons; two of these represent a hAT-like element [61] that we called “Scarce”, and one ORF with amino acid similarity to the Fot1 transposon from Fusarium oxysporum [62]. As the only full-length Scarce ORF SMAC_09680 contains a nonsense codon, it is likely that the element is no longer active, thus explaining the low copy number. Overall, the transposon load of S. macrospora is very low, much more resembling that of another homothallic fungus Gibberella zeae (anamorph Fusarium graminearum) [63] than that of N. crassa. This is also reflected in a search for regions of high similarity within the S. macrospora genome by performing a BLASTN analysis of the genome sequence versus itself (Figure S6). In this analysis, the prevalence of regions with high intragenomic similarity in S. macrospora is between those of N. crassa [43] and F. graminearum [63], all of which have significantly fewer intragenomic regions of high similarity than the repeat-rich genome of Magnaporthe grisea [64]. This finding correlates well with the low transposon count of the S. macrospora genome. Taken together with the fact that we were able to assemble long contigs and that the assembly size correlates well with the genome size determined by pulsed-field gel electrophoresis, this suggests that the low repeat content is not an assembly artefact. In addition to class I and II transposable elements, five different non-coding short repeat sequences (Smini1 to Smini5,150–670 bp, Table 4) were detected. Two of these have partial sequence identity (Smini3 and Smini4) because of overlapping sequences. To verify that these repeats are real and not due to assembly problems, at least two copies for each repeat were PCR amplified and sequenced. For all tested repeats, their presence within the predicted genomic context was confirmed. Ten copies of repeat Smini5 are within ORFs, two are outside of ORFs, and another two overlap with ORFs. At least six of these ORFs show similarities to retrotransposon sequences. Some Smini1 and Smini3 repeats possess 5 bp target site duplications suggesting that they may be transposons or integrated elements caused by transposition. In some cases, point mutations may have modified target site duplications, or recombination may have occurred as has been shown for Aspergillus niger [65]. As Smini1 and Smini3 are both uniform in size (with the exception of some truncated elements), they may be solo-LTRs rather than mini-transposons such as the guest element of N. crassa [66]. Altogether, these five short repeat types cover only 56. 8 kb of the genome (0. 14%). In N. crassa and a few other ascomycetes (e. g. P. anserina [67], [68], M. grisea [69], F. graminearum [63], and Leptosphaeria maculans [70]), the RIP machinery detects pairs of repeated segments during premeiosis, introduces C∶G to T∶A mutations and can trigger DNA methylation of the mutated repeats in the vegetative cells resulting from ascospores, presumably by virtue of the increased AT content [71]. We analyzed the entire S. macrospora genome sequence for the presence of RIP footprints by calculating RIP indices [72] on the concatenated contigs and scaffolds (Figure S7). In contrast to the situation in N. crassa, where large regions mutated by RIP make up the centromeric DNA (K. M. Smith, L. R. Connolly and M. Freitag, unpublished data), we found no large blocks of AT-rich regions with the typical RIP bias (e. g. , TpA/ApT >1. 0). The only large region with atypical dinucleotide distribution was scaffold 0, which contains the mtDNA. Here, both TpA/ApT and (CpA+TpG) / (ApC+GpT) were close to 1, suggesting DNA composition more reminiscent of bacteria or budding yeast. Our results suggest the absence of large regions mutated by RIP in the S. macrospora genome. Previous analyses have shown that there is no active RIP in S. macrospora (Kück et al. , unpublished data). However, an ortholog of the N. crassa rid gene, the only gene known to be important for RIP [73], is present in the S. macrospora genome (Table S9), indicating that S. macrospora might have been able to undergo RIP during some time of its evolution; alternatively, RIP may occur at such low levels that it is difficult to detect in typical transformation and selfing experiments. RID homologs are involved in sexual development in two other fungi, Ascobolus immersus [74] and Aspergillus nidulans [75], suggesting that the S. macrospora protein may carry out a function independent of RIP. In N. crassa, two other genome defense mechanisms in addition to RIP have been identified, namely meiotic silencing by unpaired DNA (MSUD or “meiotic silencing”) and a form of RNAi (“quelling”) [76], [77]. All N. crassa genes identified in these processes have orthologs in S. macrospora suggesting that S. macrospora might be able to perform different varieties of genome defense (Table S9). The fact that endogenous genes can be silenced via introduction of transgenic constructs that result in double-stranded RNA molecules indicates an active RNAi-like mechanism [78]. Nevertheless, transformants with ectopically integrated copies for genes involved in meiosis (which might be subject of MSUD) or other processes (which might be subject to RNAi) have been successfully generated in different laboratories working with S. macrospora for years. Silencing of the resident and/or ectopically located gene functions has never been observed or described (e. g. [21], [25], [30], [79], [80]). This suggests that S. macrospora might possess gene silencing mechanisms but that they are perhaps less active, at least with respect to transgenes, than in N. crassa. Apart from genome defense mechanisms, there are a number of conserved processes in eukaryotes that are involved in maintaining genome integrity and regulating genome activity at the chromatin level [81]. We annotated chromatin-associated proteins, histone modification proteins, genes involved in the structural maintenance of chromosomes as well as centromere and kinetochore proteins and found that S. macrospora contains essentially the same set of genes as N. crassa (Table S9). Like its close relative, S. macrospora has single genes for the histone H3 K9 methyltransferase (DIM5), the heterochromatin protein 1 (HP1) and the DNA methyltransferase DIM-2, suggesting that heterochromatin formation and DNA methylation in S. macrospora are similar to what has been observed in N. crassa [43]. Taken together, these data indicate that S. macrospora contains the typical, conserved eukaryotic machinery for genome maintenance. Despite the absence of active RIP, this fungus appears to prevent the spreading of transposons and other repeated sequences as indicated by the low content of these elements within the genome. Since the 1950s, S. macrospora has been used as a model system for the analysis of fungal sexual development and meiosis, and a number of developmental genes have been characterized at the molecular level [14], [82]. We searched for genes known to be involved in development or in signaling cascades in S. macrospora and other fungi and found that S. macrospora contains homologs to all conserved genes as expected, further confirming the quality of the genome sequence. Specifically, we looked for orthologs to known genes for fungal sexual development, meiosis, GTP-, phospholipid- and calcium-signaling, motor proteins, senescence, photoreceptors and light signaling (Tables S10, S11, S12, S13, S14). In the case of photoreceptor-coding genes, it was found that S. macrospora contains homologs to known or putative fungal photoreceptors (Table S10). S. macrospora is able to undergo sexual development both in the dark as well as under white light [83]; however, in the light perithecial necks of Sordaria and Neurospora species exhibit positive phototropism in order to aim the active discharge of ascospores away from the growth substrate [84], [85]. In N. crassa, this photoresponse is mediated by the blue light photoreceptor WC-1 [84], [86]–[88]. Photoresponses often involve multiple photoreceptors, e. g. photoreceptors for red and blue light are present in one protein complex in A. nidulans [89], [90]. To test whether wavelengths other than blue light also play a role in regulating neck phototropism, we tested the photoresponse of S. macrospora to green and red light. Under red light, perithecial necks were oriented in random directions similar to that of perithecia grown in complete darkness, but perithecial necks showed a strong positive phototropism in response to green light (Figure S8). Our results suggest that perithecial neck phototropism in S. macrospora is regulated by blue light, similar to photoresponses in N. crassa [91], and additionally by green light, a response not yet observed in other fungi. The photoreceptors responsible for this phenotype remain to be uncovered; possible candidates are two putative rhodopsin-like green light photoreceptors (SMAC_02424 and SMAC_06025) that are orthologs of ORP-1 and NOP-1 in N. crassa, respectively [81], [92]. Senescence in fungi has been observed in the model organism P. anserina, in strains of N. crassa and N. intermedia [93], [94], but not in S. macrospora. A search for homologs to genes that are known to be involved in the aging process in P. anserina revealed that for the majority of the genes clear homologs are present in S. macrospora (Table S11). This includes genes that are required for mitochondrial protein quality control, programmed cell death, DNA repair, ROS scavenging, mitochondrial dynamics, and respiration, among other processes. Two genes not identified in S. macrospora are the apoptosis-related genes PaAif1 and PaAmid2. PaAIF1 (apoptosis-inducing factor) and PaAMID2 (AIF-like mitochondrion-associated inducer of death) are putative NADH oxidoreductases. In mammals, AMID is present in mitochondria, and its overexpression induces cell death [95]. The third protein that is missing in S. macrospora is the SAM-dependent O-methyltransferase PaMth1. An accumulation of this protein was detected in the mitochondria and in total protein extracts of senescent P. anserina wild type strains [96], [97]. Investigation of substrate-specificity of the protein hints to a protecting role of this methyltransferase against the generation of reactive oxygen species [98], [99]. While PaMth1 overexpressing strains show a significantly elongated life span, PaMth1 deletion strains are short-lived. However, S. macrospora does not show a restricted lifespan despite the lack of a PaMth1 homolog, indicating that the aging process in P. anserina is not conserved in other members of the Sordariales, and that the P. anserina aging genes that are present in S. macrospora may function in other cellular pathways. Fungi have long been used as model systems to study the molecular mechanisms of meiosis, and S. macrospora has played a prominent role in these investigations due to its simple sexual life cycle, large meiotic products (ascospores) and the production of an ordered tetrad of ascospores that allows the differentiation between pre- and postreduction segregation of alleles [14], [82]. Comparison of the predicted S. macrospora genes with the S. cerevisiae and Schizosaccharomyces pombe genomes [100], [101] allowed the identification of 92 “meiotic” genes. Reciprocal best hit BLASTP similarity searches against the predicted ORFs of S. macrospora, N. crassa and P. anserina showed that the 92 genes display orthologs in all three species (Table S15) [81], [102]. Nine of the genes were already characterized in S. macrospora (Table S15). The most conserved proteins include enzymes that are implicated in the recombination process and the proteins involved in sister-chromatid cohesion. In contrast, structural proteins like the components of the synaptonemal complex (SC) are poorly conserved despite the fact that the SC is as conserved during evolution as meiosis itself. This is similar to findings in other groups of organisms, e. g. mammals and plants [103]. Remarkably, S. macrospora, N. crassa, and P. anserina, like other filamentous fungi [104] possess only the RecA ortholog RAD51 and lack a recognizable DMC1, the meiosis-specific homolog of RAD51, thought to play an essential role in strand invasion [105]. The meiotic regulators are also poorly conserved (Table S15): among the three meiotic-specific transcription factors in yeast (Abf1p, Ume6p and Ndt80p) only an Ndt80p homolog is identifiable. Thus, S. macrospora has a conserved set of meiotic core genes whereas the regulators are more diverged, probably indicating life style-specific adaptations. We also searched for genes that may be involved in GTP-dependent and/or phospholipid or calcium signaling as well as known fungal developmental genes and genes encoding motor proteins, and found for all groups that the gene content of the S. macrospora genome is similar to that of N. crassa, and thus in most cases larger than that of S. cerevisiae (Tables S12, S13, S14). This shows that S. macrospora is a useful model organism for studying developmental processes because it contains the full repertoire of higher eukaryotic genes involved in signaling and regulatory networks. Nevertheless, there are several groups of genes where S. macrospora differs from other fungi and that warrant a closer look because they allow insights into fungal evolution and biology. These are described below. Two features in which S. macrospora differs from its close relative N. crassa are the lack of both asexual spores (“mitospores” or conidia) and heterokaryon incompatibility reactions. Searches in the S. macrospora genome for conserved genes that are involved in these processes revealed that homologs for conidiation genes are present (Table S16). These homologs seem to encode functional proteins, as they are not enriched in missense or nonsense mutations. Furthermore, quantitative real time PCR analysis for orthologs of six genes involved in conidiation in N. crassa revealed that these genes are expressed both during vegetative growth and sexual development in S. macrospora (Figure S9). Of course, additional unknown genes that are essential for conidiation may be missing or mutated in S. macrospora. Another possibility is that S. macrospora is able to conidiate, but does not do so under laboratory conditions. This would be analogous to the situation of Aspergillus fumigatus, which was recently shown to undergo sexual development when grown under suitable conditions [106], [107]. A third possibility, discussed below, might be that S. macrospora no longer produces conidia due to an unfavorable combination of heterokaryon incompatibility genes. Filamentous fungi can undergo hyphal fusion (anastomosis, [108]) between individuals of different genotypes leading to the formation of a mycelium containing genetically different nuclei (heterokaryon). In many ascomycetes such as N. crassa, P. anserina, and A. nidulans, the viability of these heterokaryons is genetically controlled by a set of heterokaryon incompatibility (het) loci. A het locus can be defined as a locus at which heteroallelism cannot be tolerated in a heterokaryon [109], thus a fusion between two individuals that differ genetically at a het locus results in a nonself recognition reaction which leads to phenotypes ranging from inhibited, abnormal growth to cell death [110]. Heterokaryon incompatibility (HI) has been shown to prevent the spread of viruses and the exploitation of aggressive phenotypes and is believed to reduce the risk of resource plundering between individuals [111]–[114]. However, heterokaryon formation can also have benefits for the individuals involved, e. g. the formation of functional diploids and mitotic genetic exchange in the parasexual cycle [115]. Several het loci have been characterized at the molecular level, and a conserved region of about 150 residues has been identified within various HI proteins. This domain is termed the HET domain [116]. The parts of het genes not encoding the HET domain are highly polymorphic; they ensure nonself recognition and are evolving very rapidly whereas the HET domain triggers cell death [117]. In addition to het domain genes, several other genes function as het loci, among them the mating-type genes in N. crassa, which act as het genes during vegetative cell fusion but are required to be different during sexual cell fusion [118]–[120]. Vegetative incompatibility has not been observed in S. macrospora [35]. Nevertheless, S. macrospora harbors genes for homologs to known het genes in other fungi (Table S17). A rather surprising finding was that in the case of het-c, pin-c, and a tol-related HET domain gene, not one, but two closely linked copies for each of these genes are present in the S. macrospora genome (Figure 5, Table S17). This is in contrast to all other filamentous ascomycetes which encode only one homolog of the het-c gene [121]. In addition to het-c, a second, closely linked HET domain-encoding gene named pin-c is essential for the HI reaction in N. crassa. It was shown that nonallelic genetic interactions between het-c and pin-c mediate nonself recognition while the severity of the HI depends on allelic interactions at the het-c locus [122]. In S. macrospora, the genomic region that is orthologous to the het-c/pin-c locus in N. crassa contains two copies of pin-c (SMAC_07217 and SMAC_07219) and one full-length (SMAC_07220) and one partial (SMAC_07218) copy of het-c (Figure 5). BLASTP comparison shows that the two PIN-C proteins from S. macrospora differ from each other to about the same degree as the N. crassa PIN-C allelic variants differ from each other (data not shown). The het-c/pin-c region is inverted in S. macrospora, and the genes at the ends of the inverted region, het-c and pin-c, are duplicated. To exclude the possibility that this is an assembly error, we amplified by PCR and end-sequenced DNA fragments spanning the regions between SMAC_07217 and SMAC_07218, between SMAC_07218 and SMAC_07219, between SMAC_07219 and SMAC_07220, and between SMAC_07228 and SMAC_07229. In all cases, we obtained PCR fragments of the expected size and sequence thereby validating that this gene order is not an assembly error but represents the wild type situation. Interestingly, the intergenic region between SMAC_07217 and SMAC_07218 contains two copies of the Smini1 repeat; thus, the duplication in this region may have originated from a transposition event. Phylogenetic analysis of the duplicated PIN-C homologs and the duplicated TOL-related proteins indicates that for pin-c, the duplication arose after the divergence of Sordaria from Neurospora, because the two pin-c copies are more similar to each other than to either of the three known pin-c alleles from N. crassa (Figure 5). In N. crassa, two copies of het-c are only present in one cytoplasm after heterokaryon formation, and it has been shown that HET-C proteins encoded by different het-c alleles form a heterodimer complex at the plasma membrane during the HI reaction [123]. Thus, with respect to het-c and pin-c, the genomic situation in S. macrospora resembles that of a heterokaryon in N. crassa (Figure 6), but no obvious signs of HI, e. g. compartmentalization and cell death, are evident in S. macrospora. However, mild HI reactions in N. crassa can lead to less severe phenotypes, e. g. aconidial strains [124]–[127]. In S. macrospora, the second het-c copy is incomplete and the ortholog of het-6, another gene involved in HI in N. crassa, contains internal stop codons so that a full HI reaction might be prevented by only partially functional het genes. Thus, we hypothesize that the lack of conidiation in S. macrospora may be due to “cryptic” or “mild” HI caused by the presence of more than one copy of putative HI genes in the genome (Figure 6). However, as indicated above, this is just one of several hypotheses to explain the fact that S. macrospora is aconidiate despite possessing orthologs to all known conidiation genes. Another point worth considering is that S. macrospora is homothallic and encodes mating type genes in one locus that are present in separate mating-type idiomorphs in N. crassa [128]. This situation would result in severe HI in vegetative cells of N. crassa mediated by the TOL protein. Only in tol mutants both mating type idiomorphs are tolerated in one vegetative cytoplasm [120]. Introgression of the N. crassa tol into N. tetrasperma caused HI and disrupted the pseudohomothallic nature of this fungus indicating that the native N. tetrasperma tol does not mediate HI [129]. Interestingly, the S. macrospora TOL, SMAC_08253, has only 40% amino acid identity to its N. crassa ortholog, an extremely low value compared to the average 89% identity in coding regions at the DNA level [37]. Probably this very divergent TOL does not mediate HI and allows co-existence of all mating type genes within vegetative cells. Thus, HI in S. macrospora might be attenuated (“cryptic” HI) or abolished by mutations in critical HI-mediating genes to cope with or allow the presence of otherwise incompatible genes within one genome. A second genomic locus that is important for HI in N. crassa and N. tetrasperma contains the het-6 and un-24 (rnr-1) genes. In this case, the two known alleles, Oak Ridge (OR) and Panama (PA), of both genes in both species differ not only in the sequences of the alleles, but also in the gene order within the het-6/un-24 locus, which was caused by an inversion of a block of five genes including un-24 [116], [130], [131]. An analysis of the orthologous region in S. macrospora revealed the same gene order as in the OR allele (Figure S10A). Phylogenetic analysis of both genes showed that the different allelic versions of N. crassa and N. tetrasperma cluster together as has been shown previously [131], while the S. macrospora genes occupy a basal position relative to the two Neurospora species (Figure S10B). This suggests that the OR allele represents the ancient gene order, and that the PA allele arose from an inversion after separation of Sordaria and Neurospora, but before speciation of N. crassa and N. tetrasperma; otherwise one would have to postulate two independent inversion events of the same genomic region leading to the OR gene order which is rather unlikely. Polyketides and non-ribosomal peptides are the most prominent classes of fungal secondary metabolites [6]. They comprise a wide variety of chemical structures, and a number of them have pharmaceutical applications, but their biological functions remain largely unknown [132], [133]. Most filamentous fungi harbor several genes encoding polyketide synthases (PKS) as well as non-ribosomal peptide synthases (NRPS) in their genomes. Apart from the pks and nrps genes, the biosynthesis of a polyketide or non-ribosomal peptide usually requires additional genes that encode, for example, enzymes that modify the products of the PKSs and NRPSs. These genes are often clustered together with the corresponding pks or nrps gene within the genome [134]. In order to determine the potential of S. macrospora for the biosynthesis of secondary metabolites, we searched the predicted proteins for the occurrence of typical domains associated with PKS or NRPS proteins, and additionally also for fatty acid synthase (FAS) proteins as these have structural similarity to PKS proteins (Table S18). S. macrospora contains three putative nrps genes, three genes that fall into the fas class, and eleven putative pks genes. The numbers of nrps and fas genes are the same as in N. crassa, and the corresponding genes in the two fungi are orthologs. However, of the predicted eleven pks genes, only seven have an ortholog in N. crassa, whereas four PKS proteins have a higher sequence identity to other, more distantly related fungi. The N. crassa genome contains only eight putative pks genes one of which has no ortholog in S. macrospora [36], [135]. Thus, with respect to pks genes and putative polyketides, S. macrospora appears to possess a greater potential for the production of secondary metabolites than its close relative N. crassa (Table S18, Figure S11). Most of the S. macrospora polyketide biosynthesis genes that have been studied previously have been found to be upregulated during sexual development, and polyketides may play a role in fruiting body formation in S. macrospora [36], [78]. Therefore, we determined the expression of the remaining five pks as well as the three nrps genes during sexual development (Figure 7). The nrps genes as well as eight of the eleven pks genes are transcriptionally upregulated during sexual development. The three pks genes that are not upregulated comprise the single type III pks gene as well as two pks genes without orthologs in N. crassa. These two pks genes, SMAC_01188 and SMAC_01198, are organized in a cluster of putative polyketide biosynthesis genes (Figure 8). Despite the fact that polyketide biosynthesis genes are often clustered in filamentous fungi [134], in S. macrospora only one such cluster has been found [36], and the genome sequence shows that most pks genes of S. macrospora do not occur clustered with other polyketide biosynthesis genes. Apart from them being clustered, the two pks genes SMAC_01188 and SMAC_01198 are interesting because they do not have orthologs in N. crassa or any of the other sequenced Sordariomycete genomes (P. anserina, C. globosum, F. graminearum, M. grisea). This is true for most of the genes from the cluster spanning the region from SMAC_01188 to SMAC_01201 (Table S19). With the exception of SMAC_01192 and SMAC_01197, the clustered genes do not have identifiable homologs within the Sordariomycetes, rather their most similar homologs are found within the Eurotiomycetes (Aspergillus, Neosartorya, Penicillium) or Dothideomycetes (Phaeosphaeria). In the center of the cluster, six genes are orthologs to genes from a putative polyketide biosynthesis cluster of Phaeosphaeria nodorum (Figure 8, syn. Stagonospora nodorum, http: //www. broadinstitute. org/annotation/genome/stagonospora_nodorum/Home. html [136]). There are two likely explanations for these findings: (1) the cluster originated through gene duplication in a common ancestor of the Sordariomycetes and Dothideomycetes, and later on, massive gene loss occurred in the Sordariomycetes with the exception of S. macrospora; (2) S. macrospora acquired the cluster through horizontal gene transfer (HGT). To examine these two possibilities, we determined the sequence identity between the S. macrospora cluster proteins and their orthologs in the P. nodorum cluster as well as the sequence identity between all homologous S. macrospora and P. nodorum proteins, and found that the sequence identity between the proteins from the cluster is significantly higher (Figure 8B). This is also the case when looking at the sequence identity of proteins with the same domains as the orthologs in the cluster. A phylogenetic analysis was performed with the cluster protein SMAC_01196 that encodes a putative phenylalanine ammonia lyase (PAL), a second PAL protein SMAC_05651 present in S. macrospora, and the homologs from seven other fungi (Figure 9). As expected, SMAC_05651 groups with the corresponding proteins from the Sordariales N. crassa, C. globosum, and P. anserina, each of which encodes only one PAL protein in their genomes. However, the “additional” PAL protein SMAC_01196 from the cluster groups among the Leotiomycetes/Dothideomycetes proteins and is closest to the P. nodorum cluster protein SNOG09914. Phylogenetic analysis of the cluster protein SMAC_01190 that encodes a putative member of the UbiA prenyltransferase family, its two other S. macrospora paralogs, SMAC_02313 and SMAC_06375, and the homologs from eleven other fungi gives a similar picture: SMAC_02313 and SMAC_06375 group within the Sordariales, whereas SMAC_01190 groups with the P. nodorum protein SNOG_09915 within a section of the tree that contains proteins from the Dothideomycetes, Eurotiomycetes, and Leotiomycetes, but not Sordariomycetes (Figure S12). The findings of (1) a conserved cluster of genes with closest homologs from the Dothideomycete P. nodorum instead of members of the Sordariomycetes, (2) the significantly higher sequence similarity between S. macrospora and P. nodorum proteins from the cluster compared to the overall sequence similarity between other proteins from these species, and (3) the phylogenetic positioning of two of the clustered proteins within the Dothideomycetes rather than the Sordariomycetes are more consistent with HGT than with the hypothesis of gene duplication and subsequent gene loss even though the latter cannot be excluded [137]. Recent studies, made possible by the increasing number of fungal genome sequences, have indicated that HGT may be more common in fungi than previously thought, and that genes for secondary metabolism are especially prone to HGT [138]–[141]. Even though in many cases “non-canonic” phylogenetic tree topologies can be explained by a combination of duplication, diversification, and differential gene loss [138], [142], that still leaves a number of cases where a HGT model best fits the observed data [140], [141], [143], [144]. HGT may be one way for fungi to increase their biochemical repertoire, thereby increasing their ability to adapt to new ecological niches [137]. In the case of the S. macrospora cluster presented here, it is interesting to note that it contains two putative pks genes (SMAC_01188 and SMAC_01198), one of which (SMAC_01198) has acquired 16 frame shifts/stop codons that interrupt the open reading frame whereas the other pks gene SMAC_01188 as well as the additional ten genes that comprise the putative polyketide biosynthesis cluster represent functional genes. For seven of the twelve genes from the cluster (SMAC_01188 to SMAC_01991, SMAC_01194, SMAC_01196 and SMAC_01198), transcriptional expression was verified by cDNA sequencing, and spliced cDNAs were obtained for all of the genes including SMAC_01198 which is unlikely to yield a functional protein due to the frameshifts (data not shown). Thus, this cluster might represent a case of an evolutionary recent acquisition that was introduced into the S. macrospora genome since its divergence from the last common ancestor with N. crassa. While part of the cluster appears to be retained and under purifying selection in S. macrospora, the gene SMAC_01198 has drifted and accumulated frameshift/nonsense mutations, even though it is still transcribed. Further analyses are necessary to determine the function of this putative polyketide biosynthesis cluster in S. macrospora. Due to their high throughput and low costs, next-generation sequencing techniques have greatly changed the way large-scale sequencing projects are done. This includes e. g. re-sequencing of existing genomes for the discovery of variations, “RNA-sequencing” for transcriptome analysis, or “ChIP-Seq” for the genome-wide analysis of DNA-protein interactions [8]. Until recently, de novo genome assembly from next-generation sequences has been restricted to prokaryotic genomes [10], [11]. This is due to the fact that eukaryotic genomes are larger and often contain high amounts of repetitive sequences that cannot be assembled from read lengths that are smaller than the length of the repeats. With the recent release of the Giant Panda genome [13] it has become obvious that even more complex eukaryotic genomes can be sequenced and assembled from short reads. Here, we present a high-quality draft of the S. macrospora genome, assembled solely from next-generation sequences, showing that de novo assembly from Solexa paired-end reads in combination with 454 sequence reads is feasible, cost-effective and fast, at least for compact eukaryotic genomes with few repetitive sequences. Additionally, the S. macrospora genome revealed several features that are of interest with respect to fungal evolution, namely its complement of het genes as well as polyketide biosynthesis genes. In the case of the closely linked het-c and pin-c genes, it was found that S. macrospora contains additional copies that might have arisen from inversion/duplication events. In other fungi, the presence of non-identical het alleles within one cytoplasm leads to HI, which in its extreme results in cell death [109], [115]. In contrast, S. macrospora is able to cope with this situation as no obvious HI phenotypes are observed in this fungus. However, we suggest that the aconidial phenotype of S. macrospora may be the result of “cryptic HI” caused by the presence of incompatible het genes within a single genome. Furthermore, analysis of a second het gene locus shows how the analysis of closely related genome sequences can help to pinpoint evolutionary events, in this case the occurrence of an inversion after separation of Sordaria and Neurospora but before speciation of N. crassa and N. tetrasperma. The analysis of predicted polyketide biosynthesis genes showed that S. macrospora contains more pks genes than its close relative N. crassa, and therefore probably has a wider biochemical repertoire available. One putative polyketide biosynthesis cluster might have been acquired through HGT, and this fits with previous results that show that HGT is probably rather widespread in fungi both for the transfer of single genes, clustered genes like polyketide biosynthesis genes, or even larger stretches of DNA up to whole chromosomes as was found in the phytopathogenic fungus Nectria haematococca [139]–[141], [143], [144]. These findings support the theory that HGT plays a role in fungal evolution and might be a source of genetic variation that allows fungi to adapt to different ecological niches [137]. The sequenced reference strain is Sordaria macrospora k-hell from the strain collection of the Department of General and Molecular Botany at the Ruhr-Universität Bochum. The strain was grown on cornmeal medium as previously described [38]. Genomic DNA from S. macrospora was prepared by following a modified previously published method [145]. Mycelium was frozen in liquid nitrogen, pulverized, and incubated in equal volumes of lysis buffer (0. 2 M sodium borate, 30 mM EDTA, 1% SDS, pH 9. 0) and phenol at 60°C for 5 min. After centrifugation, the supernatant was treated with RNase, and afterwards with an equal volume phenol/chloroform (1∶1). After centrifugation, genomic DNA was purified from the supernatant by cesium chloride density gradient centrifugation. To construct libraries of two different insert sizes, 5 µg DNA each were sonicated with a Branson sonicator. Sonicated DNA was separated through 2% NuSieve agarose gels and fragments of ∼300 and ∼500 bp were purified. After generation of blunt-end fragments, A-overhangs were added, adaptors ligated, and the fragments were PCR amplified [146]. The resulting libraries were sequenced on an Illumina Genome Analyzer with a paired-end module generating reads of 36 bases. Four lanes from the 300 bp library and three lanes from the 500 bp library resulted in 3. 4 Gb of sequence data (Table 1, Figure S1). Roche/454 sequencing was performed with 50 µg genomic DNA at Eurofins MWG GmbH (Ebersberg, Germany). This resulted in 415 Mb of sequence data with an average read length of 367 bp (Table 1). The 454 raw data were extracted from the sff file and converted to a fasta file using sff_extract. py (written by Jose Blanca and Bastien Chevreux, http: //bioinf. comav. upv. es/sff_extract/index. html). Assembly of the Solexa reads only as well as the combined Solexa and 454 reads was carried out with the Velvet assembler [42]. A description of the parameters used with Velvet can be found in Text S1 and Figure S1. An assembly of only the 454 data with the Celera Assembler 5. 3 was performed by Eurofins MWG GmbH (Ebersberg, Germany). Comparison of the S. macrospora genome with the N. crassa genome [43] was done with BLAST [147] and visualized with Combo [148]. Comparative assembly of the S. macrospora genome along the genome sequences of N. crassa, N. discreta (http: //genome. jgi-psf. org/Neudi1/Neudi1. home. html) and N. tetrasperma (http: //genome. jgi-psf. org/Neute1/Neute1. home. html) genomes was done with Mercator [44]. Assembly of the mitochondrial genome and the rDNA unit was done with CodonCode Aligner version 3. 0. 3 (http: //www. codoncode. com/aligner/), details can be found in Text S1 and Figure S3. Gene models were predicted independently with the ab initio predictors AUGUSTUS, GeneMark+ES, SNAP, and the evidence-based predictor Genewise [149]–[153]. The ab initio SNAP and AUGUSTUS parameters were trained on all N. crassa gene models while GeneMark performs an iterative self-training procedure. The Genewise predictions were generated from N. crassa proteins aligned to the genome by first aligning the proteins with TBLASTN, choosing the S. macrospora locus with only the best alignment for each protein and then refining the alignment and splice-sites with Genewise. The processing of outputs from these tools was completed with custom scripts utilizing tools from the BioPerl toolkit [154]. The resulting GFF annotation from each of the prediction programs was used as input to Evigan, a program that integrates the four sources of gene evidence [45]. For each of the predicted proteins, the protein with the highest sequence identity in GenBank was determined using BLASTP [147] (Table S2). Additionally, putative domains were predicted with the HMMER (version 2. 3. 2) program hmmpfam using the hidden Markov models from the pfam database [49], [50] and with the InterProScan function from Blast2GO [51], [52]. The resulting data can be found in Table S8. Putative localization of the predicted proteins was determined with WoLF PSORT [155], putative signal peptides and signal anchors were predicted with SignalP 3. 0 [156], and transmembrane domains with HMMTOP [157] and TMHMM [158] (Table S2, Text S2). tRNAs were predicted using a combination of Infernal 1. 0, tRNAscan-SE, and TFAM 1. 0 [159]–[161]. Orthologous groups of genes among the five fungal species S. macrospora, N. crassa [43], N. discreta (http: //genome. jgi-psf. org/Neudi1/Neudi1. home. html), P. anserina [102], and C. globosum (http: //www. broadinstitute. org/annotation/genome/chaetomium_globosum, Chaetomium globosum Sequencing Project, Broad Institute of Harvard and MIT http: //www. broad. mit. edu) were identified with OrthoMCL [53]. Searches for transposons and repeat elements were done with BLAST [147] and by searches in Repbase (http: //www. girinst. org/) [59]. For comparison of different genomic regions (CDSs, introns and upstream regions) from S. macrospora, N. crassa, N. discreta and N. tetrasperma, a Mercator alignment [44] of the genome sequences was performed and the parts of the alignment corresponding to the genomic regions were used to compute pairwise identities and evolutionary distances. Only those upstream regions were used that do not overlap with a protein coding region, and each region was used only once even if it is upstream of two divergently transcribed genes to avoid double-counting. The sequence and annotation data are available under the accession numbers CABT01000001-CABT01004783. The sequence reads that were used for the assembly of the S. macrospora genome were submitted to the NCBI sequence read archive (accession number SRA010462). For comparison of vegetative growth versus sexual development, growth and harvesting of S. macrospora and N. crassa, RNA preparation, reverse transcription and quantitative real time PCR were as described previously [19], [162]. Multiple alignments were created in CLUSTALX [163] and trimmed with Jalview [164], and the same alignment was used for analysis by distance-matrix (DM), maximum parsimony (MP) or Bayesian methods. Phylogenetic analyses were made with PAUP version 4. 0b10 for Windows (D. L. Swofford, distributed by Sinauer Associates, copyright 2001 Smithsonian Institution) for DM and MP analyses, and with MrBayes [165], [166]. DM and MP analyses were performed as described using 10,000 bootstrap replicates, Bayesian analysis was performed with at least 250,000 generations [167]. Consensus trees were graphically displayed with TREEVIEW or Dendroscope [168], [169].

Fungi have immense impacts on ecosystems and affect many aspects of society. They are used as convenient organisms for fundamental research because their typically haploid genetics enable straightforward phenotyping of mutations and because most fungal cells can differentiate the entire organism. Fungi have compact genomes with few repetitive sequences, and their genomes should be much easier to assemble from short sequence reads than genomes of mammals or higher plants. To test this idea, we used Solexa and 454 sequencing to generate ∼4 Gb of raw sequence data from the filamentous fungus Sordaria macrospora. De novo assembly yielded 5,097 contigs. This assembly was improved by comparison with reference genomes of three closely related Neurospora species, resulting in placement of ∼40 Mb of genome sequence in 152 scaffolds. From comparisons of predicted proteins we conclude that S. macrospora carries a conserved set of genes for signaling and development, which should encourage its further use as a model organism for morphogenesis and meiosis. We demonstrate that de novo assembly of fungal genomes from short reads is cheap and efficient. Species that are not traditionally considered “model organisms” but await genome sequencing for comparative and functional genomics analyses are at last amenable to in-depth genome-wide analyses.

Abstract Introduction Results/Discussion Materials and Methods

computational biology/genomics microbiology

2010

De novo Assembly of a 40 Mb Eukaryotic Genome from Short Sequence Reads: Sordaria macrospora, a Model Organism for Fungal Morphogenesis

Class 1 integrons are widespread genetic elements that allow bacteria to capture and express gene cassettes that are usually promoterless. These integrons play a major role in the dissemination of antibiotic resistance among Gram-negative bacteria. They typically consist of a gene (intI) encoding an integrase (that catalyzes the gene cassette movement by site-specific recombination), a recombination site (attI1), and a promoter (Pc) responsible for the expression of inserted gene cassettes. The Pc promoter can occasionally be combined with a second promoter designated P2, and several Pc variants with different strengths have been described, although their relative distribution is not known. The Pc promoter in class 1 integrons is located within the intI1 coding sequence. The Pc polymorphism affects the amino acid sequence of IntI1 and the effect of this feature on the integrase recombination activity has not previously been investigated. We therefore conducted an extensive in silico study of class 1 integron sequences in order to assess the distribution of Pc variants. We also measured these promoters' strength by means of transcriptional reporter gene fusion experiments and estimated the excision and integration activities of the different IntI1 variants. We found that there are currently 13 Pc variants, leading to 10 IntI1 variants, that have a highly uneven distribution. There are five main Pc-P2 combinations, corresponding to five promoter strengths, and three main integrases displaying similar integration activity but very different excision efficiency. Promoter strength correlates with integrase excision activity: the weaker the promoter, the stronger the integrase. The tight relationship between the aptitude of class 1 integrons to recombine cassettes and express gene cassettes may be a key to understanding the short-term evolution of integrons. Dissemination of integron-driven drug resistance is therefore more complex than previously thought. Integrons are natural genetic elements that can acquire, exchange and express genes within gene cassettes. The integron platform is composed of a gene, intI, that encodes a site-specific recombinase, IntI, a recombination site, attI, and a functional promoter, Pc, divergent to the integrase gene [1] (Figure 1). Gene cassettes are small mobile units composed of one coding sequence and a recombination site, attC. Integrons exchange gene cassettes through integrase-catalyzed site-specific recombination between attI and attC sites, resulting in the insertion of the gene cassette at the attI site, or between two attC sites, leading to the excision of the gene cassette (s) from the gene cassette array [2]–[6]. Multi-resistant integrons (MRI) contain up to eight gene cassettes encoding antibiotic resistance. To date, more than 130 gene cassettes have been described, conferring resistance to almost all antibiotic classes [7]. MRI play a major role in the dissemination of antibiotic resistance among Gram-negative bacteria, through horizontal gene transfer [8]. Five classes of MRI have been described on the basis of the integrase coding sequence, class 1 being the most prevalent [8]. Gene cassettes are usually promoterless, and their genes are transcribed from the Pc promoter, as in an operon (Figure 1), the level of transcription depending on their position within the integron [9], [10]. Among class 1 MRIs, several Pc variants have been defined on the basis of their −35 and −10 hexamer sequences. Four Pc variants have been named according to their sequence homology with the σ70 promoter consensus and their estimated respective strengths, as follows: PcS for ‘Strong’, PcW for ‘Weak’ (PcS being 30-fold stronger than PcW), PcH1 for Hybrid 1 and PcH2 for Hybrid 2, these two latter Pc variants containing the −35 and −10 hexamers of PcW and PcS in opposite combinations (Table 1), and having intermediate strengths [11]–[13]. More recently, a new variant was reported to be significantly stronger than PcS [14], and we therefore named it ‘Super-Strong’ or PcSS. Three other Pc variants have been described but their strength has not been determined; for simplicity, we named these Pc promoters PcIn42, PcIn116 and PcPUO, as they are carried by integrons In42 and In116 and by plasmid pUO901, respectively [15]–[17]. Nesvera and co-workers found a C to G mutation 2 bp upstream of the −10 hexamer in PcW and showed that this mutation increased promoter efficiency by a factor of 5 [18]. This mutation creates a ‘TGN’ extended −10 motif that is known to increase the transcription efficiency of σ70 promoters in E. coli [19]. Also, class 1 integrons occasionally harbor a second functional promoter named P2, located in the attI site and created by the insertion of three G residues, optimizing the spacing (17 bp) between potential −35 and −10 hexamer sequences [9] (Figure 1). Given the diversity of Pc variants and the range of their respective strengths, an identical array of gene cassettes should be differently expressed depending on the Pc variant present in the integron platform. However, the distribution of Pc variants among the numerous class 1 integrons has never been comprehensively studied. In class 1 MRIs, the Pc promoter is located within the integrase coding sequence (Figure 1). Some of the base substitutions in the −35 and/or −10 hexamer sequences defining the different Pc variants actually correlate with amino acid changes in the IntI1 sequence. These variations in the IntI1 protein sequence could potentially influence integrase recombination activity and define different IntI1 catalytic variants. We first performed an extensive in silico examination of all class 1 integron sequences available in databases in order to determine the prevalence of Pc variants and, therefore, the prevalence of IntI1 variants. We then estimated the strength of all Pc variants and Pc-P2 combinations in the same reporter gene assay, as well as the excision and integration activity of the main IntI1 variants. We found a very unequal distribution of the Pc variants, and a negative correlation between the strength of the Pc variant and the recombination efficiency of the corresponding IntI1 protein. We analyzed the sequences of 321 distinct class 1 integrons containing the complete sequences of both gene cassette arrays and Pc-P2 promoters (see Materials and Methods). When considering only the −35 and −10 hexamer sequences, we found no more than the eight variants identified previously. However, their distribution was highly uneven, four variants (PcW, PcS, PcH1 and PcH2) totalling 98. 4% of the sequences analyzed (Table 1). The most frequent Pc variant was PcW (41. 7%), followed by PcH1 (28%), PcS (24. 3%) and PcH2 (4. 4%). The four other Pc variants, all more recently described, were extremely rare (Table 1). The most prevalent Pc variant among class 1 integrons appeared to be the weak PcW, but in 58% of the analyzed PcW-containing integrons this promoter was associated with either a ‘TGN’ extended −10 motif [20] (hereafter designated variant PcWTGN-10) or the second gene cassette promoter P2 (Table 2). These two features were much less frequent with the other Pc variants (Table 2). The dataset also contained two other extremely rare Pc configurations, designated PcWTAN-10 and PcH1TTN-10, in which the second base upstream of the −10 hexamer was replaced by an A or a T instead of C, respectively, as well as two other rare forms of P2, designated P2m1 and P2m2, for ‘P2 mutated form 1’ and ‘P2 mutated form 2’ (Table 1 and Table 2). Altogether, on the basis of the −35 and −10 hexamers and the sequence upstream of the −10 box, we identified 13 Pc variants, four of which were also found associated with a form of the P2 promoter (Table 2). Until recently, the promoter strength of only 4 of the 8 known variants (PcSS, PcS, PcH1 and PcW) had been estimated, but variants strength had never been compared in the same assay [11], [14]. We therefore examined the capacity of all the Pc variants and the different Pc-P2 configurations to drive the expression of the lacZ reporter gene cloned in a transcriptional fusion with a 254-bp fragment containing the Pc variant and the P2 promoter region (see Materials and Methods). We found, in agreement with the results of a previous study [11] and those of another study published during the course of this work [13], that PcS was about 25-fold stronger than PcW and 4. 5-fold stronger than PcH1, while PcH2 lay between PcH1 and PcS, being 3. 8-fold stronger than PcH1. PcPUO and PcIn42 were of similar strength to PcW, and PcIn116 was very weak (Figure 2A). The PcSS variant, previously described as being stronger than PcS [14], was about 12-fold less efficient in our experimental conditions (Figure 2A). This latter result was not wholly unexpected, as PcSS contains a down-promoter mutation in the −35 hexamer relative to PcS (Table 1; [21]). We found that the presence of the TGN-10 motif increased PcW efficiency 15-fold, approaching that of PcH2, whereas it had no significant effect on PcS or PcH2 activity (Figure 2B), probably because these promoters are already maximally efficient. On the other hand, the C to A mutation in PcWTAN-10 severely reduced PcW activity (as already observed for the activity of an Escherichia coli promoter [19]), and the C to T mutation in PcH1TTN-10 slightly increased PcH1 efficiency (1. 7-fold; data not shown). To evaluate the contribution of P2 to gene cassette expression, we first created transcriptional lacZ fusion with sequences containing a combination of an inactive PcS (hereafter named PcS*, see Materials and Methods) and the P2 variants, in order to assess their specific strength. We found that P2 was active and 7-fold stronger than PcW (Figure 2C), in keeping with previous studies [11]. P2m1 and P2m2 appeared to be inactive (data not shown) and their influence on gene cassette expression was not investigated further. When the weakest Pc variants (PcW and PcH1) were associated with P2, β-galactosidase activity was increased but was equivalent to that of P2, indicating that, in the PcW-P2 and PcH1-P2 combinations, PcW and PcH1 do not contribute significantly to the expression of gene cassettes, which is mainly driven by P2. By contrast, when P2 was associated with the strongest variants, PcS and PcWTGN-10, β-galactosidase activity decreased slightly (Figure 2C). A recent report described a small increase in the expression of a gene cassette when PcS was combined with P2 [13], but these authors used different methods to measure promoter strength, which may explain the discrepancy with our results. In class 1 MRIs, Pc is located within the intI1 coding sequence, and several of the substitutions generating the different Pc variants affect the IntI1 amino acid (aa) sequence. The aa changes involve aa 32 or 39 for the main variants and aa 31,32,38 and/or 39 for the rare variants (Table 3). Some Pc variants produce the same IntI1 variant, e. g. PcW/PcH1 and PcS/PcH2 (Table 3). Altogether, 10 IntI1 variants are generated from 13 Pc variants, three of which (IntI1R32_H39, IntI1R32_N39 and IntI1P32_H39) represent almost 96% of the IntI1 variants (Table 3). In order to estimate the impact of the aa differences on IntI1 activity, we first cloned the intI1 gene of the three main IntI1 variants, IntI1R32_H39, IntI1R32_N39 and IntI1P32_H39, under the control of the arabinose-inducible promoter ParaB (see Materials and Methods). However, we anticipated that the two convergent promoters, namely Pc (contained in the intI sequence) and ParaB, might interfere with each other. Thus, to estimate IntI1 protein recombination activity independently of potential promoter interference, we introduced mutations that inactivated the Pc promoters without affecting the IntI1 aa sequence (see Materials and Methods). The resulting integrases were named IntI1*R32_H39, IntI1*R32_N39 and IntI1*P32_H39 (Table 3). We then estimated the excision activity of these integrases by measuring their capacity to catalyze recombination between two attC sites located on a synthetic array of two cassettes, attCaadA7-cat (T4) -attCVCR-aac (6′) -Ib, and resulting in the deletion of the synthetic cassette, cat (T4) -attCVCR, and the expression of tobramycin resistance mediated by the gene aac (6′) -Ib (see Materials and Methods; [22]). As shown in Figure 3, the three integrases exhibited very different excision activities (1. 8×10−2 to 1. 3×10−5), IntI1*P32_H39 and IntI1*R32_N39 being respectively 336- and 51-fold less efficient than IntI1*R32_H39. Thus, replacing R32 by P32, or H39 by N39 drastically reduces the capacity of the integrase to promote recombination between the attCaadA7 and attCVCR sites. The strongest effect was observed when a proline was present at position 32. P32 is also found in the integrase IntI1P32_N39, a much less frequent variant of IntI1 (Table 3). We therefore created this latter IntI1 variant and measured its excision activity. IntI1*P32_N39 was 27-fold less active than IntI1*R32_N39, showing the same negative effect of P32 on excision activity (Figure 3). Class 1 integrase is also able to catalyze the integration of gene cassettes by promoting recombination between attI and attC sites [5]. We therefore tested the ability of the different IntI1 variants to catalyze recombination between attI and the two attC sites used for the excision activity assay (attCaadA7 and attCVCR), in an assay based on suicide conjugative transfer previously developed [6] and since extensively used [23]–[25] (see Materials and Methods). Surprisingly, the range of integration activity of the four IntI variants tested in this study was rather narrow (4. 5×10−3 to 2. 3×10−4) compared to their excision activity, independently of the nature of the attC site (Figure 3). IntI1*R32_H39 and IntI1*R32_N39 exhibited similar integration activities in the two reactions performed, and the R32P substitution appeared to be detrimental for the activity of both integrases, but far less than for their excision activity. This effect seemed a bit stronger with IntI1*P32_N39 than with IntI1*P32_H39 (integration frequency was reduced by roughly 8-fold compared to 3-fold, respectively; Figure 3). To show that the observed differences in excision and integration activities of the four integrases tested were not due to variations in the amounts of integrase but indeed to the nature of the aa at positions 32 and 39, we performed SDS-Page western blot analysis. We found that IntI1*R32_H39, IntI1*R32_N39 and IntI1*P32_N39 were equally produced and that IntI1*P32_H39 was slightly more strongly expressed in our experimental conditions (Figure S1 and Text S1). However, the latter had one of the weakest recombination activities (Figure 3). Therefore, the observed differences in excision activity among the IntI1 variants were due not to differences in protein abundance but to differences in protein activity and/or folding. In this study we found marked polymorphism of the gene cassette promoter Pc (13 variants), corresponding to ten variants of the class 1 integrase IntI1. The 13 Pc variants were defined on the basis of the −35 and −10 hexamers and the sequence upstream of the −10 box. Indeed almost 20% of the 321 integrons analyzed here harbored a TGN-10 motif that characterized an extended −10 promoter. This feature was mainly associated with the weak PcW variant (41. 8% of PcW-containing integrons) and increased the efficiency of this promoter by a factor of 15. In view of its frequency and its strength difference relative to PcW, we propose that this promoter, designated PcWTGN-10, be considered as a Pc variant distinct from PcW. Furthermore, 9% of the 321 integrons contained the P2 promoter, which was almost exclusively associated with the PcW variant (17. 2% of PcW-containing integrons, Table 2). As in previous studies, we found that transcriptional activity was mainly driven by P2 in the PcW-P2 combination [9], [11]. We also observed the same effect with PcH1. Altogether, there are no fewer than 20 distinct gene cassette promoter configurations for class 1 integrons, but their frequencies are very different. Five main combinations emerged from the dataset, defining five levels of promoter strength. The distribution and strength of the gene cassette promoters were as follows: PcW-P2<PcW≈PcWTGN-10<PcS≈PcH1 (distribution, Table 2) and PcW<PcH1<PcW-P2<PcWTGN-10<PcS (respectively 4. 5-, 7-, 15- and 25-fold more active than PcW; Figure 1 and Figure 2). The multiplicity of gene-cassette promoters displaying different strengths indicates that a given antibiotic resistance gene cassette will be differently expressed depending on which Pc variant is present in the integron. For example, we used an E. coli strain containing a class 1 integron with PcW, PcS or PcWTGN-10, and with aac (6′) -Ib as the first cassette. The tobramycin MIC was 8-fold higher when the cassette was expressed from PcS or PcWTGN-10 than from PcW (data not shown). Our findings indicate that, in class 1 integrons, gene cassette expression is mainly controlled by the strongest Pc variants (PcS, PcH2, PcWTGN-10 and PcW-P2, in 55% of cases). Another important and previously unnoticed feature of class 1 integrons is the variability of the IntI1 primary sequence linked to the diversity of Pc variants. Among the 10 IntI1 variants identified, three (IntI1R32_H39, IntI1R32_N39 and IntI1P32_H39) accounted for almost 96% of class 1 integrases (Table 3). We found that these three main IntI1s displayed similar integration efficiencies, independently of the attC sites tested, whereas they had extremely different excision activities, depending on the nature of the amino acid at position 32 and/or 39. The R32P and H39N substitutions each drastically reduced the capacity of the integrase to promote recombination between the attCaadA7 and attCVCR sites (by 336- and 51-fold, respectively). In the integrase of the Vibrio cholerae chromosomal integron VchIntIA, the aa found at the position equivalent to residue 32 is basic, while the aa at position equivalent to residue 39 is a histidine (K21 and H28, respectively [24]), showing that, among IntI1 variants, IntI1R32_H39 is its closest relative. The crystal structure of VchIntIA bound to an attC substrate showed that these amino acids are located within an α-helix involved in attC binding [26]. This α-helix is conserved in the predicted structure of IntI1 and presumably plays the same role in recombination [24]. Thus, mutations of aa 32 and 39 in IntI1 might perturb the binding and thus undermine the recombination efficiency of attC×attC. The positively charged aa R32 may also play a role in the interaction with the attC site in the attI×attC recombination reaction. Indeed, a R32P substitution in both IntI1*R32_H39 and IntI1*R32_N39 reduced the integration frequency, but to a lesser extent than in an excision reaction (Table 3 and Figure 3). In contrast, aa H39 does not seem to be involved in the integration reaction. The attI×attC and attC×attC recombination reactions may thus involve different regions of the integrase. Indeed, Demarre and collaborators isolated two IntI1R32_H39 mutants, IntI1P109L and IntI1D161G, that showed much higher integration efficiencies [24]. Interestingly, we found a correlation between Pc strength and integrase excision activity: the weaker the Pc variant, the more active the IntI1. Among the four integrases tested, IntI1R32_H39, which was the most prevalent IntI1 in our dataset (Table 3), had the most efficient excision activity and also displayed higher excision than integration activity. Integrons with this integrase contain either the PcW variant, leading to a weak expression of the gene cassette array, or the PcH1 variant, associated with slightly higher expression (4. 5-fold). PcW-containing integrons could compensate for a low level of antibiotic resistance expression by the high excision efficiency of IntI1R32_H39, which confers a marked capacity for cassette rearrangement, in order to place the required gene cassette closer to Pc. In a recent study, Gillings et al suggested that chromosomal class 1 integrons from environmental β-proteobacteria might be ancestors of current clinical class 1 integrons [27]. The integrons they described all encoded IntI1R32_H39 and contained the PcW variant. We suspect that, under antibiotic selective pressure, these “ancestor” integrons may have evolved to enhance gene cassette expression, without modifying the potential for cassette reorganization, either through a single mutation (conversion of PcW to PcH1) or by the creation of a second promoter, P2, that is seven times more active. The high frequency of PcH1 (27. 3%) likely reflects its successful selection. P2 probably arises less frequently, as it requires the insertion of three G. We have recently shown that the expression of IntI1 is regulated via the SOS response, a LexA binding site overlapping its promoter [22]. Interestingly, when P2 is created, the insertion of three G disrupts the LexA binding site, probably leading to constitutive expression of IntI1. In a context of stronger antibiotic selective pressure, the need to express gene cassettes more efficiently could have led to the selection of more efficient Pc sequences (such as PcS and PcWTGN-10) at the expense of IntI1 excision activity, resulting in the stabilization of successful cassette arrays. This hypothesis is consistent with the observation that integrons bearing IntI1R32_N39 or IntI1P32_H39 tend to harbor larger gene cassette arrays than those bearing IntI1R32_H39 (Figure S2). The tight relationship between the aptitude of class 1 integrons to recombine and to express gene cassettes may be one key to understanding short-term integrase evolution. Different antibiotic selective pressures might select different evolutionary compromises. Thus, integron-driven drug resistance is more complex than previously thought. Compilation of the class 1 promoter sequences was performed in the entire Genbank nucleotide collection (nr/nt) using the alignment search tool BLASTn (http: //www. ncbi. nlm. nih. gov/BLAST) and the sequences of the intI1 and/or attI1 from the In40 integron as reference [28] (GenBank accession number AF034958). This data extraction was performed on 2009-02-01. Three other published but non-deposited sequences [14], [29], [30] were added to the 1351 sequences collected above. Of these 1351 sequences, only 434 contained both the Pc and P2 promoter sequences. Among the latter 434 sequences, we identified the integrons that displayed both identical gene cassette arrays and identical Pc/P2 sequences, independently of their bacterial origin. This analysis led to the isolation of 321 unique class 1 integron sequences that were further studied (Table S1). The bacterial strains and plasmids are listed in Table 4. Cells were grown at 37°C in brain-heart infusion broth (BHI) or Luria Bertani broth (LB) supplemented when necessary with kanamycin (Km, 25 µg/ml), ampicillin (Amp, 100 µg/ml), tobramycin (Tobra, 10 µg/ml), chloramphenicol (Cm 25 µg/ml), DAP (0. 3 mM), glucose (1%), arabinose (0. 2%). Mutations of the Pc and P2 promoter sequences were generated by assembly PCR with overlapping primers that contained the desired mutation and two external primers, int4b and ΔORF11 (Table 5). The two primary PCR products were then used in an equimolar ratio as templates for a second PCR step with the two external primers. Each transcriptional fusion plasmid was transformed into E. coli strain MC1061 to measure β-galactosidase enzyme activity. Assays were performed with 0. 5-ml aliquots of exponential-phase cultures (OD600 = 0. 6–0. 8) as described by Miller [31] except that the incubation temperature was 37°C. Experiments were done at least 5 times for each strain. A synthetic array of two cassettes attCaadA7-cat (T4) -attCVCR-aac (6′) -Ib preceded by the lac promoter is carried on plasmid p6851. This construction confers chloramphenicol resistance from the cat gene encoding chloramphenicol acetyltransferase from Tn9, here followed by a phageT4 rho-independent terminator, to prevent transcriptional read-through. The excision assay is based on the capacity of the integrase to catalyze recombination between the attC sites, resulting in the deletion of the synthetic cassette cat (T4) -attCVCR and expression of the tobramycin resistance gene aac (6′) -Ib from the lac promoter [22]. IntI1 proteins were expressed from the pBad-intI1* plasmids. A stationary-phase liquid culture of E. coli strain MG1656, carrying both p6851 and one of the pBad-intI1*, grown over-day in LB broth supplemented with antibiotics and glucose, was diluted 100-fold in LB broth supplemented with antibiotics plus either glucose or arabinose and was grown overnight. Recombinants were selected on LB-Tobra plates. Excision frequency was measured by determining the ratio of TobraR to KmR colonies. The assay was based on the method described in [6] and since extensively used [23]–[25]. Conjugation is used to deliver the attC site carried onto a suicide vector from the R6K-based pSW family [32] into a recipient cell expressing the IntI1 integrase and carrying the attI site on a pSU38 plasmid derivative (all plasmids are listed in Table 4). Briefly, the RP4 (IncPα) conjugation system uses the donor strain β2163 and the recipient MG1656, which does not carry the pir gene, and thus cannot sustain replication of pSW plasmids after conjugation. Recombination between attI and attC sites within the recipient cell leads to the formation of cointegrates between pSW and pSU38 plasmid. The number of recipient cells expressing the pSW marker (CmR) directly reflects the frequency of cointegrate formation. IntI1 proteins were expressed from the pBad-intI1* plasmids. Conjugation experiments were performed as previously described [5]. Integration activity was calculated as the ratio of transconjugants expressing the pSW marker CmR to the total number of recipient KmR clones. attC-attI cointegrate formation was checked by PCR with appropriate primers (primers 35 and 36; Table 5) on two randomly chosen clones per experiment. Background values were established by using recipient strains containing an empty pBad in place of the pBad-intI1*, and were 6×10−7 and 6×10−8 for the attI×attCVCR and attI×attCVCR assays, respectively. At least five experiments were performed for each recombination assay.

Integrons are widespread bacterial genetic elements able to capture and express gene cassettes that often encode antibiotic resistance determinants. Gene cassettes are usually promoterless and are transcribed from a common promoter, Pc. Pc is located within the coding sequence of the integron integrase, IntI, which is the key element catalyzing the integration and excision of gene cassettes. Several Pc variants, associated with different integrase amino acid sequences, have been described, but the influence of these differences on integrase activity has never been investigated. Here, we show that Pc is highly polymorphic, conferring a wide range of antibiotic resistance. Furthermore, we found that different Pc variants are associated with different integrase excision activities: the weaker the Pc variant, the more active the integrase. These results point to evolutionary compromises between the expression and mobility of drug resistance determinants located on integrons.

Abstract Introduction Results Discussion Materials and Methods

microbiology/microbial evolution and genomics microbiology/medical microbiology infectious diseases/antimicrobials and drug resistance

2010

Inverse Correlation between Promoter Strength and Excision Activity in Class 1 Integrons

Dengue is a major public health problem worldwide and continues to increase in incidence. Dengue virus (DENV) infection leads to a range of outcomes, including subclinical infection, undifferentiated febrile illness, Dengue Fever (DF), life-threatening syndromes with fluid loss and hypotensive shock, or other severe manifestations such as bleeding and organ failure. The long-standing World Health Organization (WHO) dengue classification and management scheme was recently revised, replacing DF, Dengue Hemorrhagic Fever (DHF), and Dengue Shock Syndrome (DSS) with Dengue without Warning Signs, Dengue with Warning Signs (abdominal pain, persistent vomiting, fluid accumulation, mucosal bleeding, lethargy, liver enlargement, increasing hematocrit with decreasing platelets) and Severe Dengue (SD; dengue with severe plasma leakage, severe bleeding, or organ failure). We evaluated the traditional and revised classification schemes against clinical intervention levels to determine how each captures disease severity using data from five years (2005–2010) of a hospital-based study of pediatric dengue in Managua, Nicaragua. Laboratory-confirmed dengue cases (n = 544) were categorized using both classification schemes and by level of care (I–III). Category I was out-patient care, Category II was in-patient care that did not meet criteria for Category III, which included ICU admission, ventilation, administration of inotropic drugs, or organ failure. Sensitivity and specificity to capture Category III care for DHF/DSS were 39. 0% and 75. 5%, respectively; sensitivity and specificity for SD were 92. 1% and 78. 5%, respectively. In this data set, DENV-2 was found to be significantly associated with DHF/DSS; however, this association was not observed with the revised classification. Among dengue-confirmed cases, the revised WHO classification for severe dengue appears to have higher sensitivity and specificity to identify cases in need of heightened care, although it is no longer as specific for a particular pathogenic entity as was the traditional schema. Dengue is an arthropod-borne viral disease with increasing prevalence in the last three decades, resulting in serious public health problems worldwide. With no vaccine or specific treatment to mitigate the natural history of the disease, a tool that can help clinicians detect and provide timely treatment is of utmost importance. The traditional World Health Organization (WHO) classification for dengue was implemented since 1974, based on experience with pediatric dengue in Thailand, and was then revised in 1997 [1], [2]. It classifies dengue disease as Dengue Fever (DF), Dengue Hemorrhagic Fever (DHF), and Dengue Shock Syndrome (DSS). The case definition for DHF lists the presence of four criteria: fever, hemorrhagic manifestations, thrombocytopenia (platelets ≤100,000 cells/mm3), and evidence of plasma leakage (pleural effusion, ascites, hemoconcentration ≥20% or hypoproteinemia). In turn, DHF is divided into four grades (DHF I–IV), where Grades III and IV are DSS, with hypotensive shock or narrow pulse pressure plus clinical signs of shock. It has proved to be very useful, with an emphasis on volume replacement for hemodynamic instability. However, limitations have been noted regarding its complexity and applicability, particularly in patients with severe symptoms [3], [4], [5], [6], [7], [8]. The recognition of these limitations led the Tropical Disease Reseach branch (TDR) of the WHO in 2006–7 to sponsor a multicenter study in seven countries in Asia and Latin America [9], and from this study emerged a new classification schema [10]. The new classification is divided into Dengue without Warning Signs, Dengue with Warning Signs, and Severe Dengue. In this study, we evaluated the capacity of the traditional classification and the revised classification to detect severe cases of dengue, compared to standardized clinical intervention levels. This evaluation was performed with information from ∼550 laboratory-confirmed dengue patients 6 months to 14 years old seen at the National Pediatric Reference Hospital in Managua, Nicaragua, from 2005 to 2010. A cross-sectional study was performed in the Hospital Infantil Manuel de Jesús Rivera (HIMJR), the National Pediatric Reference Hospital, in Managua, Nicaragua. A total of 544 children who attended the HIMJR between July 2005 and January 2010 with laboratory-confirmed dengue were studied. These patients were between 6 months and 14 years of age, had fever or history of fever less than 7 days, and one or more of the following signs and symptoms: headache, arthralgia, myalgia, retro-orbital pain, positive tourniquet test, petechiae, or signs of bleeding. Patients with a defined focus other than dengue were excluded. Additional exclusion criteria included: a) children weighing less than 8 kg, b) children less than 6 months of age, and c) children 6 years of age and older displaying signs of altered consciousness at the time of recruitment. Patient data such as vital signs, clinical data, and radiographic or ultrasound results were collected on a daily basis by trained medical personnel until discharge. A blood sample was collected daily for a minimum of three days for Complete Blood Count (CBC) with platelets, blood chemistry, and diagnostic tests for dengue. Between days 14 and 21 after onset of symptoms, a blood sample was taken for convalescent follow-up. Hospital admission criteria for study participants is detailed in Text S1. Criteria for admission to the Intensive Care Unit (ICU) included patients with shock despite appropriate fluid management with crystalloids and colloids, patients requiring vasoactive amines, patients using a mechanical ventilator, or patients requiring continuous monitoring due to hemodynamic instability. Over the years, a few patients were not able to be admitted to the ICU despite meeting ICU admission criteria due to the lack of space in the ICU. The protocol for this study was reviewed and approved by the Institutional Review Boards (IRB) of the University of California, Berkeley, and of the Nicaraguan Ministry of Health. Parents or legal guardians of all subjects provided written informed consent, and subjects 6 years of age and older provided assent. All information was collected every 12 hours in Case Report Forms (CRFs) designed to follow the patients' progress. Each CRF was completed by an infectious disease pediatrician and supervised by a second physician. Following this supervision, the CRFs were systematically monitored and then their information was entered into an Access 2003 database by double-date entry, with quality control checks performed daily and weekly. Thus, all data were collected prospectively over the entire course of illness following the same protocol and were reviewed carefully in real time to minimize any missing data. The data were then analyzed by illness episode; there were no missing signs or symptoms by episode. A case was considered positive for dengue when laboratory tests met one or more of the following criteria: 1) Dengue viral RNA was detected by RT-PCR, 2) Dengue virus (DENV) was isolated, 3) Seroconversion of DENV-specific IgM was detected by MAC-ELISA in paired acute and convalescent samples, and 4) DENV-specific antibody titer by Inhibition ELISA [11], [12], [13] demonstrated a 4-fold or greater increase between acute and convalescent sera. Primary DENV infections were those in which acute antibody titer was <10 or convalescent antibody titer was <2,560, and secondary infections were those in which antibody titer was ≥10 (acute) or ≥2,560 (convalescent) as determined by Inhibition ELISA. The traditional WHO classification is defined as follows: Dengue Fever (DF), Dengue Hemorrhagic Fever (DHF), and Dengue Shock Syndrome (DSS), whereas the revised WHO classification consists of Dengue without Warning Signs, Dengue with Warning Signs, and Severe Dengue (Table 1). Regarding the traditional classification, laboratory-confirmed cases that met the case definition for dengue but did not comply with the criteria for DHF or DSS were classified as DF. With respect to the revised classification, we interpreted SD to include compensated shock based on dengue case management algorithms in the 2009 WHO Guidelines [10]. A computerized algorithm was developed to classify laboratory-confirmed dengue patients according to the traditional and revised classifications; this algorithm compiled the presence or absence of all signs and/or symptoms as well as results of clinical laboratory tests and radiography/ultrasound and thereby determined the level of severity according to each of the classifications (Table S1). The following definitions were used for each of the warning signs: Abdominal pain: abdominal tenderness and continuous pain (not intermitent), on some occasions diffuse. Persistent vomiting: more than three episodes of vomiting in 12 hours, preventing adequate oral hydration. Clinical accumulation of liquids: pleural effusion and ascites diagnosed clinically, confirmed with imaging techniques (ultrasound for ascites, gallbladder wall thickening, and pleural effusion, and/or X-rays for pleural effusion). Mucosal bleeding: bleeding gums or conjunctiva, epistaxis, vaginal bleeding, bleeding from digestive, respiratory or urinary system (kidneys); mucosa defined as respiratory, vaginal, digestive, conjunctival and urinary tract mucosa. Lethargy: evaluated as an alteration of consciousness with a Glasgow score less than 15 or a Blantyre score less than 5. Irritability: irritability or restlessness. Hepatomegaly: the liver edge palpated by the clinician more than 2 cm below the costal margin. Increased hematocrit concurrent with rapid decrease in platelet count: increase in hematocrit together with a decrease of >10,000 platelets/mm3 in 24 hours with respect to previous measurement or concurrent with platelet count ≤100,000 cells/mm3. The level of agreement for detection of severe cases of dengue between the traditional and revised classifications was determined using the Kappa index, as was the concordance between the clinical diagnosis by the physicians and the diagnosis generated by the computer algorithm. A Kappa value <0. 00 was considered “poor agreement, ” 0. 00–0. 20 as “slight agreement, ” 0. 21–0. 40 as “fair agreement, ” 0. 41–0. 60 as “moderate agreement, ” 0. 61–0. 80 as “substantial agreement, ” and 0. 81–0. 99 as “almost perfect agreement” [14]. To determine the sensitivity, specificity, and positive and negative predictive values of the traditional and revised WHO classifications for the detection of severe cases of dengue, each classification schema was compared to standardized clinical intervention levels. Three reference levels (gold standard) were established based on the type of intervention the patient required in accordance with the DENCO study sponsored by TDR [9]. Category I were those patients who were managed as outpatients (did not present criteria for hospitalization). Category II were hospitalized patients who received intravenous fluids for rehydration or maintanance and did not suffer organ damage. Category III were patients hospitalized in the Intensive Care Unit (ICU), administered inotropic drugs or ventilation, or who experienced organ failure. Patients classified as DHF or DSS were considered severe, and those classified as DF were considered non-severe. In the case of the revised WHO classification, patients classified as Severe Dengue were considered severe, and those with Dengue with or without Warning Signs were considered non-severe. With respect to the reference levels, severe dengue cases were considered as patients managed with Category III care. All data was stored in Microsoft Office Access version 2003 and analyzed using Stata Intercooled 9. 0 (StataCorp LP, College Station, Texas), with a 95% confidence level. According to the traditional WHO classification, the majority of the patients were classified as DF (385; 70. 8%), while the remaining 29. 2% were divided between DHF (106; 19. 5%) and DSS (53; 9. 7%). In contrast, with the revised scheme, although the majority of cases were classified as Dengue with Warning Signs (266; 48. 9%), a large percentage of patients were classified as Severe Dengue (242; 44. 5%) and only a small percentage (36; 6. 6%) were classified as Dengue without Warning Signs (Figure 1). The level of agreement between the traditional and revised classifications for the detection of severe cases of dengue was fair (kappa 0. 25, CI95% 0. 17–0. 32, p<0. 001) (Table 3), and the percentage of observed agreement (64. 1%) was somewhat higher than that expected by chance alone (52. 3%). In the traditional classification, the majority of DF cases were treated at Category II, as they were hospitalized and received some type of intravenous (IV) rehydration. However, it is striking that 28. 1% (108/385) of these patients received Category III intervention, as, despite being classified as DF, they showed severe clinical manifestations warranting ICU transfer or administration of inotropic drugs (Table 4). Ninety percent (48/53) of DSS cases were managed according to Category III care. However, most DHF cases (76. 4%; 81/106) were treated at Category II care, and as such, sensitivity for detecting severe cases of dengue (DHF/DSS) was low (39. 0%, CI95% 31. 8–46. 6) and specificity was moderate (75. 5%, 70. 7–79. 8) (Table 4). When the revised classification was compared with level of care (as the gold standard), 61. 1% (22/36) of patients with Dengue without Warning Signs were treated as outpatients (Category I). In 38. 9% (14/36) of remaining cases, 36. 1% (13/36) fell under Category II care, and only one patient classified as Dengue without Warning Signs received Category III care. Sixty-seven percent (163/242) of Severe Dengue cases corresponded to Category III care, although it is noteworthy that 8 children (3. 3%) of those classified as Severe Dengue were treated as outpatients (Category I). The fact that Severe Dengue cases mostly fall under Category III and non-severe cases of dengue (Dengue with and without Warning Signs) are categorized as Category I and II allows for a high level of sensitivity (92. 1%, 87. 1–95. 6) and a moderate level of specificity (78. 5%, 73. 9–82. 6) for the detection of severe dengue cases (Table 5). In this sense, the revised classification is more sensitive than the traditional classification, but equally specific. In 2009, the concordance between the clinician' s diagnosis and classification according to the algorithms was evaluated prospectively (n = 212). With respect to the traditional classification, physicians had no difficulty classifying patients with DF or DSS, which matched the diagnosis generated by the computer algorithm in 95. 6% (174/182), and 83. 3% (5/6) of cases, respectively. Difficulty was encountered in classifying DHF patients, however, where only 12. 5% (3/24) of cases were classified correctly, with 66. 7% (16/24) being diagnosed as DSS by physicians. This incongruence meant that the level of agreement between clinical diagnosis and the computerized classification algorithm was moderate (kappa = 0. 46,0. 38–0. 55, p<0. 001) (Table 6), with an observed agreement (85. 8%) higher than expected by chance (73. 6%). When comparing the physicians' clinical diagnosis with the computer-generated algorithm of the revised WHO classification, 72. 2% (13/18) of patients with Dengue without Warning Signs and 87. 8% (94/107) of patients with Severe Dengue were correctly classified. A lower percentage of cases were correctly diagnosed in those patients with Dengue with Warning Signs, where 66. 7% (58/87) of cases were correctly classified. The level of agreement was substantial (kappa = 0. 62,0. 53–0. 71, p<0. 001) (Table 7), with the observed agreement (77. 8%) much higher than the expected (41. 3%). In general, it was found that physicians had fewer difficulties classifying patients when using the revised classification scheme. During the study period, DENV-1, DENV-2, and DENV-3 circulated among the patients. Association of disease severity with serotype was investigated using the two classification schemes. Using the traditional WHO classification protocol, it was found that the proportion of DHF and DSS cases was significantly greater (p<0. 001, Fisher' s exact test) in patients with DENV-2 infections as compared to the other serotypes (28. 6% DHF and 22. 3% DSS in DENV-2 infections versus 15. 6% DHF and 11. 1% DSS in DENV-1 infections and 16. 4% DHF and 3. 8% DSS in DENV-3 infections) (Table 8). Likewise, DENV-2 was most associated with evidence of plasma leakage, such as ascites, pleural effusion, and gallbladder wall thickening (p<0. 001), as well as thrombocytopenia (p<0. 001) (Table S2). In contrast, no significant difference between the proportion of severe cases and serotype was observed when the revised classification was applied (p = 0. 104, Fisher' s exact test), with 51. 1%, 52. 8% and 40. 8% of severe dengue in patients with DENV-1, DENV-2, and DENV-3 infections, respectively (Table 8). This study shows that the sensitivity and specificity of the traditional WHO classification for the detection of severe cases of dengue was 39. 0% and 75. 5%, respectively, while the sensitivity and specificity of the revised classification was 92. 1% and 78. 5%, respectively. A fair level of agreement (kappa = 0. 25, p<0. 001) was observed between the traditional and revised classifications for detection of severe cases of dengue. Evaluation of physicians' clinical diagnosis resulted in moderate agreement (kappa = 0. 46, p<0. 001) with the traditional classification and substantial agreement (kappa = 0. 62, p<0. 001) with the revised classification. However, whereas the traditional classification demonstrated a significant association of DENV-2 infection with DHF/DSS, no such association with Severe Dengue was observed with the revised classification scheme. The traditional WHO classification includes two major entities, DF and DHF/DSS. This classification was largely based on experience with a pediatric population in Southeast Asia, though currently dengue has spread to other tropical and subtropical regions and clinical presentation of the disease has changed. Dengue varies widely in clinical manifestations, and the classification of severity therefore depends on the presence and detection of particular symptoms and signs. While the traditional classification requires the presence of four criteria to qualify as a case of DHF, situations have been observed where all four criteria are not present, resulting in problems with the classification and detection of severe cases. Indeed, many authors have reported difficulty in complying with the traditional classification for documenting clinical presentations of dengue such as hemorrhagic manifestations [4], [15], thrombocytopenia [15], [16], [17], [18], [19], and fluid leakage [4], [15], [20], [21], [22]. For instance, with respect to the latter, it is often difficult to demonstrate that hemoconcentration is ≥20%, as there are places where it is not possible to perform daily CBC; in addition, a physician-ordered intervention during the course of the illness, such as administration of intravenous fluids, can alter hematocrit levels and thus hemoconcentration [15]. Another complication is that few institutions in dengue-endemic countries have records of a normal hematocrit value for each patient; therefore, some investigators have used a population hematocrit value as a baseline or the hematocrit value during the convalescent phase or at discharge to define hemoconcentration via comparison with the highest hematocrit observed during the acute phase of the disease [4], [17]. Use of a population baseline enables increased documentation of plasma leakage, but may be less specific since DENV-negative cases can present with elevated hematocrit [17]; whereas using the convalescent hematocrit value as baseline requires retrospective classification of dengue cases. Despite the widespread recognition of the usefulness of the traditional classification, difficulties in documenting all of the clinical manifestations required to define severe cases of dengue has resulted in alternative designations of certain clinical presentations seen in dengue, such as “Dengue with Signs Associated with Shock” (DSAS) [13] and “Dengue with Severe Bleeding” (DFB) [8]. As a result of this situation, Bandyopadhyay [6] proposed the creation of a multicentric prospective study in various countries of Asia and Latin America to describe the varying clinical presentations of dengue and to determine whether revision of the traditional WHO classification was necessary. Such a study was conducted from 2006–2007, and from it emerged a revised proposal for dengue classification [9]. Three levels of care were used in this study as the reference or gold standard and were based on the type of care needed and the condition in which patients presented. Category III represented patients with a severe condition and served as the comparison for DHF/DSS cases in the traditional classification and Severe Dengue cases in the revised classification. This methodology [9] was used as a basis for the study reported here. In our study, 71% of patients were classified as DF, 19. 5% as DHF, and only 9. 7% as DSS according to the traditional classification, while only 6. 6% of patients were classified as Dengue without Warning Signs, 48. 9% as Dengue with Warning Signs, and 44. 5% as Severe Dengue according to the revised classification. The difference in the percentage of severe cases between the two classification schemes can be explained by the existence of 62 patients with hypotension for age who did not present with platelet count ≤100,000, hemoconcentration or hemorrhagic manifestations, and were thus classified as DF, and 101 patients with compensated shock who were also classified as DF using the traditional scheme. These cases can serve to explain the low sensitivity (39. 0%) of this classification scheme for detecting severe cases of dengue. The revised classification does not require the presence of the four criteria to determine severity, so the presence of shock, independent of thromobocytopenia or hemoconcentration, is sufficient for a case to be designated severe, and this explains the higher numbers of these cases within this group of patients. The high sensitivity of the revised classification (92. 1%) for detecting severe cases of dengue can be explained by the same reason, that is, the presence of a single criterion for defining a severe case. This feature allows better case capture and increased admission to health units, though it results in not all cases being truly “severe”, as expressed by a moderate positive predictive value (67. 4%). This may overload health units in countries such as Nicaragua where large numbers of patients are admitted, disease evolution is carefully observed and monitored, and patients are discharged slowly once cases of severe disease have been ruled out. This over-estimation of severe cases of dengue may overwhelm hospitals and health centers, particularly during outbreaks or periods of high incidence, thus resulting in overextension of medical personnel and resources of each unit, but would avoid deaths due to the disease. In our pediatric cohort study of dengue in Nicaragua, during the years 2004–2008, the percentage of patients with dengue who were transferred to the study hospital from our study health center varied between 11% and 36%, but during 2009, the year where the revised WHO dengue classification scheme was implemented in the cohort study, the percentage of transferred cases rose to 83% (A. Balmaseda, G. Kuan, E. Harris, unpublished data). The revised classification has a specificity of 78. 5% for detecting severe cases of dengue, which is virtually identical to that of the traditional classification (75. 5%), with the exception that the revised classification scheme has a significantly higher negative predictive value. This feature of the revised classification may allow the clinician to better discern a patient who does not have a severe case of dengue. From the treating physician' s viewpoint, the revised classification may be useful because it allows the patient to be classified and treated in real-time, that is, during their hospital stay, whereas with the traditional classification scheme, the majority of cases tended to be retrospectively classified so as to detect the presence of the four criteria that define severity (DHF/DSS). This is reflected in the observed difficulty in correctly classifying the patient according to the traditional classification schema, expressed by the low level of agreement with the clinician' s diagnosis (kappa = 0. 46). Ideally, one would hope for a classification that is both sensitive and specific for detection of severe cases of dengue, in order to avoid oversaturation of health units, especially at the secondary level, but neither of the two classification schemes evaluated here possess these characteristics. Choosing a highly sensitive test maximizes the capture of severe cases, but requires subsequent evaluation during the patients' hospitalization to determine their real condition. The traditional WHO classification allows characterization of the pathophysiology of severe cases of dengue as the syndrome of DHF/DSS. This focus on a particular syndrome is useful for investigating viral and immunological risk factors. For example, in this data set, DENV-2 was found to be significantly associated with DHF/DSS; this information is useful for clinicians and public health officials to keep in mind for future epidemics, as well as for designing possible follow-up investigations. However, this association was not observed when the new classification was used, presumably because the definition of severe dengue is so broad. DENV2 has been associated in this and numerous other studies with the defining features of DHF/DSS [23], [24] and with DHF/DSS to a greater extent than other serotypes [24], [25]. Therefore, it is not surprising that DENV2 is significantly associated with severity (DHF/DSS) using the previous classification scheme, but not with the revised classification scheme, which is no longer specific for DHF/DSS or the key clinical manifestations (e. g. , thrombocytopenia, shock). “Severe dengue” and “Dengue with warning signs” are very broad definitions that make it difficult to determine the pathophysiology of the disease. Therefore, it is important to analyze the frequency of warning signs and of severe manifestations, respectively, in order to obtain a clearer picture of the disease profile. Similarly, the specific syndrome of plasma leakage (DHF/DSS), so characteristic of the critical phase of dengue, is lost in the new classification; thus, for studies of viral, host, and immunological determinants of dengue pathogenesis, a more specific definition than “severe dengue” will need to be implemented. Additional implications of the new classification scheme exist with respect to epidemiological surveillance, since the traditional and revised categories are not equivalent and may initially lead to a difficult transition. According to results obtained in this study, we believe the revised classification is most useful for the physician for the detection of severe cases of dengue, but its utility in pathphysiological and epidemiological studies needs further evaluation in future research.

Dengue is a mosquito-transmitted viral disease that is a major public health problem worldwide. Dengue virus (DENV) infection leads to Dengue Fever (DF) and a spectrum of life-threatening syndromes with fluid loss and hypotensive shock or other severe manifestations. Recently, the traditional World Health Organization (WHO) dengue classification scheme (classic DF, Dengue Hemorrhagic Fever (DHF), and Dengue Shock Syndrome (DSS) ) was replaced with Dengue without Warning Signs, Dengue with Warning Signs and Severe Dengue (SD). Using data from 544 laboratory-confirmed dengue cases recruited over five years of a hospital-based study of pediatric dengue in Managua, Nicaragua, we evaluated the traditional and revised classification schemes against clinical intervention levels (I–III) to determine how each captures disease severity. The sensitivity and specificity to capture Category III care for DHF/DSS were 39. 0% and 75. 5%, respectively, and for SD were 92. 1% and 78. 5%, respectively. Interestingly, DENV-2 was significantly associated with DHF/DSS; however, this association was not observed with the revised classification. This study indicates that among dengue-confirmed cases, the revised WHO classification appears to have higher sensitivity and specificity for identifying cases in need of heightened care, although it is no longer as specific for a particular pathogenic entity as was the traditional schema.

Abstract Introduction Materials and Methods Results Discussion

medicine infectious diseases virology dengue fever neglected tropical diseases dengue biology microbiology viral diseases viral disease diagnosis

2011

Evaluation of the Traditional and Revised WHO Classifications of Dengue Disease Severity

When confronted with poor oxygenation, cells adapt by activating survival signaling pathways, including the oxygen-sensitive transcriptional regulators called hypoxia-inducible factor alphas (HIF-αs). We report here that HIF-1α also regulates the life cycle of Epstein-Barr virus (EBV). Incubation of EBV-positive gastric carcinoma AGS-Akata and SNU-719 and Burkitt lymphoma Sal and KemIII cell lines with a prolyl hydroxylase inhibitor, L-mimosine or deferoxamine, or the NEDDylation inhibitor MLN4924 promoted rapid and sustained accumulation of both HIF-1α and lytic EBV antigens. ShRNA knockdown of HIF-1α significantly reduced deferoxamine-mediated lytic reactivation. HIF-1α directly bound the promoter of the EBV primary latent-lytic switch BZLF1 gene, Zp, activating transcription via a consensus hypoxia-response element (HRE) located at nt -83 through -76 relative to the transcription initiation site. HIF-1α did not activate transcription from the other EBV immediate-early gene, BRLF1. Importantly, expression of HIF-1α induced EBV lytic-gene expression in cells harboring wild-type EBV, but not in cells infected with variants containing base-pair substitution mutations within this HRE. Human oral keratinocyte (NOK) and gingival epithelial (hGET) cells induced to differentiate by incubation with either methyl cellulose or growth in organotypic culture accumulated both HIF-1α and Blimp-1α, another cellular factor implicated in lytic reactivation. HIF-1α activity also accumulated along with Blimp-1α during B-cell differentiation into plasma cells. Furthermore, most BZLF1-expressing cells observed in lymphomas induced by EBV in NSG mice with a humanized immune system were located distal to blood vessels in hypoxic regions of the tumors. Thus, we conclude that HIF-1α plays central roles in both EBV’s natural life cycle and EBV-associated tumorigenesis. We propose that drugs that induce HIF-1α protein accumulation are good candidates for development of a lytic-induction therapy for treating some EBV-associated malignancies. Epstein-Barr virus (EBV) is a ubiquitous human gamma herpesvirus that infects over 90% of the world’s population. In healthy hosts, primary infection after childhood often results in infectious mononucleosis (IM). Following primary infection, EBV establishes a life-long latent infection in a tiny subset of its host’s memory B cells where its genome is maintained as an episome that replicates in synchrony with the host’s cellular DNA (reviewed in [1,2]). Latency is characterized by expression of, at most, a small number of viral protein-encoding genes (EBNAs and LMPs), two non-coding RNAs (EBERs), and some micro (mi) RNAs (reviewed in [3]). Latent EBV infection is associated with some malignancies in humans, including nasopharyngeal carcinoma (NPC), some gastric cancers (GC), a subset of Burkitt lymphomas (BL), diffuse large B-cell lymphomas (DLBCL), and post-transplant lymphoproliferative diseases (PTLD) (reviewed in [1,4, 5]). Several EBV-encoded latency proteins and miRNAs have been shown to contribute to cell transformation and tumorigenesis [1,3]. Like other herpesviruses, EBV’s long-term success requires it to undergo lytic as well as latent modes of infection during its life cycle. While latent infection permits persistence of the virus for the life of the host, lytic replication enables production of infectious virus necessary for transmission from cell to cell and host to host. Thus, EBV occasionally reactivates out of latently infected B cells. Physiological inducers of EBV reactivation include B-cell antigen receptor (BCR) activation leading to plasma cell differentiation [2], butyrate [6,7], and transforming growth factor β (TGF-β) [8,9]. Subsequently, EBV infects differentiated cells within the normal oropharyngeal epithelial where infection is usually lytic [1,2, 10]. EBV reactivation is initiated by transcriptional activation of one or both of the viral immediate-early (IE) gene promoters, Zp and Rp, leading to production of its two IE proteins, Zta (the product of the BZLF1 gene; also called Z, ZEBRA, and EB1) and Rta (the product of the BRLF1 gene; also called R), respectively. Synthesis of Zta is sufficient to induce reactivation in most EBV-positive (EBV+) cell lines [11], while Rta induces reactivation in some cell lines [12,13]. Rta and Zta are transcription factors that then activate each other’s promoters [12,14,15] and, subsequently, activate expression of EBV’s early (E) genes, including BMRF1, a viral DNA polymerase processivity factor [also called early-antigen diffuse (EAD) ], and BGLF4, a virus-encoded protein kinase (reviewed in [16]). Given that expression of the BZLF1 gene serves as the primary gatekeeper to the viral latent-to-lytic switch in most EBV+ cell lines, transcriptional regulation of Zp has been studied extensively. Numerous cis-acting elements and their cognate trans-acting factors have been identified that contribute to silencing during latency and activation in response to inducers (reviewed in [16]). Poor oxygenation, i. e. , hypoxia, contributes to tumor progression and resistance to conventional chemotherapy (reviewed in [17–19]). The mechanisms by which cells respond to hypoxic environments are known (reviewed in [20,21]). Under normal oxygen tension corresponding to approximately 21% O2, cellular transcription factors called hypoxia-inducible factor alphas (HIF-αs) are synthesized but rapidly degraded via the ubiquitin-dependent proteasome pathway. Three distinct genes encode the HIF-αs (HIF-1α, HIF-2α, and HIF-3α). Hydroxylation of specific proline residues by oxygen-dependent cellular prolyl hydroxylases (e. g. , PHD2, encoded by the EGLN1 gene) marks these proteins for ubiquitin-mediated degradation. The hydroxylation reaction catalyzed by PHDs also involves the conversion of α–ketoglutarate to succinate, Fe2+ to Fe 3+, and O2 to CO2, with vitamin C required for the regeneration of Fe2+. Under hypoxic conditions (or in the presence of iron chelators or competitors), PHDs fail to hydroxylate HIF-αs, resulting in accumulation of these proteins to high levels. Stabilized HIF-αs form heterodimers with their constitutively present binding partner, HIF-1β [also called aryl hydrocarbon nuclear receptor translocator (ARNT) ], translocate to the nucleus, and sequence-specifically bind to hypoxia-response elements (HREs) located within the promoter regions of cellular genes involved in angiogenesis, anaerobic metabolism, and erythropoiesis. The roles hypoxia and HIF-1α play in the life cycle of Kaposi’s sarcoma herpesvirus (KSHV), another member of the gamma herpesvirus family, have been extensively studied (reviewed in [22,23]). Analogously, Jiang et al. [24] reported that incubation of the EBV+ marmoset-derived B-cell line, B95-8, in 2% oxygen conditions leads to induction of Zta synthesis within one-to-two days, and Murata et al. [25] confirmed that hypoxia (1% oxygen; 36 h) induces BZLF1 gene expression in human EBV+ Akata B cells and LCLs as well as B95-8 cells. Here, we report that drugs that mimic hypoxia induce lytic EBV infection in some EBV+ epithelial and B-cell lines by a HIF-1α-dependent mechanism. HIF-1α induces the switch to lytic-gene expression through directly activating BZLF1 gene expression by sequence-specific binding to an HRE located within Zp. We further show that HIF-1α can play important roles in EBV’s natural life cycle and tumorigenesis induced by this virus. These findings suggest a new class of drugs that may be useful in the development of a lytic-induction therapy for treating patients with some EBV-associated malignancies. Our long-term objective is to find drugs suitable for use in EBV-targeted oncolytic therapy [26,27]. Thus, we chose to mimic hypoxia by incubating cells with deferoxamine (DFO; also called Desferal) or L-mimosine (Mim; also called Leucenol), two drugs that inhibit prolyl hydroxylase activity by chelating iron [28]. The EBV+ cell lines examined were Burkitt lymphoma-derived Sal and KemIII and gastric carcinoma-derived AGS-Akata and SNU-719. SNU-719, Sal, and KemIII retain their original-infecting EBV genomes. SNU-719 cells have type I latency plus LPM2A, Sal cells have Wp-restricted latency, and KemIII have type III latency. In an initial experiment, we found that incubation of Sal cells with mimosine promoted both stabilization of HIF-1α and induction of synthesis of the immediate-early (IE) lytic EBV antigen, Zta (Fig 1A). However, because mimosine is not FDA-approved for internal use, we largely focused on DFO in subsequent experiments. Incubation of all four of these cell lines with DFO for 24 h promoted stabilization of HIF-1α protein along with inducing synthesis of Zta (Fig 1B–1E). Quantitation of the efficiency of EBV reactivation by staining cells for presence of Zta indicated that 15%-30% of AGS-Akata cells were induced into lytic-gene expression within 24 h of addition of 200 μM DFO, while 1 ½%-to-3% of Sal, SNU-719, and KemIII cells were induced within this time frame (S1 Fig; Fig 1D and 1E). MLN4924 (Pevonedistat), an inhibitor of the NEDD8-activating enzyme (NAE), also blocks degradation of HIF-αs. By preventing NEDDylation of the cullin-RING E3 ubiquitin ligases (CRLs), MLN4924 inhibits degradation of approximately 20% of cellular proteins, including the HIF-αs, whose levels are regulated in part via the proteasome degradation pathway (reviewed in [29]). We found that MLN4924 reactivation of EBV into lytic infection at a roughly similar efficiency to DFO in the EBV+ GC-derived cell lines (Fig 1B). Thus, we conclude that two classes of HIF-α stabilizing drugs with different off-target effects can both induce lytic EBV infection. We also asked whether temporary stabilization of HIF-αs resulted in abortive lytic infection or activation of the complete lytic replication cycle. Removal of DFO after 24 h led to loss of HIF-1α, as expected, yet the lytic cycle continued to progress, leading to high-level synthesis of EAD and the late (L) EBV-encoded viral capsid antigen (VCA, also called p18) by 48 h (Fig 1C). These data also suggest that the percentage of cells reactivated by DFO might well be higher than observed after only 24 h. Thus, we conclude that HIF-α-stabilizing drugs can induce lytic EBV infection in EBV+ cells of lymphocytic and epithelial origin and in a variety of latency types. While DFO induced high-level accumulation of HIF-1α protein in all four of these cell lines, it only induced HIF-2α protein to moderately high levels in SNU-719 cells (Fig 1D). Analysis of RNA-sequencing data of mRNA purified from SNU-719 cells indicated that HIF-1α mRNA was 5. 5-fold more abundant than HIF-2α mRNA and 43-fold more abundant than HIF-3α mRNA. Thus, even though we could detect some HIF-2α protein in DFO-treated SNU-719 and AGS-Akata cells, it was probably present at considerably lower levels than was HIF-1α protein. HIF-1α was also the predominant isoform of the three HIF-αs detected at the RNA level in primary tissues exhibiting EBV tropism. Transcriptome analysis of four high EBV+ primary gastric cancers from the TCGA cohort [30] indicated that HIF-1α mRNA was, on average, 2. 9-fold more abundant than HIF-2α mRNA (range 1. 9-fold to 4. 9-fold), with HIF-3α mRNA undetectable above background level. A similar analysis of 17 EBV+ endemic Burkitt lymphomas [31] indicated HIF-1α mRNA was, on average, nine-fold more abundant than HIF-2α mRNA (range 2-fold to 19-fold), with HIF-3α mRNA detectable above background in only one of these 17 tumors (at 1/20th of the HIF-1α level). Further, HIF-1α accounts for almost all of the HIF-α-related mRNA present in primary human B-cells throughout the various stages of B-cell differentiation into plasma cells (e. g. , see data presented below). Thus, although HIF-2α may contribute to EBV’s life cycle under some conditions in epithelial cells, HIF-1α appears to be the predominant HIF-α expressed in cell types of physiological relevance to EBV. Given this finding, most of the studies presented here were performed with HIF-1α. We occasionally confirmed our findings with HIF-2α and did not conduct further studies with HIF-3α. If HIF-1α induces EBV reactivation, one would expect most Zta+ cells to also be HIF-1α+. To determine the level of coincidence between Zta+ and HIF-1α+ cells, we performed dual immunofluorescence staining (IFS) assays. Consistent with our hypothesis, we found that almost all of the Zta+ cells were also HIF-1α+ in AGS-Akata cells that had been incubated with DFO for 24 h (Fig 2). The occasional Zta+, HIF-1α-negative cell we observed was likely the consequence of AGS-Akata cells exhibiting some spontaneous reactivation (e. g. , S1B Fig). Thus, we conclude that DFO efficiently induces EBV reactivation in AGS-Akata cells, at least in part, by stabilizing HIF-1α. To demonstrate a direct causal role of HIF-1α in reactivation, we evaluated induction of synthesis of lytic EBV antigens after addition of HIF-1α. AGS-Akata cells were co-transfected with: (i) a plasmid expressing an oxygen-insensitive variant of HIF-1α that contains alanine substitutions in the proline residues targeted for hydroxylation by PHDs; and (ii) a plasmid expressing HIF-1α’s heterodimeric partner, HIF-1β/ARNT. Addition of HIF-1α/HIF-1β was sufficient to strongly induce synthesis of Zta and EAD (Fig 3A). We also performed a reciprocal experiment. Knockdown of HIF-1α expression by 80%-90% in AGS-Akata cells resulted in a comparable level of loss of DFO-induced synthesis of Zta and EAD (Fig 3B, lanes 1–6). Similar findings were observed in Sal cells infected with these lentiviruses (Fig 3B, lanes 7–10). Thus, we conclude that DFO-directed induction of lytic EBV infection is mediated largely by HIF-1α. HIF-1α induces KSHV reactivation by directly enhancing expression of its ORF50 gene, the orthologue of EBV’s BRLF1 gene [32,33]. Thus, we asked whether HIF-1α reactivates EBV by inducing transcription from Rp and/or Zp. HEK 293T cells were transiently co-transfected with plasmids expressing the oxygen-insensitive variant of HIF-1α, HIF-1β, and reporters driving luciferase expression from Rp or Zp. We used an Rta expression plasmid as a positive control since Rta is a potent transcriptional activator of both Zp and Rp [15]. While addition of HIF-1α/HIF-1β activated transcription from the Zp-luc reporter approximately 24-fold, it activated the Rp-luc reporter similarly to the four-fold activation observed with the negative control TATA-luc reporter (Fig 4). As expected, Rta robustly activated both reporters. Thus, HIF-1α/HIF-1β heterodimers activate transcription from Zp approximately six-fold above the non-specific level observed in this assay while failing to activate specifically transcription from Rp. Thus, in contrast to KSHV, we conclude that HIF-1α regulates lytic EBV infection by activating expression of the BZLF1 gene, not the BRLF1 gene. To determine how HIF-1α activates BZLF1 gene expression, we performed an in silico analysis of Zp and noted a single consensus HRE located from nt -83 through -76 relative to the Zp transcriptional initiation site (Fig 5A). To examine whether HIF-1α-dependent transactivation of Zp mapped to this sequence, we constructed a set of base-pair substitution mutant variants of our WT luciferase reporter, pWTZp-luc (Fig 5B). These mutations were designed to avoid disrupting bases that overlap the adjacent ZIIR silencing element [34,35]. Reporter assays performed with these variants of pZp-luc showed that the WT and ZIIR mutant promoters were activated by HIF-1α/HIF-1β approximately five- to eight-fold above the non-specific activation observed with the minimal TATA box-containing control promoter while none of the 3-bp substitution mutants in the putative HRE were activated above this non-specific level (Fig 5C). Even the 1-bp substitution mutation present in mutant M1 significantly reduced activation by HIF-1α/HIF-1β. Analysis of the basal activity of these mutants in the absence of HIF-1α and of a non-overlapping mutant only altered in nt -77 and -76 of the Zp HRE ruled out the possibility that these HRE mutations were affecting binding of a repressor (S2 Fig). Similar results were obtained when we used an expression plasmid that encodes an oxygen-insensitive variant of HIF-2α in place of the HIF-1α one (Fig 5D). Thus, we conclude that Zp contains a transcriptionally functional HRE that includes nt -79 through -81. HREs act as sequence-specific binding elements for HIF-α/β heterodimeric complexes. To demonstrate that HIF-1α/HIF-1β heterodimers bind to the Zp HRE, we performed in vitro DNA-binding assays. Our protein source of HIF-1α/HIF-1β complexes was nuclear extract obtained from EBV-negative AGS cells incubated for 24 h with 200 μM CoCl2, an iron competitor. A radiolabeled, double-stranded oligonucleotide containing a consensus HRE sequence, 5’-CACGTC-3’, served as probe (Fig 6C, HRE WT). We identified the HIF-1α-containing protein-DNA complex by showing it was lost by incubation with a HIF-1α-specific antibody (Fig 6A). Competition electrophoretic-mobility-shift assays (EMSAs) were performed by pre-incubation of the extract with various amounts of the unlabeled, double-stranded WT or mutant (MT) oligonucleotides indicated in panel C. WT Zp HRE-containing oligonucleotide competed for binding the HIF-1α/HIF-1β complex as well as the consensus WT HRE oligonucleotide (Fig 6B, lanes 9–11 vs. lanes 3–5, respectively) while the 3-bp mutant variant of this consensus HRE oligonucleotide failed to compete (Fig 6B, lanes 6–8 vs. lane 2). Likewise, a 3-bp mutant variant of the Zp HRE-containing oligonucleotide corresponding to the M3 mutation that abolished HIF-1α/HIF-1β-dependent transcriptional activation of Zp-luc (Fig 5) also largely failed to compete for binding HIF-1α/HIF-1β complexes (Fig 6B, lanes 12–14 vs. lane 2). Thus, the trans-activation and DNA-binding activities of HIF-1α co-localize to the HRE present within Zp. We next performed ChIP assays to show HIF-1α binds Zp in the physiological context of whole EBV genomes. SNU-719 and Sal cells incubated (+) or not (-) with 200 μM DFO for 24 h served as the source of chromatin given this treatment induces abundant accumulation of HIF-1α in these cells (Fig 7A). Quantitative PCR analysis of these samples following chromatin precipitation with HIF-1α-specific versus IgG control antibody indicated that this HIF-1α-specific antibody precipitated Zp approximately four-fold more efficiently than did the anti-IgG antibody (Fig 7B and 7C), yet failed to increase significantly precipitation of EBV DNA located approximately 4. 8 kbps upstream of Zp (Neg. Cntl.). Thus, we conclude that the EBV BZLF1 gene contains a transcriptionally functional, HIF-1α-binding HRE within Zp. To confirm that HIF-1α induction of lytic EBV infection truly occurs via binding to this Zp HRE rather than indirectly via downstream signaling events, we constructed two independent HRE variants of EBV containing the M2 and M4 substitution mutations analyzed in our reporter assay (Fig 5) within the context of the p2089 BAC [36]. 293T cells were transfected in parallel with these two EBV HRE mutant BACs alongside their parental WT EBV BAC and selected for resistance to hygromycin to establish the cell lines 293T-EBV M2,293T-EBV M4, and 293T-EBV WT, respectively. Confirming our observation with AGS-Akata cells (Fig 3A), co-transfection of 293T EBV-WT cells with plasmids expressing the oxygen-insensitive variant of HIF-1α along with HIF-1β efficiently induced expression of EBV IE and E genes (Fig 8A, lane 2 vs. lane 1). Strikingly, co-transfection of HIF-1α/HIF-1β expression plasmids into 293T cells latently infected with either the M2 or M4 HRE mutant variant of EBV failed to induce synthesis of lytic EBV antigens above the background level of spontaneous reactivation (Fig 8A, lane 4 and lane 6, respectively). To rule out non-HRE-related causes for our negative finding, we also transfected these cell lines with plasmids expressing Zta (Fig 8B) or Rta (Fig 8C). When either of these EBV IE proteins was provided, all three cell lines exhibited similar high-level expression of both the non-transfected IE gene and the EAD-encoding gene, BMRF1 (Fig 8B, lanes 2,4, and 6; Fig 8C, lanes 2,4, and 6). We also recovered the viral DNAs from these mutant-infected cell lines and thoroughly analyzed their genomes for second-site mutations; none were found by either DNA sequencing or restriction fragment pattern analysis. Thus, mutation of the Zp HRE within the context of whole EBV genomes disables HIF-1α-dependent induction of BZLF1 gene expression. We conclude that HIF-1α induces lytic reactivation in EBV primarily (possibly, exclusively) via direct binding to this single HRE located within Zp. Some HREs respond to both of the two major HIF-α isoforms whereas others primarily or solely respond to only one of them [18]. As indicated above, we observed that HIF-1α RNA is more abundant that HIF-2α RNA in all of the EBV+ epithelial and B cell lines and tumors we have examined to date. Nevertheless, given our finding that HIF-2α can also activate Zp in reporter assays (Fig 5), it remains possible that some conditions exist (e. g. , chronic hypoxia) in which HIF-2α is the more physiologically important HIF-α regarding some aspects of EBV’s life cycle. Thus, we examined likewise whether latent EBV genomes can also be induced into lytic infection using a plasmid that expresses an oxygen-insensitive variant of HIF-2α. As with HIF-1α, we observed high-level induction of EBV IE and E gene expression in the cells latently infected with the WT EBV genome, but not cells latently infected with the Zp HRE M2 mutant (Fig 8D, lane 2 vs. lane 4, respectively). Noteworthy is the fact that neither HIF-1α nor HIF-2α directly activated BRLF1 gene expression in cells infected with Zp HRE mutant genomes; if they had, we should have seen synthesis of Zta and EAD as well as Rta protein as was observed in Fig 8C. This finding demonstrates that HIF-α/β heterodimers fail to activate transcription from Rp in the context of full-length latent EBV genomes as well as in reporter assays (Fig 4). Thus, we conclude that, when present, either of the two major HIF-α isoforms can mediate EBV reactivation via the Zp HRE. Why might EBV have evolved to contain an HRE within Zp? To answer this question, we examined whether the appearance of HIF-1α protein during differentiation of normal epithelial and B cells coincides with the cell types in which lytic EBV infection takes place. In B cells, lytic EBV reactivation occurs when memory B-cells begin to differentiate into plasma cells [2]. To determine when functionally active HIF-1α protein is present in B cells, we mined existing microarray data sets obtained from B cells harvested at eight different stages of differentiation, ranging from naïve B cells to fully differentiated plasma cells (Fig 9). HIF-1α mRNA is present at high levels in all of these stages, declining somewhat only during the very last stage. However, functionally active HIF-1α protein, as measured by expression of the HIF-1α-activated genes VEGFA and PDK1, dramatically increases in the post-memory cell preplasmablast and plasmablast stages, respectively. These are the same stages during B-cell differentiation when expression of both ZEBs plummets (possibly due, in part, to HIF-1α also activating synthesis of miR-429 [37], a down-regulator of ZEB levels [38,39]), and expression of Blimp-1α and XBP-1 dramatically increases. Thus, the stages during B-cell differentiation when EBV reactivates are coincident with the stages when three of the Zp activators (HIF-1α, Blimp-1α [40], and XBP-1s [41,42]) appear and two of the major Zp repressors (ZEB1 and ZEB2 [43,44]) disappear. Another stage of EBV’s natural life cycle involves the infection of differentiated epithelial cells by EBV (either free virions or virus produced in reactivated EBV+ B cells) [10]. Expression of Blimp-1α is also induced during epithelial cell differentiation, synergizing with KLF4 to activate transcription from both Zp and Rp [40,45]. To determine whether HIF-α protein accumulation is induced by epithelial cell differentiation, we incubated telomerase (TERT) -immortalized human normal oral keratinocyte (NOK) cells with the differentiation-inducing agent, methylcellulose (MC) (Fig 10A). Both HIF-1α and HIF-2α protein, along with some Blimp-1α, appeared within 2 h of MC addition; they remained present for at least 12 h. Thus, their stabilization may be among the earliest events to occur during epithelial cell differentiation, hours before the appearance of involucrin, another marker of epithelial cell differentiation. The kinetics of appearance of HIF-1α and Blimp-1α were similar in MC-treated NOK-Akata, cells infected with EBV (Fig 10B). This latter finding suggests that regulation of the stabilization of HIF-1α protein during epithelial cell differentiation occurs independently of the presence of EBV. We examined likewise hTERT-infected human gingival epithelial (hGET) cells. In this case, HIF-1α protein and Blimp-1α were both abundantly present, along with involucrin, 48 h after addition of MC (Fig 10C, lane 2); accumulation of HIF-1α protein was within a few-fold of that observed when these cells were incubated with 50 μM DFO (Fig 10C, lane 4). HIF-1α protein also accumulated together with Blimp-1α and involucrin when NOK cells were induced to differentiate by growth in organotypic culture (Fig 10D). Thus, both HIF-1α and Blimp-1α are present in differentiated cells of the types present in the human oral cavity. Thus, we conclude that HIF-1α protein accumulates during the course of both epithelial and B-cell differentiation, likely contributing to activation of BZLF1 gene expression along with other inducers of Zp activation. Given the above findings, we hypothesized that hypoxic regions within growing EBV+ tumors accumulate HIF-1α, thereby increasing the probability that latent EBV infection will reactivate into lytic replication. To test this hypothesis, we examined EBV+ B-cell lymphomas [similar in phenotype to human diffuse large B-cell lymphomas (DLBCLs) ] that had been induced in NSG (NOD/LtSz-scid/IL2Rγnull) mice by inoculation with human cord blood that had been infected with the M81 strain of EBV by co-culture for 90 min immediately prior to injection [46]. If our hypothesis is valid, EBV+ tumor cells located distally from blood vessels (i. e. , in poorly oxygenated regions) are more likely to reactivate into lytic replication than ones located near them. Latently EBV-infected B-cells were identified by IFS for the latent EBV protein, EBNA2 (Fig 11A); lytic infection was identified by IFS for the lytic EBV protein, Zta (Fig 11B and 11C; see also S3 Fig for adjacent serial sections, including H&E staining). The blood vessels were identified by co-staining with a CD31 (PECAM-1) -specific antibody that detects endothelial cells. Hypoxic regions were identified by co-staining with a Hypoxyprobe-specific antibody in mice that had been treated with Hypoxyprobe 90 min prior to sacrifice. Strikingly, whereas the EBV+ EBNA2 cells were located throughout these tumor sections, as expected, Zta+ cells were only occasionally observed within three cell widths of blood vessels. The distributions of distances of EBNA2+ versus Zta+ cells from their nearest blood vessel are summarized in Fig 12: while most EBNA2+ cells were located within 30 μm of a blood vessel, most Zta+ cells were located beyond this distance (p < 10−19). Dual staining for Zta and Hypoxyprobe pictorially documented that most of the Zta+ cells were located within or near hypoxic regions of the tumors, distal to blood vessels (Fig 11C, S3 Fig). IFS of xenografts generated by injection of gastric cancer-derived SNU-719 cells into the flanks of NSG mice (e. g. , S4 Fig) and IHC staining of serial sections of M81-induced tumors (e. g. , S5 Fig) produced similar findings. Thus, we conclude that the probability of a latently infected cell reactivating in vivo is considerably higher when it is located in an oxygen-deficient environment. The HRE identified here overlaps the previously identified ZIIR element of Zp [34,35] (Fig 5B). Thus, one possibility was that binding of HIF-1α to this HRE activates transcription from Zp by displacing the yet-to-be-identified ZIIR repressor. Inconsistent with this hypothesis was our finding that mutations known to relieve ZIIR-mediated repression affected neither HIF-1α- nor HIF-2α-induced activation of transcription from Zp (Fig 5C and 5D, respectively) unless they also impinged upon the HRE element (S2 Fig). Furthermore, HRE mutations that abolished HIF-α-induced reactivation of EBV had no effect on the frequency of spontaneous reactivation (Fig 8), a frequency enhanced in ZIIR mutant variants of EBV [35]. Thus, we conclude that the HRE and ZIIR elements are genetically distinguishable, independently acting regulatory elements of Zp, with HIF-α proteins functioning as transcriptional activators via binding to the HRE. The sequence encompassing the HRE present in the promoter of KSHV’s latent gene, ORF73 [encoding latency-associated nuclear antigen (LANA) ], is identical to that of the EBV Zp HRE we identified here, with both HREs being responsive to both major HIF-α isoforms [48] (Figs 5 and 8). However, HIF-1α was the predominant HIF-α expressed at the RNA level in all of the EBV+ primary tumors and cell lines we have examined to date. Consistent with this finding, the gastric cancer-derived cell lines, SNU-719 and AGS-Akata, were the only ones in which we detected HIF-2α protein upon incubation with DFO (Fig 1D). Previous reports of others likewise indicated preferential accumulation of HIF-1α protein with exposure to hypoxia in EBV+ LCLs that contain little HIF-2α mRNA [49] and in EBV+ NPC-derived cell lines that contain some HIF-2α mRNA [50]. Thus, we conclude that HIF-1α is the primary HIF-α of physiological relevance to EBV’s natural life cycle and in EBV+ tumors. Much literature exists indicating HIF-1α plays central roles in regulating both lytic infection and tumorigenesis by KSHV (reviewed in [22,23]). Functional HREs are present within the promoter regions of KSHV’s latent gene, ORF73/LANA, as well as its IE lytic gene, ORF50/RTA, and lytic ORF34-ORF37 gene cluster [32,48,51]. HIF-1α complexes with LANA to activate ORF50 gene expression during hypoxia, inducing lytic KSHV replication [33], yet a SUMOylated form of LANA inhibits HIF-1α induction of RTA synthesis to maintain latency during normoxia while still enabling HIF-1α to promote angiogenesis [52]. As with KSHV, the relationship between EBV’s latent gene products and HIF-1α is also complex. EBNA3A and EBNA-LP bind PHDs, blocking their catalytic activity and, thereby, inhibiting oxygen-dependent degradation of HIF-1α [53]. LMP1 promotes accumulation of HIF-1α by signaling PHD1 and PHD2 degradation pathways [54,55]. EBNA3A stabilizes HIF-1α via protein-protein interactions [53], a complex somewhat analogous to the LANA/HIF-1α complex. However, these above-mentioned EBV-encoded proteins are clearly not necessary for HIF-1α-induced activation of BZLF1 gene expression given we showed here that HIF-1α can induce transcription from Zp in EBV-negative cells and Zta synthesis in EBV+ Sal cells that are in a Wp-restricted latency in which these proteins are not expressed. In latency types in which these above-mentioned proteins stabilize HIF-1α, other factors are likely also present in the cells to inhibit HIF-1α from inducing lytic reactivation. EBNA1, present in all EBV+ cells, also has been reported to enhance HIF-1α activity, most likely indirectly via its effects on AP-1 [56]. We propose that HIF-1α plays central roles in regulating both lytic replication and tumorigenesis by EBV. Regarding EBV’s natural life cycle, we hypothesize that B cells from the naïve B-cell through memory B-cell stages lack functional HIF-α activity as well as Blimp-1α and XBP-1s (Fig 9) (other known inducers of BZLF1 gene expression), while containing several known direct or indirect repressors of this gene [9,43] and inhibitors of Zta activity [57,58] (Fig 13). Thus, EBV infection tends to go latent. However, if an EBV-infected B cell begins to undergo plasma cell differentiation, the virus may reactivate due to the appearance of functionally active HIF-1α along with these other activators and loss of these repressors. Likewise, when epithelial cells differentiate, they accumulate HIF-αs along with Blimp-1α and KLF4 (known inducers of BRLF1 as well as BZFL1 gene expression [40,45]) and lose repressors of these genes such as the ZEBs (Fig 13). Thus, when EBV virion particles or EBV-infected B cells come into close contact with differentiated epithelial cells within the oral cavity, the introduction of EBV genomes into these cells can lead to lytic replication and production of infectious virus, helping to spread the virus from cell-to-cell and host-to-host. How might HIF-1α and the Zp HRE contribute to tumorigenesis by EBV in vivo? We propose that HIF-αs contribute via two routes to tumor growth in EBV+ cancers. As is true of most tumors, EBV+ tumors develop hypoxic regions as they enlarge (e. g. , Fig 11C, S3A Fig; [59–61]), leading to accumulation of the HIF-αs whose genes are being expressed. These HIF-αs then activate expression of a variety of cellular genes involved in angiogenesis and anaerobic metabolism that help the tumor to continue to enlarge (reviewed in [62]). In the case of EBV+ tumors, presence of HIF-α also contributes to activation of BZLF1 gene expression, leading to EBV lytic-gene expression in some tumor cells. We showed here by examining EBV-induced lymphomas and EBV+ gastric cancer xenografts that Zta+ cells were preferentially located in regions of the tumors that were clearly hypoxic as indicated by Hypoxyprobe staining or, presumably, hypoxic because they were located distal to blood vessels (Figs 11 and 12, S3–S5 Figs). Thus, we propose the following model: Hypoxic regions develop in EBV+ tumors as they grow in size, leading to accumulation of HIF-1α and, in some cases, HIF-2α. Prior to angiogenesis, HIF-α increases the frequency of lytic EBV infection in these hypoxic regions, with these lytic-infected cells secreting a variety of cellular and viral factors, some of which contribute to the enhancement of tumor growth (reviewed in [47]). The goal of chemotherapy is to kill cancer cells while minimizing harm to healthy cells. Treatment of some EBV+ cancers with minimally toxic drugs that rapidly and efficiently induce EBV lytic-gene expression, in combination with prodrugs such as ganciclovir (GCV), may be one way to achieve this goal ([16,63] and references cited therein). Based upon the findings presented here, we propose that briefly targeting the PHDs or other enzymes that regulate degradation of HIF-1α (e. g. , NAE) may be useful as part of a strategy to achieve efficient EBV lytic-induction therapy. Transient expression of HIF-1α induced sufficient Zta synthesis to promote expression of EBV early- and late-lytic genes (Fig 1C). Furthermore, these expressed early-lytic genes included BGLF4 [as indicated by the presence of phosphorylated forms of EAD (e. g. , Figs 3 and 8) ], the gene that encodes the EBV-PK that can phosphorylate GCV [64]. Intrinsic features of EBV and HIF-1α make this strategy feasible: (i) Once BZLF1 gene expression is activated by an inducer, Zta synthesis usually continues after the inducer is removed because of its positive feedback loop with Rta (e. g. , Fig 1C); and (ii) The HIF-αs are rapidly degraded once HIF-α-stabilizing drugs are removed (e. g. , Fig 1C). Thus, brief treatment with a stabilizer of HIF-1α may well be sufficient, reducing potential adverse reactions due to off-target effects of the drug and HIF-1α. Quite likely, one may be able to increase considerably the percentage of the EBV+ cells reactivated by using DFO, MLN4924, or another HIF-1α stabilizer in combination with other drugs known to activate BZLF1 gene expression via different cellular signaling pathways (e. g. , HDAC inhibitors). We were fortunate to identify here an already FDA-approved drug as a possible candidate for use in lytic-induction therapy. DFO and the FDA-approved oral iron-chelators deferasirox and deferiprone, are used to treat iron overload and toxicity that result from frequent blood transfusions [65]. DFO-based therapy is also emerging as a tool for treating a variety of diseases, including persistent anemia, impaired angiogenesis resulting from diabetes mellitus, and numerous neurodegenerative disorders (reviewed in [66]). The mouse experiments were approved by the UW-Madison Institutional Animal Care and Use Committee (protocol #M005197-A01) and conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals. The mice were sacrificed by cervical dislocation under isoflurane anesthesia. The UW IRB classified the work with human tissues and cells as exempt. Sal cells (a gift from Alan Rickinson via Bill Sugden) were derived from an EBV+ BL; they are co-infected with wild-type and EBNA2-deleted EBV genomes and maintain a Wp-restricted latency [67]. KemIII cells (a gift from Alan Rickinson via Jeff Sample), derived from an EBV+ BL, are currently in type III latency and express LMP1. These B-cell lines were maintained in RPMI-1640 medium supplemented with 10% FBS and 100 units/ml penicillin and 100 μg/ml streptomycin (pen-strep; Life Technologies). SNU-719 cells (obtain from Jin-Pok Kim via Bill Sugden), derived from an EBV+ gastric carcinoma, retain their original EBV genome [68]; they were maintained likewise. AGS-Akata cells, an EBV-infected clonal derivative of AGS cells (derived from a human gastric carcinoma; obtained from ATCC) [69], were maintained in F12 medium (Life Technologies) supplemented with 10% fetal bovine serum (FBS; Atlanta Biologicals) and pen-strep additionally supplemented with 400 μg/ml of G418. 293T (obtained from ATCC) is a human embryonic kidney (HEK) cell line expressing the early genes from SV40 and adenovirus. These cells were maintained in DMEM (Life Technologies) supplemented with 10% FBS and pen-strep. 293T cells harboring the B98. 5 strain of EBV in BAC p2089 [36] or HRE mutant variants thereof were maintained in DMEM additionally supplemented with 100 μg/ml hygromycin B. NOK (a gift from Karl Munger) are telomerase (hTERT) -immortalized normal oral keratinocyte (NOK) cells [13]. NOK-Akata, clone 2 (generously obtained from Bill Sugden), are NOK cells (with WT p53) that are latently infected with an Akata-GFP strain of EBV [13]. NOK (clones #1 and #3) cells are clonal isolates of NOK cells (with WT p53). These cell lines were maintained in an undifferentiated state by growth in keratinocyte serum-free medium (K-SFM; Life Technologies) supplemented with epidermal growth factor, bovine pituitary extract, pen-strep. The NOK-Akata growth medium also included 50 μg/ml G418. hTERT-transduced human gingival epithelial (hGET) cells were generated as follows. A frozen pool of primary human gingival epithelial cells (HGEPp) was obtained from CellnTEC. Upon thawing, the cells were initially grown in their specialty medium (CnT-PR; CELLnTEC) and re-frozen. These cells were then passaged in K-SFM supplemented with a ROCK inhibitor (10 μM Y-27632 Di-HCl; Selleck Chemical #50-863-6) and infected with pBABE-puro-hTERT (Addgene plasmid #1771; a gift from Bob Weinberg) [70], a recombinant retrovirus expressing human telomerase. The hTERT-transduced cells were selected by incubation with puromycin (1 μg/ml), pooled, and subsequently maintained in K-SFM. All cells were incubated at 37°C in a humidified 5% CO2 atmosphere. Plasmids pHA-HIF-1α P402A/P564A-pcDNA3 and pHA-HIF-2α P405A/P531A-pcDNA3 express oxygen-insensitive variants of HIF-1α and HIF-2α, respectively [71]; they were obtained from William Kaelin via Addgene (#18955 and #18956, respectively). A HIF-1β expression plasmid, pSV-Sport-ARNT [72], was obtained from Christopher Bradfield. Plasmid pZpWT-luc contains the nt -221 to + 30 region of Zp relative to the transcription initiation site cloned between the KpnI and HindIII sites of the luciferase reporter plasmid, pGL3-Basic (Promega) [73]. The mutant variants of it shown in Fig 5B contain the indicated base pair substitution mutations; they were generated by Quick Change methodology (Stratagene), with pZpWT-luc serving as template and synthetic oligonucleotides containing the desired mutations surrounded by 10 bases of wild-type sequence serving as primers. Plasmid pWTRp-luc contains the nt -1069 to +38 region of Rp relative to the transcriptional initiation site cloned between the KpnI and HindIII sites of pGL3-Basic. Plasmid pTATA-luc [74] served as a negative control. Plasmids pSG5-BZLF1 and pRTS15 (kindly provided by Diane Hayward) express Zta and Rta, respectively, from the SV40 early promoter [75]. Plasmid pcDNA3-BRLF1 expresses Rta from the CMV IE promoter [76]. Plasmid p2089 (a generous gift from Wolfgang Hammerschmidt) is a BAC containing the entire genome of the B95. 8 strain of EBV [36]. 293 cells infected with the M81 strain of EBV in a BAC were a generous gift from Henri-Jacques Delecluse [77]. To mimic hypoxia, cells were incubated with the indicated concentrations of CoCl2, Deferoxamine (DFO, Sigma; also called Desferrioxamine, Desferal; stock solution prepared in PBS), L-Mimosine (Mim; Sigma; stock solution prepared in 10% NaHCO3), or MLN4924 (Pevonedistat; AdooQ Bioscience #A11260; stock solution prepared in DMSO) for the indicated time periods. Whole-cell extracts (WCE) were prepared in SUMO lysis buffer [150 mM sodium chloride, 1% Nonidet P-40,0. 5% sodium deoxycholate, 0. 1% sodium dodecyl sulfate, 50 mM Tris (pH 8. 0), 50 mM sodium fluoride, 50 mM β-glycerophosphate, 2 mM sodium vanadate, 1x Complete Protease Inhibitor (Roche) ]. Proteins were separated by electrophoresis in SDS gels containing 4–20% (NuSep) or 10% (Biorad) polyacrylamide and transferred to nitrocellulose membranes (ISC Biosystem). After blocking by incubation for 1 h with 5% casein in TBST [10 mM Tris-HCl (pH 7. 4), 0. 15M NaCl, 0. 1% Tween 20], the membranes were incubated overnight at 4°C in 5% casein-TBST containing antibody specific to Zta (BZLF1,1: 250, #sc-53904; Santa Cruz), Rta (BRLF1,1: 250, #11–008; Argene), EAD (BMRF1,1: 250, #VP-E608; Vector Laboratories), or VCA/p18 (BFRF3,1: 1000, #J125; East Coast Biologics) protein. Afterward, the membranes were washed, incubated for 1 h in 5% casein-TBST containing the appropriate secondary antibody (1: 5000, horseradish peroxidase (HRP) -conjugated goat anti-mouse IgG, #G-21040, Thermo Scientific; 1: 5000 HRP-conjugated donkey anti-rabbit IgG, #NA-934, GE Healthcare; or 1: 5000, HRP-conjugated donkey anti-goat IgG, #sc-2056, Santa Cruz) washed again with TBST, incubated for 2 min in enhanced chemiluminescence (ECL) (Luminata Crescendo, #WBLUE0100; Millipore), and exposed to X-ray film (Kodak or GeneMate). To detect HIF-1α, membranes previously probed for lytic EBV antigens were washed for 1 h in TBST, incubated overnight at 4°C in 5% casein-TBST containing anti-HIF-1α polyclonal antibody (1: 500 or 1: 1000, ab103063; Abcam), washed again, incubated for 1 h in 5% casein-TBST containing the secondary antibody (1: 5000, horseradish peroxidase-conjugated donkey anti-rabbit IgG, #NA9340V; GE Healthcare), and processed as described above. HIF-2α was detected likewise with an anti-HIF-2α antibody (1: 500 or 1: 1000, ab13704; Abcam) using separate membranes from the ones used for HIF-1α. In some experiments, membranes were probed likewise with anti-Blimp-1α (1: 1,000, #9115; Cell Signaling) and anti-involucrin (1: 3000, #I9018; Sigma) antibodies. As a loading control, membranes were also probed with anti-GAPDH (1: 5000, #A00192-40; GenScript), anti-β-actin (1: 15,000, #A5441; Sigma), or anti-α-tubulin (1: 2,000, #T5168; Sigma) antibody as indicated. For IFS of cells grown in culture, cells were seeded onto cover slips placed within 10-cm dishes, incubated in medium with or without the indicated concentration of DFO for 24 h, and fixed by incubation with cold methanol: acetone (1: 1) for 10 min immediately after washing with cold PBS containing DFO or after incubation in medium without DFO for the indicated additional times. Non-specific antibody binding was blocked by incubation with Blotto [5% casein, 5% donkey serum (Sigma) ] for 2 h at room temperature. Cells were probed for Zta protein by incubation at 4°C overnight with mouse anti-BZLF1 antibody (1: 300 in Blotto, #sc-53904; Santa Cruz) followed by incubation for 2 h at room temperature with secondary antibody conjugated to a fluorescent dye (1: 500, donkey anti-mouse IgG with Alexa Fluor 488, #37114; Invitrogen). After washing, DNA was stained by incubation with 4’, 6-diamidino-2-phenylindole (DAPI, 1: 2,000), for 15 min at room temperature. Cells were stained likewise for HIF-1α by primary incubation with rabbit anti-HIF-1α antibody (1: 500, #GTX127309; GeneTex) followed by incubation with Alexa Fluor 647 anti-rabbit secondary antibody (1: 500, #A32733; Molecular Probes). Frozen M81 and SNU-719 tumor sections (8 μm and 10 μm, respectively) were fixed for IFS in cold acetone (for Hypoxyprobe and Zta co-stain) or -20°C methanol (for EBNA2, Zta, and CD31 co-stains), and blocked in PBS with 0. 1% Tween-20 and 5% goat serum (EBNA2 co-stain with CD31), 5% casein, 5% donkey serum (Zta co-stain with Hypoxyprobe), or 5% casein, 5% goat serum (Zta co-stain with CD31). Sections were then incubated in the indicated primary antibody overnight. The antibodies used were as follows: anti-Zta primary (1: 100, BZ1; Santa Cruz) or anti-EBNA2 primary (1: 50, #ab90543; Abcam), followed by goat anti-mouse IgG with Alexa Fluor 488 (1: 250, #A11001; Invitrogen); anti-CD31/PECAM1 primary (1: 50, #ab28364; Abcam) followed by goat anti-rabbit IgG with Alexa Fluor 594 (1: 500, #A11012; Invitrogen); and Hypoxyprobe primary (1: 50, #PAb2627AP; Hypoxyprobe, Inc.) followed by donkey anti-rabbit with Alexa Fluor 594 (1: 500, #A21207; Invitrogen). Images were taken and distance measurements were determined with a Zeiss AxioImager M2 microscope and Axiovision Software version 4. 8. 2. For the IHC studies (S1 and S5 Figs), the cells and M81-induced lymphomas were fixed immediately after harvest, embedded in paraffin, sectioned, deparaffinized, the antigens retrieved by incubation with 10 mM citrate buffer (pH 6. 0) containing 0. 05% Tween 20 for 20 min at 98°C, and processed as previously described [46,78,79]. Sections were probed for the indicated proteins using the following antibodies: CD20 (1: 600, clone H1; BD Biosciences); EBNA2 (1: 100, PE2; Leica Microsystems); and Zta (1: 200, BZ1; Santa Cruz). For reporter assays, 293T cells maintained in 24-well plates were co-transfected using TransIT-LT1 (Mirus Corp.) with (i) 45 ng pHA-HIF-1α P402A/P564A-pcDNA3 plus 45 ng pHIF-1β or 45 ng of each of their parental expression plasmids as controls, and (ii) 200 ng of the indicated luciferase reporter plasmid. Cells were harvested 24-to-48 h later, lysed with Passive Lysis Buffer (Promega), and luciferase activity was determined according to the manufacturer’s instructions. All assays were performed in triplicate on three or more occasions. For all other assays, expression plasmids were transfected into the indicated cells using TransIT-LT1 and the amounts of DNA indicated followed by incubation at 37°C for the times indicated prior to harvesting and processing as indicated in each figure legend. AGS-Akata cells maintained in 10-cm dishes were transiently transfected when approximately 60% confluent using TransIT-LT1 with 0. 8 μg each of five pLKO. 1-based lentiviral vector DNAs encoding shRNAs that target HIF-1α (plasmids #3808, #3809, #3810, #3811, and #10819; Thermo Scientific). As controls, cells were transfected with 4 μg of pLKO. 1 expressing the non-targeting shRNA 1864 (cntl. #1, #1864; Addgene) or NT (cntl. #2, #SHC002; Sigma-Aldrich). Two days later, cells were incubated with 200 μM DFO for 24 h, harvested, lysed in SUMO buffer, and processed for immunoblot analysis. To transduce Sal cells with these shRNA-encoding lentiviruses, the lentiviruses were first packaged into virions as described by Open Biosystems. 293T cells in 10-cm-diameter dishes were co-transfected with (i) 0. 8 μg of the five individual shRNA lentiviral vectors targeting HIF-1α or 4 μg of non-targeting shRNA cntl. #1 lentiviral vector, (ii) 1. 4 μg of pCMV-dR8. 2 dvpr (#8455; Addgene), and (iii) 0. 6 μg of a plasmid encoding vesicular stomatitis virus G protein (VSV-G) (gift from Bill Sugden). The medium containing the virus was harvested 72 h later, passed through 0. 8-μm-pore-size filters, and used to infect the Sal cells subsequently processed as described above for AGS-Akata cells except that the DFO was added three days after infection with the lentiviruses. The protein source was nuclear extract prepared as previously described [40] from AGS cells that had been incubated with 200 μM CoCl2 for 24 h. The probe was the 5’-end-labeled, double-stranded oligonucleotide, 5’- AAACGCAAGCCGCACGTCTCACAGATCC-3’ (underlined sequence indicates consensus HRE). Reactions were performed with 20 mM HEPES (pH 7. 9), 0. 1 M KCl, 6 mM MgCl2,4 μg poly (dI-dC): (dI-dC), 0. 5 mM PMSF, 0. 5 mM DTT, 8% Ficoll in a final volume of 20 μl. For immunoshift EMSAs, 10–100 μg of protein extract was pre-incubated in the reaction buffer for 20 min at 4°C with 1 μg anti-HIF-1α polyclonal antibody (#ab103063, Abcam) prior to addition of the radiolabeled probe and incubation at room temperature for 15 min. For competition EMSAs, unlabeled, competitor double-stranded oligonucleotides were pre-incubated likewise with the reaction mixture prior to addition of the radiolabeled probe. Protein-DNA complexes were separated by electrophoresis at 200 V for 2 h at 4°C in a non-denaturing 4% polyacrylamide gel with 0. 5X Tris-borate-EDTA (TBE) as the running buffer. Gels were dried and imaged on a STORM 840 phosphorimager (GE Healthcare). Chip assays were performed essentially as previously described [40] using approximately 2 x 107 SNU-719 and Sal cells grown in 15-cm dishes. Cells were incubated for 24 h with 200 μM DFO (+) or PBS (-) in medium as indicated. Protein-DNA complexes were cross-linked by incubation with 1% formaldehyde for 10 min at room temperature. Cross-linking was quenched by addition of glycine to 0. 125 M. Cells were harvested by centrifugation and snap frozen until lysed. Following lysis, nuclei were isolated by centrifugation, and chromatin was sheared by sonication to approximately 500-bp size. After centrifugation to pellet debris, chromatin was divided into aliquots incubated overnight at 4°C with 2 μg of mouse anti-HIF-1α (#ab8366; Abcam) or anti-IgG (#sc-2025; Santa Cruz) antibody as a negative control. Antibody-conjugated protein-DNA complexes were precipitated by addition of protein A Sepharose beads (Santa Cruz), the immunoprecipitates were eluted, and the cross-links were reversed. The resulting DNAs were purified using QIAquick PCR purification kits (Qiagen) and analyzed by qPCR using iTaq universal SYBR green supermix (Biorad) and the Applied Biosystems prism real-time PCR system with the following primer pairs: BZLF1: FWD, 5′-GGCTGTCTATTTTTGACACCAGC-3′, and REV, 5-AAGGTGCAATGTTTAGTGAGTTACC -3′; and 4. 8-kbps upstream of Zp transcription initiation site (negative control); FWD, 5′-AGAAGGGAGACACATCTG-3′, and REV, 5′-AACTTGGACGTTTTTGGG-3’. A standard curve was generated from the threshold cycle (CT) of the input DNA diluted to 5%, 1%, and 0. 2% with distilled water containing 100 μg/ml sheared salmon sperm DNA (Ambion), with percent input bound calculated relative to this standard curve. Assays were performed in triplicate on two separate occasions. The 3-bp substitution mutations, HRE mt2 and HRE mt4, were introduced into the Zp HRE element in the EBV-containing BAC p2089 [36] by two-step, phage λ Red-mediated recombination essentially as previously described [80]. In brief, the I-SceI-Kan cassette present in pEPkan-S2 was PCR-amplified using the following primer pairs: HRE mt2: FWD 5’- AGGCATTGCTAATGTACCTCATAGACACACCTAAATTTAGgctGTCCCAAACCATGACATCACTAGGGATAACAGGGTAATCGATTT-3’ and REV 5’-CCAAGGCACCAGCCTCCTCTGTGATGTCATGGTTTGGGACagcCTAAATTTAGGTGTGTCTATGCCAGTGTTACAACCAATTAACC-3’; HRE mt4: FWD 5’-AGGCATTGCTAATGTACCTCATAGACACACCTAAATTTAGattGTCCCAAACCATGACATCACTAGGGATAACAGGGTAATCGATTT-3’ and REV 5’-CCAAGGCACCAGCCTCCTCTGTGATGTCATGGTTTGGGACaatCTAAATTTAGGTGTGTCTATGCCAGTGTTACAACCAATTAA-3’. The Zp sequence in these primers is underlined, with the base substitutions indicated in bold italicized small letters. These PCR products were electroporated into E. coli strain GS1783 into which BAC p2089 had been previously introduced, and inserted into p2089 by homologous recombination. Induction of the I-SceI activity encoded by GS1783 led to cleavage at the unique SceI site within the BAC. Intra-molecular recombination between the two copies of Zp resulted in precise removal of the inserted pEPkan-2 sequences, leaving behind one copy of Zp. Clones containing the desired HRE mutant BACs were initially identified by PCR screening and, subsequently, by DNA sequence analysis of the Zp and Zta-coding regions of the BAC. The mutant variants of p2089 were then thoroughly checked for absence of large deletions, insertions and rearrangements by analysis of multiple restriction enzyme fragment patterns as previously described [43,81] and for extraneous base-pair substitution mutations by high throughput sequence analysis as described below after recovery of the DNAs from mutant-infected 293T cell lines. 293T cells were transfected with twice CsCl2-purified BAC DNA and selected for hygromycin-resistance as previously described [43]. By 3-to-4 weeks post-transfection, all of the colonies of cells were GFP-positive. These clones were picked, grown up, and stored in liquid nitrogen. Their ability to produce infectious virus was determined as previously described [35] following transfection with plasmids that express the EBV Zta and gp110 proteins. The titers of the mutant virus stocks ranged from 104 to 105 green Raji units (GRU) /ml. We recovered the BAC DNAs from the HRE mutant-infected 293T cell lines by Hirt extraction as previously described [43] and introduced them into E. coli strain GS500 by electroporation. Two independent colonies obtained from each of the two mutant BACs were grown, and the BAC DNAs were isolated by alkaline lysis as previously described [43]. After purification through two cycles of centrifugation in CsCl2, the highly purified BAC DNAs were sequenced using an Ion Torrent PGM (Life Technologies). We aligned the sequencing reads to the B95. 8 reference strain of EBV (V01555) with Bowtie2 [82] using default alignment parameters and removing non-aligned reads. The resulting alignments were sorted using Samtools [83]. The Genome Analysis Toolkit (GATK) Unified Genotyper (https: //www. broadinstitute. org/gatk/guide/article? id=6201) [84–86] was used to detect genetic variations compared to the EBV reference. Since regions of repetitive DNA produce incorrect alignments [87] which can manifest in downstream analyses as apparent mutations, we further investigated called mutations which occurred in the repetitive regions of the EBV genome (TRs, FRs, IR2, and IR3). A program termed EasyVariant was written and used to parse each alignment and its CIGAR string [83] that allowed both position-specific coverage depth to be calculated and percentage of each of the four nucleotides to be called at each position. Any position in which 50% or more reads indicated a mutation was treated as valid unless it occurred in a repeat region where it was likely due to an incorrect alignment. We achieved sequence coverage depth of 15 or more reads over 93% and 97% of the unique regions of the genome for HRE mt2 and mt4, respectively. The expected mutations in the HRE (mt2 and mt4) were called as such in 100% of sequence reads, and the consensus base calls within the unique regions of the genome matched the reference genome. We also performed conventional Sanger sequencing at four locations where some reads (but still less than 50%) indicated a possible frameshift mutation; in each of these cases, no mutation was found. SNU-719 transcript data, taken from Strong et al. [30], were analyzed using the RSEM algorithm (strand-specific option) for quantification of human gene expression [88] to calculate the relative levels of HIF-1α, HIF-2α, and HIF-3α RNA present in these cells. We likewise analyzed for relative expression of the HIF-αs the raw RNA-sequence reads obtained from four primary gastric carcinoma samples previously determined to contain high levels of EBV RNA [30]. These latter reads, generated through the NIH’s The Cancer Genome Atlas (TCGA) project, were obtained from the NCBI Sequence Read Archive (SRA035410, now available through the NCI Genomic Data Commons). The relative levels of the HIF-α RNAs present in primary, endemic, EBV+ Burkitt lymphomas were calculated from the data provided in Table S10 of Abate et al. [31]. The RNA expression levels of the genes shown in Fig 9A that had been generated from eight cell types ranging from naïve B cells to plasma cells (38 samples total) were retrieved from the previously reported microarray datasets [89–92]. These data were normalized using the GCRMA algorithm and visualized using GenomicScape (http: //www. genomicscape. com/microarray/browsedata. php? acc=GS-DT-2) [93]. NOK (clone #3) cells were grown in organotypic culture as previously described [45] with the cells grown at the air-liquid interface for 11 days. Whole-cell extracts were prepared from these rafts by homogenization with a pestle in RIPA buffer [150 mM NaCl, 1% Triton X-100,0. 5% sodium deoxycholate, 0. 1% SDS, 50 mM Tris (pH8. 0) ]. The resulting lysates were incubated on ice for 1/2 hour, sonicated, and centrifuged to remove debris. The supernatants were stored frozen until analyzed by immunoblotting. EBV+ B-cell lymphomas were generated in immunodeficient NSG (NOD/LtSz-scid/IL2Rγnull) mice (Jackson Labs) as previously described [46]. In brief, CD34-depleted human cord blood mononuclear cells (#CB117; AllCells) were infected in vitro with the M81 strain of EBV (2,000 GRU) by incubation at 37°C for 1. 5 h after which the infected cells were injected i. p. into 3- to 5-week-old NSG mice. Thirty-three days later, the mice were injected i. p. with 60 mg/kg of Hypoxyprobe (Hypoxyprobe) and sacrificed 1. 5 h later by cervical dislocation under isoflurane anesthesia. Portions of the harvested tumors, along with some internal organs as controls, were submerged in Optimal Cutting Temperature compound and flash-frozen in ethanol-dry ice. Other portions of tumors, along with internal organs, were formalin-fixed and paraffin-embedded for sectioning and mounted onto slides for IHC. EBV+ gastric cancer xenografts were generated by subcutaneous inoculation of 1x107 SNU-719 cells in Matrigel into the flanks of NSG mice. Thirty-three days later, the mice were injected i. p. with 60 mg/kg of Hypoxyprobe and sacrificed 1. 5 h later. Portions of the tumors were prepared as described above. P-values for the reporter assay data were determined by the Student’s t-Test method. The p-value for testing whether the distributions in Fig 12 were statistically different was determined by the Wilcoxon Rank-Sum Test using Mstat Statistical Software.

Most adults throughout the world are infected with Epstein-Barr virus (EBV), a human herpesvirus frequently associated in a latent state with some cancers of epithelial and B-cell origin such as nasopharyngeal carcinoma and Burkitt lymphoma, respectively. To develop an oncolytic therapy for treating patients with EBV-associated cancers, we need a method to efficiently induce synthesis of lytic EBV proteins. The EBV protein encoded by its immediate-early BZLF1 gene usually mediates the switch into lytic viral infection. We show here that HIF-1α, a cellular transcription factor that accumulates in cells when deprived of normal levels of oxygen, can induce lytic EBV infection. HIF-1α mediates this switch by directly binding to a specific sequence located within the BZLF1 gene promoter, activating its expression. Importantly, we also show that deferoxamine, an FDA-approved drug that inhibits degradation of HIF-1α, can induce synthesis of lytic EBV proteins in some EBV-positive epithelial and lymphocytic cell lines. These findings indicate that HIF-1α-stabilizing drugs, administered in combination with nucleoside analogues such as ganciclovir, may be helpful as part of a lytic-induction therapy for treating some patients with EBV-positive malignancies.

Abstract Introduction Results Discussion Materials and methods

blood cells medicine and health sciences immune cells pathology and laboratory medicine cardiovascular anatomy 293t cells pathogens immunology biological cultures microbiology cell differentiation dna transcription viruses developmental biology dna viruses immunologic techniques herpesviruses research and analysis methods white blood cells epstein-barr virus animal cells blood vessels medical microbiology gene expression microbial pathogens cell lines antibody-producing cells immunohistochemistry techniques cell biology b cells anatomy viral pathogens genetics biology and life sciences cellular types histochemistry and cytochemistry techniques organisms

2017

Hypoxia-inducible factor-1α plays roles in Epstein-Barr virus’s natural life cycle and tumorigenesis by inducing lytic infection through direct binding to the immediate-early BZLF1 gene promoter

Allosteric regulation has traditionally been described by mathematically-complex allosteric rate laws in the form of ratios of polynomials derived from the application of simplifying kinetic assumptions. Alternatively, an approach that explicitly describes all known ligand-binding events requires no simplifying assumptions while allowing for the computation of enzymatic states. Here, we employ such a modeling approach to examine the “catalytic potential” of an enzyme—an enzyme’s capacity to catalyze a biochemical reaction. The catalytic potential is the fundamental result of multiple ligand-binding events that represents a “tug of war” among the various regulators and substrates within the network. This formalism allows for the assessment of interacting allosteric enzymes and development of a network-level understanding of regulation. We first define the catalytic potential and use it to characterize the response of three key kinases (hexokinase, phosphofructokinase, and pyruvate kinase) in human red blood cell glycolysis to perturbations in ATP utilization. Next, we examine the sensitivity of the catalytic potential by using existing personalized models, finding that the catalytic potential allows for the identification of subtle but important differences in how individuals respond to such perturbations. Finally, we explore how the catalytic potential can help to elucidate how enzymes work in tandem to maintain a homeostatic state. Taken together, this work provides an interpretation and visualization of the dynamic interactions and network-level effects of interacting allosteric enzymes. We are interested in studying the “catalytic potential” of an enzyme—its capacity to catalyze a reaction—from a network-level perspective. An enzyme achieves its catalytic potential when all individual enzyme species are in an active form; an enzyme with allosteric regulation modulates its utilization of its catalytic potential based on ligand-binding events throughout the network in order to maintain a homeostatic state. Here, we propose that an enzyme’s utilization of its catalytic potential can be visualized by computing the fraction of total enzyme that is available to catalyze a reaction as a function of the adenylate energy charge (Fig 1A). In this section, we describe both of these properties and how they can be computed using enzyme modules and mass action kinetics. The energetic state of a cell can be measured using the adenylate energy charge [27], which represents the amount of high energy bonds available in the adenosine phosphate pool. The energy charge is given by Energy Charge = [ ATP ] + 1 2 [ ADP ] [ ATP ] + [ ADP ] + [ AMP ] (1) where [AMP], [ADP], and [ATP] represent the concentrations of those respective metabolites. Because of the number of reactions in which the adenosine phosphates participate, the energy charge is a systemic variable sensed by important enzymatic regulators [28,29] which can be more sensitive to perturbations than are reaction rates (Fig A in S1 File). To examine individual enzymatic reactions that are regulated by at least one metabolite in the combined adenosine phosphate pool (i. e. , AMP, ADP, or ATP) from a network-level perspective, we can compute properties of enzymes as a function of the energy charge. A kinetic model that explicitly represents each of the elementary steps for an enzymatic reaction (i. e. , an enzyme module) provides enough detail to compute the fraction of uninhibited enzyme primed to facilitate the conversion of substrate to product for enzymes allosterically regulated through effector molecules. This catalytically active fraction (fA) can be calculated for an enzyme from f A = ∑ i = 0 n R i + R i, A + R i, AS E total (2) where n is the number of enzymatic binding sites, Ri is the unbound enzyme in the active state (i. e. , not bound to inhibitors), Ri, A is the enzyme bound to the cofactor, Ri, AS is the enzyme bound to the substrate and cofactor, and Etotal is the total amount of enzyme. The subscript i represents the amount of activators bound to allosteric sites; for tetrameric structures like PFK and PYK, i ranges between 0 and 4 [30,31]. Here, we adopted the Monod-Wyman-Changeux (MWC) reaction framework [32] for PFK and PYK in which the allosteric activator and inhibitor can only bind to the relaxed and tense state, respectively. Both the energy charge and fA were computed from model simulations. We use mass action kinetics to model RBC glycolysis with enzyme modules (i. e. , explicitly representing the elementary reactions for ligand-binding) for HEX, PFK, and PYK (see Supplementary Material for the full reaction mechanism for each enzyme module). In the following sections, we detail the construction and validation of models with enzyme modules and examine each enzyme’s utilization of its catalytic potential in response to perturbations in ATP utilization. We constructed a model of RBC metabolism that comprises glycolysis, the Rapoport-Luebering (RL) Shunt, and the interaction of hemoglobin with 2,3-diphosphoglycerate [23]; the stoichiometric matrix for the network, all kinetic parameters, and the initial flux values are provided in S1 Data. This small-scale model allows us to study the regulatory effects on glycolysis. This model contains three allosterically regulated kinases for which enzyme modules were constructed (see Methods and Supplementary Material): hexokinase (HEX), phosphofructokinase (PFK), and pyruvate kinase (PYK). To validate each of the enzyme modules, we sought to introduce physiologically relevant perturbations that would affect the energy charge. Several external pressures—such as hypoxic conditions [33] or sheer stress experienced in vivo due to arterial constriction [34]—can result in increased release of ATP from RBCs in vivo, while internal ATP concentrations can drop by as much as 27% or 50% due to aging or the presence of acute disease states such as gastrointestinal tumors [35]. To model these behaviors, we perturbed the rate of ATP utilization (see Methods) to induce a systemic response that is qualitatively representative of the observed phenotypes (increasing and decreasing the value of the rate constant for the hydrolysis of ATP; see Methods). We built and tested models with each enzyme module individually, examining its utilization of its catalytic potential of the enzyme as the system returns to its original homeostatic point. To validate these models against previous experimental results reported in the literature, we make the assumption that the initial velocity of a reaction is proportional to the amount of catalytically active enzyme, fA [36]; the qualitative shape of a rate versus energy charge plot should then match that of an fA versus energy charge plot. The baseline RBC glycolytic model used to construct the models is based on nominal parameter values [23]. However, genetic variation in the human population leads to varying RBC metabolic dynamics in different individuals. Our next goal was therefore to explore the sensitivity of an enzyme’s catalytic potential to perturbations to model parameters. Because of its dependence on the energy charge and literature validation of its catalytic potential, PFK was used for an in depth exploration of the robustness of the catalytic potential. We first generated 50 models from randomly sampled, thermodynamically feasible concentrations values (see Methods) and perturbed the rate of ATP utilization. We examined the net rate of ATP usage (i. e. , total flux through ATP-producing reactions minus total flux through ATP-consuming reactions), the energy charge as a function of time, and the catalytic potential (Fig D in S1 File). From this analysis, we see that reaction rates (Fig E in S1 File) are not as sensitive to changes in ATP levels, while these changes are captured by the energy charge (Fig E in S1 File). The catalytic potential then allows us to incorporate this systemic information as we observe the response of PFK. However, while these randomized models were constructed with thermodynamically feasible metabolite concentrations, they do not necessarily represent physiologically feasible concentrations. Therefore, we further collected previously reported RBC and plasma metabolite levels from a series of individuals [20], enabling the construction of “personalized” RBC models (see Methods). We constructed personalized models using glycolytic metabolite concentrations and equilibrium constants for nine individuals from a previous study [20]. Using personalized models provides a sensitivity analysis that examines physiologically-feasible parameter values. The general qualitative trend for the catalytic potential plot of PFK was similar to the one using literature values (Fig 2A), but initial fA values were significantly lower in the personalized models (Fig 3A and 3B). In particular, the amount of active PFK for each individual reached a saturation point that was higher than the initial steady-state value in order to compensate for the increase in ATP utilization before returning to a final steady-state value. While we observe that there is little difference among the rate profiles (Fig 3C), we observe much greater differences in the catalytic potential plots (Fig 3A and 3B) and energy charge profiles (Fig 3D). Notably, the model for Individual #1 exhibited a much different response than the other eight personalized models (Fig 3A, 3B and 3D). We examined this behavior and determined that PFK is highly sensitive to the rate constants for the binding of ATP and F6P to PFK (outliers with over 99% confidence according to the Dixon’s Q test; see Methods for full details); these were the only rate constants that were deemed to be outliers out of all enzymatic reactions, showing that these rate constants are the parameters to which PFK is most sensitive. Finally, we examined how an enzyme’s utilization of its catalytic potential can be used to characterize the interplay among enzymes in the same model. We thus integrated the enzyme modules for all three kinases studied here (PFK, HEX, and PYK) into the base model and introduced the same ATP utilization perturbations. We examined the disturbance rejection capabilities of this complete model compared with models with fewer enzyme modules, noting increased regulation generally improved the ability of a model to maintain a homeostatic state (Fig A in S1 File) as expected [39–42]. The inclusion of multiple enzyme modules in the same model allows us to characterize how the three allosterically regulated enzymes interact in determining the system’s response to these perturbations through dynamic simulation. We characterized the catalytic potential of this complete model’s response to external perturbations (Fig 4A). We observed similar qualitative responses for each of the enzymes in the combined model as for each enzyme module individually (Fig 2). To examine the interplay between enzymes, we looked at phase portraits comparing the catalytically active enzyme fraction (fA) for each pairwise combination of enzymes (Fig 4B and S1 Video). We can see that as a greater fraction of PFK entered a more catalytically active state, a greater fraction of HEX become catalytically inactive; a similar behavior was observed for the PFK-PYK pair. We observed that HEX and PYK moved in tandem, with both enzymes moving into catalytically active or inactive states together. This behavior is likely due to the fact that these enzymes represent the boundaries of the system and therefore are linked in order to maintain system stability. The ability to mechanistically model cellular metabolism allows for the construction of predictive physiological models. However, the mechanistic results obtained from time-course plots can complicate the interpretation and analysis of systems-wide responses to relevant perturbations. To help provide a better method of elucidating this behavior, we built modularized glycolytic models with enzymes serving as regulators that allows for a new interpretation of the state of an enzyme—where it operates with respect to its maximum catalytic potential. These models were then validated against existing empirical data to understand the relationship between the catalytically active enzyme fraction and energy charge. Visualizing an enzyme’s utilization of its catalytic potential allowed for the analysis of important systems behaviors. The results presented here have two primary implications. First, we have studied glycolysis from a perspective in which enzymes are regulators. Individual kinases serve as tuning dials for the system by sensing changes in energy charge and modulating their utilization of their catalytic potentials in order to return the system to a homeostatic state. If the energy charge dropped, then mass action kinetics would dictate that more flux would be pushed through a reaction that produces ATP in order to increase the energy charge. The response of PFK showed that its regulation is strong enough to overcome the dynamics that would result from these mass action trends alone. HEX behaves as is expected due to mass action (a lower energy charge results in a reduced fraction of catalytically active enzyme), but the observed behavior of PYK is opposite what would be expected based on the law of mass action. A decrease in energy charge would intuitively result in more catalytically active PYK since that would then result in more ATP. The literature reports this expected behavior for initial velocity of PYK [38]. However, these assays did not contain FDP, an allosteric activator of PYK. We observed that an increase in energy charge led to an initial increase in FDP concentration and a corresponding increase in the amount of PYK in the catalytically active form (Figs 2C and 4A). These plots suggest that the regulation of PYK by FDP leads to this unintuitive behavior. Second, we have shown that examining an enzyme’s catalytic potential can provide additional insight into how metabolic networks maintain a homeostatic state following physiologically-relevant perturbations. A small-scale model that explicitly accounted for the regulatory mechanisms of the three glycolytic kinases allowed us to directly investigate the interplay among these three enzymes (S1 Video). When we applied this metric to examine the response of personalized models to ATP utilization perturbations, we observed differences that were not apparent simply from the rate profile. The kinases modulated the response of the system, as demonstrated by examining individual parameterization of personalized models (Fig 3). Through an examination of how PFK operates with respect to its catalytic potential, we were able to gain insight into how the regulator within a model is tuned in different individuals in order to maintain homeostasis (Fig 3A, 3B and 3D), a behavior that was not discernible through more typical metrics like rates of reaction (Fig 3C). Hence, the catalytic potential plots describe how enzymatic entities respond to system-wide changes in order to drive the cell towards a homeostatic state after environmental alterations. Upon further investigation, we determined that the utilization of catalytic potential for Individual #1 was different than the others due to differences in the binding affinities of ATP and F6P to PFK, indicating that the PFK module was most sensitive to these parameters. Thus, the catalytic potential helped provide insight into how subtle differences among individuals can lead to differing systemic responses to perturbations that push the system away from the homeostatic state. The use of kinetic models to study the dynamics of cellular metabolism presents many well-documented challenges and limitations [43,44]. Many of these issues revolve around attempting to parameterize biochemical processes that may not be well understood [43], one of the reasons that we adopted several simplified approaches in this study. Here, we employed the use of so-called “enzyme modules” (explicit representations of all ligand-binding reactions [21,22]) for the allosterically regulated kinases in glycolysis, a modeling formulation which allowed us to compute the catalytically active enzyme fraction. We used the same reaction mechanisms (predicted by a computational method) from the previous study using enzyme modules [22] because our focus here was on interpreting the output from enzyme modules. Many alternative mechanisms exist [45], and the impact of employing different mechanisms on computing fA could be explored in the future. Mass action kinetics were used for the other enzymes in the network and represent an approximation previously examined in the literature [22]. While the final reaction step for each enzyme module could be represented by two bimolecular steps [46], we have used a simplified termolecular step (i. e. , all bound molecules are released in a single reaction step) due to a lack of high-confidence kinetic parameters. Kinetic models of metabolism are generally stiff systems [12], and the inclusion of enzyme modules exacerbates this issue due to the addition of several reactions with concentration variables that span several orders of magnitude (PFK module: 24 reactions; HEX module: 8 reactions; PYK: 34 reactions; see Supplementary Material for full mechanisms). Finally, the size of a model inevitably impacts the behavior of a model; we have chosen to draw our system boundary at the end of glycolysis, thereby not accounting for any downstream effects on the activity of PYK (such as flux leaving the pyruvate node and entering the citric acid cycle remnant reactions). The RBC metabolic network consists of well-studied metabolic pathways and their associated metabolites. New methods for the visualization of regulatory behaviors—such as the catalytic potential plot introduced here—can lead to new insights and discoveries. We have evaluated the utilization of an enzyme’s catalytic potential as a sensor which can be used to visualize the state of that enzyme in the context of the metabolic network. Viewing enzymes as regulators through which we can tune the system response opens the door for us to investigate what the optimal state might be and how that state helps maintain homeostasis. The base glycolysis network included all 10 glycolytic enzymes and lactate dehydrogenase; the complete stoichiometric matrix is provided in S1 Data. Reaction rates were defined using mass action kinetics, representing enzyme catalysis as a single step. These simplified reactions were systematically replaced with enzyme modules following the procedure outlined by Du et al. [22]. Additionally, a phosphate exchange reaction was incorporated into the glycolytic network utilizing parameters obtained from Prankerd et al. [48]. Similarly, the Rapoport-Luebering Shunt was included in some models to account for the presence of hemoglobin, whose binding to oxygen is regulated by 2,3-diphosphoglycerate (2,3-DPG). Incorporation of this shunt was accompanied by parameter changes as previously described [23]. All model parameters are provided in S1 Data. Regulation was manually incorporated into the enzyme reactions. Initial conditions from the glycolysis and hemoglobin MASS toolbox example data were used in conjunction with equilibrium constants which were obtained from from various sources (see Supplementary Material). These values were subsequently utilized to solve for new kinetic parameters by setting the following constraint: d x → d t = S · v (x; k) = 0 (3) where d x → / d t is the concentration rate of change with respect to time for metabolites, S is the stoichiometric matrix, and v (x; k) is a vector containing reaction fluxes as a function of metabolite concentrations (x) and rate constants (k). The parameters for all enzyme modules were determined using the methods described by Du et al. [22]. In short, the workflow includes: (1) defining all ligand-binding events and their associated equilibrium constants, (2) symbolically solving the resultant steady state mass balance, (3) solving for the pseudo-first-order elementary rate constant (kPERC) [23] of each enzymatic reaction using the overall flux state as a constraint, and (4) using the estimated kPERCs to approximate steady state concentration values for each enzyme form (e. g. , enzyme bound to all combinations of ligands). The kPERC for a reaction is estimated using the following equation: k i = v i Π i reactants i - Π i products i / K eq (4) where ki is the kPERC for reaction i and vi is the flux through that reaction [23]; reactions assumed to be irreversible were assigned an arbitrarily high K − eq (Mathematica allows for the assignment of infinity). We constructed a total of ten different models with varying amounts of regulation, spanning from the base glycolytic model with no enzyme modules (and therefore no regulation) to a model with three enzyme modules and the Rapaport-Luebering Shunt. The remaining models represented each combination of the three kinase modules. Enzyme module incorporation was accompanied by the deletion of the original single-step reaction in order to avoid redundant reactions. Stability for all systems was verified by simulating the network and ensuring that a steady-state point was found for all metabolites. In order to mimic a physiologically-relevant perturbation away from the homeostatic state, we simulated a 50% increase in ATP utilization for 1,000 hours and a 15% decrease in ATP utilization [33–35]. These magnitudes were chosen because they resulted in observable changes in the energy charge which could then be used to qualitatively assess the impact on the system. Changes in ATP utilization were applied by changing the rate (kATP) associated with ATP hydrolysis: ATP + H 2 O ⇌ k ATP ADP + H + P i (5) where Pi represents inorganic phosphate which was modeled as a variable quantity to allow the system to respond to these perturbations. Increasing this rate decreases the amount of available ATP and ADP. We calculated the sum of squared error (SSE) for each model in order to quantify the total deviation of the output from its setpoint, which is zero. The resulting quantity (i. e. , the SSE) is compared between models, with a smaller value indicating better disturbance rejection capabilities. Rate pools for enzymes were defined as the rate at which enzyme produced product. This was accomplished by defining a pool from the product’s ODE consisting solely of the terms contributing to product formation. In other words: rate enzyme = ∑ v formation (7) where vformation represents the forward rate of the enzyme reaction and possesses units of mmol/L ⋅ hr. Defining the rate pools in this manner neglected effects of reversible reactions contributing to the formation of product. A negative value corresponds to a net-consumption of ATP.

Enzymatic rate laws have historically been used to simulate the dynamics of complex metabolic networks with regulated reactions represented by allosteric rate laws. Here, we use detailed elementary reaction descriptions of regulatory enzymes that allow for the explicit computation of the fraction of the enzymes that are in a catalytically-active state. The fraction of the enzyme that is in the active state represents the time-dependent utilization of the enzyme’s “catalytic potential, ” its capacity to catalyze a reaction. We apply this interpretation to red blood cell glycolysis, examining how three key kinases with allosteric regulation modulate their utilization of their catalytic potential based on ligand-binding events throughout the network in order to maintain a homeostatic state. We then examine how an enzyme modulates its utilization of its catalytic potential using personalized data as a case study, visualizing the systems-level properties of a kinetic model.

Abstract Introduction Results Discussion Methods

allosteric regulation medicine and health sciences enzymes metabolic processes metabolic networks enzymology glycolysis physiological processes homeostasis network analysis enzyme metabolism enzyme kinetics enzyme chemistry computer and information sciences proteins enzyme regulation biochemistry physiology biology and life sciences metabolism

2018

Network-level allosteric effects are elucidated by detailing how ligand-binding events modulate utilization of catalytic potentials

The neurotrophin-3 (NT-3) receptor tropomyosin receptor kinase C (TrkC/NTRK3) has been described as a dependence receptor and, as such, triggers apoptosis in the absence of its ligand NT-3. This proapoptotic activity has been proposed to confer a tumor suppressor activity to this classic tyrosine kinase receptor (RTK). By investigating interacting partners that might facilitate TrkC-induced cell death, we have identified the basic helix-loop-helix (bHLH) transcription factor Hey1 and importin-α3 (karyopherin alpha 4 [KPNA4]) as direct interactors of TrkC intracellular domain, and we show that Hey1 is required for TrkC-induced apoptosis. We propose here that the cleaved proapoptotic portion of TrkC intracellular domain (called TrkC killer-fragment [TrkC-KF]) is translocated to the nucleus by importins and interacts there with Hey1. We also demonstrate that Hey1 and TrkC-KF transcriptionally silence mouse double minute 2 homolog (MDM2), thus contributing to p53 stabilization. p53 transcriptionally regulates the expression of TrkC-KF cytoplasmic and mitochondrial interactors cofactor of breast cancer 1 (COBRA1) and B cell lymphoma 2–associated X (BAX), which will subsequently trigger the intrinsic pathway of apoptosis. Of interest, TrkC was proposed to constrain tumor progression in neuroblastoma (NB), and we demonstrate in an avian model that TrkC tumor suppressor activity requires Hey1 and p53. The neurotrophins nerve growth factor (NGF), brain-derived neurotrophic factor (BDNF), neurotrophin-3 (NT-3), NT-4/5, and their respective receptors neurotrophin receptor p75 (p75NTR) and tropomyosin receptor kinases (TrkA), B, and C have been notably studied for their critical role in neurodevelopment [1]. Yet as TrkA, B, and C are tyrosine kinase receptors (RTKs), their deregulated functions in cancer have been investigated [2]. The overall view is that their kinase activity confers them the ability to activate mitogen-activated protein kinase (MAPK) and phosphoinositide 3-kinase (PI3K) /AKT pathways known to promote cell survival, proliferation, and differentiation under physiological conditions and to contribute to tumor progression when constitutively activated in cancers [2]. The kinase domains of TrkA, B, and C are indeed involved in oncogenic translocations or mutated in cancers (for review [2]). In line with the pharmaceutical rush to design antitumoral treatments based on RTK inhibition, drugs targeting TrkA, B, and C have been under development [3]. Nevertheless, TrkC expression has been paradoxically associated with favorable outcome in pediatric neoplasia, namely neuroblastoma (NB) and medulloblastoma, and was more recently shown to act as a tumor suppressor in colon cancer ([4] and for review [5–8]). We and others have indeed proposed that TrkC has a dual functionality: (i) In presence of its ligand NT-3, TrkC behaves as a classical RTK, transducing positive signals; (ii) in absence of NT-3, TrkC does not stay inactive but rather triggers apoptosis [9,10]. TrkC thus belongs to the functional family of" dependence receptors." These receptors play a crucial role in constraining the adequate number of cells in a tissue in which the ligand is expressed in a limited amount during neurodevelopment but also during tumorigenesis: Cells in excess that carry an unbound dependence receptor undergo apoptosis [11]. It was demonstrated in different types of tumors that (i) the silencing of the dependence receptor by epigenetic mechanisms or genetic alterations or (ii) the overexpression of the ligand confers to the tumor cells a survival selective advantage: The dependence receptor is then no longer able to trigger apoptosis. TrkC expression was indeed shown to be epigenetically silenced in colon tumors [4,6]. Along the same line, we also demonstrated that a large proportion of high-grade NB tumors shows an autocrine production of NT-3 as a mechanism to constitutively block TrkC proapoptotic function. It was thus proposed that interfering with ligand–receptor (NT-3/TrkC) interaction, either by gene silencing or the use of a blocking antibody, is associated in different animal models with the inhibition of tumor growth and metastasis [12]. The mechanism for TrkC proapoptotic activity has been investigated in recent years [9,10,13]. Upon ligand withdrawal, TrkC appears to be cleaved by caspase-like proteases at 2 sites (D495 and D641) within its intracellular domain, and the released fragment (TrkC 496–641, called the" killer-fragment" [TrkC-KF]) is necessary and sufficient to promote apoptosis. We demonstrated recently that this fragment interacts with cofactor of breast cancer 1 (COBRA1), which shuttles TrkC-KF to the mitochondria [13]. Once at the mitochondria, TrkC-KF and COBRA1 activate B cell lymphoma 2–associated X (BAX) and induce mitochondrial outer membrane permeabilization (MOMP), the release of cytochrome c, and the subsequent apoptosome activation [13]. Here, we show that TrkC-KF is not only cytoplasmic as described previously but is also observed in the nucleus. TrkC-KF is translocated to the nucleus by importins. A 2-hybrid screen allowed us to identify that TrkC-KF then interacts with Hey1, a basic helix-loop-helix (bHLH) transcription factor originally described as an effector of the NOTCH pathway. Hey1 and TrkC-KF bind on mouse double minute 2 homolog (MDM2) promoter and negatively regulate MDM2 transcription. This decrease of MDM2 expression favors p53 stabilization, which triggers the transcription of TrkC proapoptotic partners acting at the mitochondria. We finally show in an avian model of NB tumor progression that Hey1- or p53-silencing abrogates TrkC tumor suppressor activity. We have previously shown that in absence of its ligand NT-3, TrkC is cleaved by caspase at 2 sites (D495 and D641) within its intracellular domain, leading to the release of several intracellular fragments. This caspase-dependent cleavage can be detected both in vitro and in vivo and is required for apoptosis induction, since the mutation of the caspase sites inhibits apoptosis induced by TrkC [9,12,13]. TrkC cleavage by caspases leads to the generation of 3 fragments: TrkC 1–495, TrkC 496–641, and TrkC 642–825 (Fig 1A). In various cell lines, including the murine Neuro2a (N2A) and human SHEP NB cell lines enforced expression of the internal caspase-generated fragment TrkC 496–641 (named TrkC-KF) was associated with cell death induction, while TrkC 1–495 and TrkC 642–825 displayed no proapoptotic activity [9,13]. In addition to its mitochondrial localization described earlier [13], the green fluorescent protein (GFP) –tagged TrkC-KF (TrkC-KF-GFP) was detected in the nucleus of N2A cells (Fig 1B). As a control, full-length TrkC (TrkC-FL-GFP), the C-terminal cleavage fragment (TrkC-642-825-GFP), and the intracellular fragment of an unrelated receptor—Neogenin (Neo-IC-GFP) —were mostly detected outside the nucleus of transfected cells (Fig 1B). We used GFP-fused fragments as none of the commercial antibodies or antibodies we generated were able to detect endogenous TrkC-KF. We verified that Flag-tagged TrkC-KF was also observed both in the cytoplasm and in the nucleus upon cellular fractionation (S1A Fig). In a yeast 2-hybrid screen using TrkC-KF as bait and a mouse embryonic cDNA library as prey, we identified importin-α3 (karyopherin alpha 4 [KPNA4]) (S1B Fig) [13]. Importins are cargo proteins shuttling cytoplasmic proteins into the nucleus [14,15]. A proximity ligation assay (Duolink) using a pan-importin antibody and an anti-GFP antibody allowed us to detect a close interaction between TrkC-KF-GFP and endogenous importins (Fig 1C and 1D), suggesting that TrkC-KF is interacting with importins. We thus treated N2A cells with Ivermectin, a pan-importin inhibitor, and performed a fractionation experiment (Fig 1E and S1C Fig). As a control, we used a version of Neo-IC deleted for its nuclear export sequence (Neo-IC-ΔNES) but with an intact nuclear localization sequence (NLS), which is mostly localized in the nucleus [16]. We observed that the amount of TrkC-KF was greatly reduced in the nucleus upon treatment with Ivermectin, while the cytoplasmic pool was not significantly affected (Fig 1E and S1C Fig). As a positive control, Neo-IC nuclear translocation was also affected by Ivermectin treatment. TrkC-KF thus appears to be shuttled in the nucleus by importins. Importins need to first recognize NLSs in the proteins they are supposed to shuttle [14,15]. Two putative NLSs could be mapped in TrkC-KF sequence, and we thus generated constructs bearing 1 (KFΔNLS1) or the 2 (KFΔNLS1/2) mutations of these putative sites (Fig 1F). While the mutation of NLS1 had no effect on TrkC-KF nuclear translocation, mutation of NLS1/2 greatly reduced the amount of TrkC-KF in the nuclear fraction of transfected SHEP cells (Fig 1G and S1D Fig). Furthermore, the mutation of NLS1/2 (TrkC-KFΔNLS1/2) is sufficient to partially but significantly inhibit TrkC proapoptotic activity (Fig 1H) without affecting its functionality. Indeed, TrkC-KFΔNLS1/2 is able to bind COBRA1, its cytoplasmic partner, as wild-type TrkC-KF does when overexpressed in cells (S1E Fig), suggesting that this mutant is still functional. TrkC-KF nuclear translocation seems therefore necessary for its proapoptotic activity in SHEP cells. In addition, no nuclear export sequence (NES) could be found in TrkC-KF, suggesting that once in the nucleus, TrkC-KF does not return in the cytoplasm. To monitor the role of TrkC-KF in the nucleus, we investigated whether it is able to transactivate gene transcription. To first assay this, TrkC-KF was fused to a Gal4 DNA-binding domain (DBD), and SHEP cells were forced to express TrkC-KF-Gal4DBD together with a construct encoding a luciferase reporter gene under the control of the upstream activating sequence (UAS) -GAL4 promoter. As shown in Fig 1I and S1F Fig, TrkC-KF-Gal4DBD is unable to transactivate the UAS-GAL4, unlike deleted in colorectal cancer intracellular domain (DCC-IC), as shown previously [17]. This result suggests that TrkC-KF has no intrinsic transcriptional activity per se. As TrkC-KF does not seem to have a transcriptional activity, we looked for putative nuclear interacting partners in the 2-hybrid screen mentioned in Fig 1, using TrkC-KF as bait. We focused on Hey1, which was identified as a putative partner of TrkC-KF in the screen (S1B Fig). Hey1 is a transcription factor that belongs to the bHLH-Orange (bHLH-O) family of transcriptional repressors, together with Hey2 and HeyL [18]. NOTCH pathway activation increases Hey1 expression, leading to the transcriptional inhibition of downstream targets. Therefore, Hey1 is a critical effector of the NOTCH pathway, being involved in cardiac and vascular development [19]. Hey1 is mostly nuclear but has also been detected in the cytoplasm [20]. We showed by confocal microscopy that TrkC-KF-GFP colocalizes with Hey1 tagged with red fluorescent protein (Hey1-RFP) in the nucleus of N2A cells (Fig 2AB). Silencing Hey1 with a designed small interfering RNA (siRNA) was associated with a strong reduction of Hey1 mRNA (S2A Fig) or protein (S2B Fig) without affecting the level of other members of the bHLH-O transcription factor family members Hey2 and HeyL (S2A and S2C Fig). We further showed in SHEP cells that TrkC-KF-GFP and endogenous Hey1 are interacting, by using an anti-Hey1 antibody in a proximity ligation assay (Duolink) (Fig 2C and 2D). No signal of interaction was detected when TrkC 642–825 was used instead of TrkC-KF or when Hey1 was silenced with an siRNA (Fig 2C and 2D). Furthermore, TrkC-KF-GFP nuclear localization was not affected by invalidation of endogenous Hey1 with the siRNA (S2D and S2E Fig). Thus, Hey1 specifically interacts with TrkC-KF and does not seem to be required for TrkC-KF nuclear translocation. We next confirmed by coimmunoprecipitation performed in human embryonic kidney 293 T (HEK293T) cells that Hey1 interacts with TrkC-KF-GFP and full-length TrkC (TrkC-FL-GFP) (Fig 2E). Furthermore, HeyL, a transcription factor closely related to Hey1, fails to interact with TrkC-KF of TrkC-FL, showing that TrkC specifically interacts with Hey1 only (Fig 2E). Interestingly, when cotransfected with Hey1, TrkC-KF-GFP was markedly detected in the nucleus in the DNA-bound fraction (Fig 2F). Thus, Hey1 interaction with TrkC-KF allows their joint binding to DNA complex. In a previous study, we set up conditions in order to transiently express TrkC-FL without inducing an artefactual dimerization and subsequent activation of the kinase domain, as it may be the case with overexpressed RTKs [12,13]. In this setting, we observed in absence of NT-3 that the expression of TrkC-FL or TrkC-KF induces cell death in various cancer cell lines, including N2A NB cells [13]. We show here that silencing of Hey1 by RNA interference abrogates cell death induced by TrkC-FL or TrkC-KF in N2A cells (Fig 3A). As a control, the apoptosis induced by another dependence receptor, Patched (Ptc), is not affected by Hey1 silencing (Fig 3A). Rather than forcing TrkC expression, we investigated whether cell death induced upon NT-3 withdrawal is also dependent on Hey1. We have previously shown that SHEP cells are expressing both NT-3 and TrkC and that silencing of NT-3 is associated with TrkC-induced apoptosis in these cells [12]. We therefore silenced Hey1 in NT-3-depleted SHEP cells. As shown in Fig 3B, caspase-3 activation induced by NT-3 deprivation is similarly abrogated by cosilencing of endogenous Hey1. We had also previously demonstrated that various other NB cell lines overexpress NT-3 [12]. Among them, CLB-Ga [21] and LAN6 [22] express NT-3 within the same range as SHEP cells (S3A Fig and [12]). Despite the fact that LAN6 also expresses another Trk receptor, TrkA (S3A Fig), we demonstrate that interfering with endogenous NT-3 by RNA interference also triggers caspase-3 activity in LAN6 cells and that this caspase-3 activity can also be abrogated by cosilencing of Hey1 (S3B and S3C Fig). As p75NTR has been suggested to mediate TrkC-induced apoptosis [10], we wondered whether its cosilencing could abrogate the apoptosis induced by NT-3 deprivation in SHEP cells as well as Hey1 did. As shown in S3D and S3E Fig, the caspase-3 activity measured by endogenous NT-3 invalidation is not altered upon treatment with an siRNA targeting p75NTR. p75NTR may thus be dispensable, at least in our model, for the apoptosis triggered by TrkC via Hey1 in absence of NT-3. In order to work with cells constitutively knock-out for Hey1, we then invalidated endogenous Hey1 in SHEP cells by clustered regularly interspaced short palindromic repeat (CRISPR) /CRISPR-associated protein 9 (CAS9) editing and obtained 2 independent clones in which Hey1 expression was fully abolished. As a control, we used clones that had undergone the same selection process but without the transfection of the guide RNA (gRNA) (Fig 3C). We labeled the various SHEP clones and parental cells with an anticleaved caspase-3 antibody after siRNA NT-3 or siRNA control treatment. Hey1 knock-out clones displayed a much-reduced staining compared to the clones still expressing Hey1 (Fig 3D and 3E). Along this line, in a WB with an anticleaved poly [ADP-ribose] polymerase (PARP) antibody, we observed that PARP cleavage (cPARP) is greatly reduced in Hey1 knock-out clones upon siRNA NT-3 treatment (Fig 3F and S3F Fig). Altogether, these data indicate that the transcription factor Hey1 acts as a specific proapoptotic partner/effector of endogenous TrkC. Of interest, Hey1 has been previously identified in a screen aimed at finding new activators of p53 [23]. This study indeed demonstrated that Hey1 stabilizes p53 by down-regulating the expression of the p53 antagonist, MDM2. Along this line, we were able to detect an increase in the amount of p53 protein in SHEP cells forced to express Hey1, and this increase was more important when Hey1 was coexpressed with TrkC-FL or TrkC-KF (Fig 4A and S4A Fig). Conversely, coexpression of Hey1 with the intracellular uncleavable form of TrkC (TrkC-IC-D495N/D641N [TrkC-IC-DM]) did not increase p53 protein amount in SHEP cells (Fig 4A and S4A Fig). In order to determine whether p53 could be involved in the apoptosis mediated by TrkC, we silenced p53 by siRNA and assessed whether TrkC-FL or TrkC-KF could still be proapoptotic. Of interest, silencing p53 abrogates TrkC-FL- or TrkC-KF-mediated apoptosis (Fig 4B and S4B Fig). We then took advantage of HCT116 cells, which have been knock-out for p53, and their parental wild-type counterparts [24]. In p53 constitutively knock-out HCT116 cells, TrkC-FL and TrkC-KF were both unable to trigger apoptosis compared to what is seen in p53 wild-type HCT116 cells (Fig 4C). Again, rather than forcing TrkC expression, we silenced NT-3 in SHEP cells. As shown in Fig 4D and S4C Fig, silencing of NT-3 by siRNA is associated with an increased p53 protein level, whereas this is not the case when NT-3 and Hey1 are cosilenced. Similarly, cosilencing of NT-3 and p53 by RNA interference blocked NT-3 deprivation-induced apoptosis, demonstrating that p53 is necessary for TrkC apoptotic signaling (Fig 4E). As shown in S4D and S4E Fig, we also confirmed that Hey1 and p53 are necessary to unliganded TrkC–induced apoptosis in another NB cell line that expresses both NT-3 and TrkC: CLB-Ga cells (S3A Fig and [12]). We then used a chemical inhibitor of p53-dependent transcriptional activation, pifithrin-α [25], and treated SHEP cells with an siRNA targeting NT-3 or an siRNA control. We detected an increased amount of the apoptotic cPARP fragment in siRNA NT-3-treated cells, and this effect was reversed upon treatment with pifithrin-α (Fig 4F and S4F Fig). This result suggests that p53 transcriptional activation is required to mediate TrkC-induced apoptosis. Finally, in order to determine whether p53 stabilization is mediated by Hey1 transcription repressor function, we coexpressed TrkC-KF with a mutant version of Hey1 bearing 3 point mutations (Hey1-RK3: R50K, R54K, R62K). This triple mutation has been shown to affect Hey1 DNA-binding basic domain and consequently Hey1 transcriptional activity [26]. We observed that TrkC-KF when expressed with Hey1-RK3 is no longer able to induce p53 stabilization, supporting the view that Hey1 transcriptional activity is required for p53-dependent TrkC-KF proapoptotic activity (Fig 4G and S4G Fig). Because we (i) failed to detect any interaction between MDM2 protein and neither Hey1 nor TrkC-KF by proximity ligation assay (S5A Fig), (ii) observed that Hey1 DBD appears to be important for TrkC-KF/Hey1-mediated stabilization of p53 (Fig 4G), and (iii) identified MDM2 in a chromatin immunoprecipitation sequence (ChIP-Seq) aimed at screening Hey1 binding sites in Hey1 overexpressing cells (Gene Expression Omnibus [GEO] accession number GSE60699 [27]), we hypothesized that Hey1/TrkC-KF may transcriptionally regulate MDM2 expression. We first measured MDM2 expression by quantitative real-time PCR (RT-QPCR) in SHEP cells transfected with various expression plasmids. As described previously by Huang and colleagues, forced expression of Hey1 is able to deregulate MDM2 expression [23]. Of interest, TrkC-FL itself is also able to deregulate MDM2 expression (Fig 5A). As a control, the TrkC 642–825 fragment, the uncleavable form of TrkC (TrkC-DM), TrkC-FL in presence of caspase inhibitor z-vad, or an unrelated overexpressed receptor Ptc does not significantly alter MDM2 expression (Fig 5A). We further assessed the importance of Hey1 in TrkC-mediated MDM2 repression. As shown in Fig 5B, silencing of endogenous Hey1 by an siRNA in SHEP cells, forced to express either TrkC-FL or TrkC-KF, restores MDM2 expression. TrkC-FL- and TrkC-KF-mediated down-regulation of MDM2 was also observed at the protein level, as shown by western blot on transfected SHEP cells (Fig 5C and S5B Fig). Again, as a control, TrkC 642–825, TrkC-DM, or Ptc had no effect on MDM2 protein level (Fig 5C and S5B Fig). As illustrated in Fig 5D, the MDM2 gene has 2 promoters and an enhancer box (E-box) described as a putative binding site for various transcription factors, including bHLH-O factors like Hey1 [28]. We designed various primers all along the MDM2 promoter region and performed chromatin immunoprecipitation (ChIP) on SHEP cells expressing Flag-tagged versions of TrkC-KF and/or Hey1. ChIP with an antibody targeting endogenous Hey1 resulted in a slight enrichment of the promoter region amplified by primers located in close proximity to the E-box (Fig 5E). As a negative control, no enrichment was observed after the use of primers designed in 5′ or in 3′ of MDM2 promoter region (Fig 5E). Interestingly, TrkC-KF favors endogenous Hey1 binding to MDM2 promoter, as observed by the increased DNA enrichment in TrkC-KF-Flag transfected cells when compared to nontransfected cells expressing endogenous Hey1 only (Fig 5E). When SHEP cells were forced to express TrkC-KF-Flag or Hey1, a similar enrichment of the same promoter region was observed when chromatin was pulled down with either an anti-Flag antibody (targeting TrkC-KF) or an anti-Hey1 antibody (targeting Hey1) (Fig 5F). Together, these results support the view that TrkC-KF and Hey1 interact and bind to the same promoter region near the E-box of MDM2 promoter. To more formally address this question, we silenced Hey1 in TrkC-KF-Flag-expressing SHEP cells. Silencing of Hey1 fully reversed the DNA enrichment observed, indicating that TrkC-KF binding on MDM2 promoter is dependent on its interaction with Hey1 (Fig 5G and 5H). Finally, to assess direct binding of Hey1 and TrkC-KF to the MDM2 promoter E-box, we proceeded to an oligonucleotide pull-down assay using biotin-labeled double-stranded oligonucleotides homologous to the promoter region spanning MDM2 E-box. Oligonucleotides were mutated (mut) or not (WT) on the E-box sequence (Fig 5I) and incubated with lysates of SHEP cells expressing Hey1-Flag and TrkC-KF-GFP. As illustrated in Fig 5J and 5K, we could demonstrate the association of Hey1 (Fig 5J) and TrkC-KF (Fig 5K) with the oligonucleotide containing the wild-type E-box and much less with the mutated E-box oligonucleotide control. The number of bound oligonucleotides is increased when both Hey1 and TrkC-KF are expressed (Fig 5J and 5K). Conversely, silencing of endogenous Hey1 strongly inhibits TrkC-KF binding to the oligonucleotides corresponding to MDM2 promoter (Fig 5L). These data further confirm the association of Hey1 with TrkC-KF in the promoter region spanning MDM2 E-box. Together, these results show that TrkC-KF and Hey1 interact on MDM2 promoter and inhibit MDM2 transcription. We described in a previous study the shuttling of TrkC-KF at the mitochondria by COBRA1, in which both partners activated BAX and induced MOMP, the subsequent release of cytochrome c, and apoptosome activation [13]. What is then the role of the nuclear pathway and p53 stabilization by TrkC-KF/Hey1 interaction? Are the mitochondrial and the nuclear pathways redundant, or are they sequentially activated? We first expressed TrkC-FL in N2A cells and invalidated Hey1 by siRNA to abrogate cell death. We observed that in this setting, transient overexpression of COBRA1 is sufficient to largely restore apoptosis (S5C Fig). Conversely, Hey1 expression does not significantly restore cell death upon COBRA1 silencing (S5D Fig). These results suggest that the nuclear Hey1/p53 pathway is acting upstream the COBRA1/BAX mitochondrial pathway. We made the hypothesis that p53 activation may transcriptionally supply the proteins that are essential for the mitochondrial pathway triggered by TrkC. Indeed, COBRA1 promoter has been previously identified as a target of p53 in a genome-wide ChIP assay [29]. We measured COBRA1 expression by RT-QPCR and observed that invalidation of endogenous NT-3 by RNA interference (i. e. , the activation of TrkC/Hey1/p53 pathway) increases the amount of COBRA1 mRNA in SHEP cells (S5E Fig). This effect was reversed upon coinvalidation of NT-3 with Hey1 or p53 (S5E Fig). This result suggested that, indeed, p53 is important to allow the expression of COBRA1. In order to determine whether p53 is responsible for this transcriptional up-regulation of COBRA1, we identified 2 putative p53 binding sites [30] on the COBRA1 promoter and designed various pairs of primers spanning different regions of the COBRA1 promoter (S5F Fig). The chromatin of SHEP cells transfected with either control or TrkC-KF and Hey1 was pulled down with an anti-p53 antibody, and the region encompassing the 2 putative p53 sites was more amplified than the 5′ or 3′ region of the promoter (S5G Fig). These results suggest that p53 indeed binds to the COBRA1 promoter and contributes to the supply of COBRA1 proteins in the cytoplasm so that a pool of TrkC-KF fragments produced by the caspase cleavage in the cytoplasm can be shuttled at the mitochondria by COBRA1 proteins. We also showed in our previous study that TrkC-KF and COBRA1, once anchored at the mitochondrial membrane, activate BAX—but not B cell lymphoma 2 killer (BAK) —to trigger MOMP [13]. BAX is a well-characterized target of p53 [31], so we also designed pairs of primers among which 1 pair spanned the p53 binding site (S5H Fig). The chromatin region amplified by this pair of primers was greatly amplified in SHEP cells transfected with TrkC-KF and Hey1 (S5I Fig). Conversely, BAK promoter, another p53 target [32], was not amplified when this experiment was repeated with primers spanning the p53 binding site on the BAK promoter (S5J and S5K Fig). These results are consistent with the fact that the TrkC proapoptotic pathway does not require BAK but requires BAX and COBRA1. Together, these data support the idea that the nuclear apoptotic pathway triggered by TrkC-KF with Hey1 and p53 is essential to provide the adequate number of TrkC proapoptotic partners in the cytoplasm to finally induce MOMP and apoptosome activation. We previously demonstrated that TrkC-mediated apoptosis constrains tumor growth in NB [12] and proposed that some NB cells escape from TrkC-induced apoptosis by up-regulating NT-3. We thus took advantage of an avian model in which we showed that interference with NT-3/TrkC is associated with NB growth inhibition [12,33]. SHEP cells were inoculated on the highly vascularized chorioallantoic membrane (CAM) of E10 chicken embryos (Fig 6A). Five days later, a primary tumor was formed at the inoculation site. When we inoculated parental SHEP cells or the control clone of SHEP cells, silenced for NT-3 by RNA interference, the size and weight of the tumors were reduced in comparison with tumors generated by scramble siRNA-transfected cells (Fig 6B and 6C). We inoculated 2 independent clones knock-out for Hey1 (CRISPR/CAS9 edited as shown in Fig 3). The weight of the tumor did not vary significantly in Hey1 knock-out clones upon NT-3 invalidation by siRNA, suggesting that Hey1 is necessary for TrkC to limit tumor progression in absence of NT-3 (Fig 6B). We also observed that cosilencing of NT-3 with either Hey1 or p53 by RNA interference in SHEP cells also reverses siRNA NT-3-induced tumor suppressive effect (Fig 6C), as it had previously been demonstrated upon NT-3 and TrkC cosilencing [12]. The reduction in size of tumors formed by NT-3-silenced SHEP cells was associated with high apoptosis, as shown by TUNEL staining performed on tumor cryosections (Fig 6D and 6E). As expected, this induction of apoptosis triggered by NT-3 silencing is reversed when Hey1 or p53 is invalidated (Fig 6D and 6E). These results demonstrate in vivo that TrkC tumor suppressor activity requires Hey1 and p53. To analyze whether silencing of the proapoptotic pathway induced by TrkC/Hey1 in absence of NT-3 could be associated with patient poor prognosis, we analyzed various transcriptomic analyses performed on human NB tumors (S6 Fig). It had previously been shown in a limited number of human samples that TrkC expression is associated with favorable outcome in NB [34] and that Hey1 expression is greatly reduced in NB when compared with benign tumors [35]. We made the same observation on a larger cohort, published by T. Wolf on the National Center for Biotechnology Information (NCBI) GEO, analyzed by Agilent-Microarray 44K (GSE45480,649 samples) for both TrkC and Hey1 (S6A and S6B Fig). TrkC and Hey1 expression is significantly lower in aggressive stage 4 NB tumors than in stage 1–3 NB tumors. We further calculated the intergrade median of expression for NT-3, TrkC, and Hey1. We then selected 3 profiles of tumors based on the mode of action of NT-3 inhibiting the death induced by TrkC and Hey1: (i) tumors which express low levels of NT-3 (beyond the intergrade median of NT-3 expression), high levels of TrkC, and high levels of Hey1—i. e. , tumors in which TrkC is prone to induce apoptosis (NT-3low, TrkChigh, Hey1high); (ii) tumors in which the 3 genes are expressed at a low level or silenced—i. e. , this death pathway is blocked (NT-3low, TrkClow, Hey1low); and (iii) tumors with other profiles. As shown in S6C Fig, the proportion of tumors in which the TrkC death pathway is" ON" (NT-3low, TrkChigh, Hey1high) is high in low-grade NB1–3 tumors but decreases when the grade increases (22% in NB1–3 and 12% in NB4). Conversely, the percentage of tumors in which the death pathway is" OFF" (NT-3low, TrkClow, Hey1low) is limited in low-grade tumors and increases with tumor aggressiveness in NB (S6C Fig). This result is in agreement with our hypothesis that a functional TrkC proapoptotic pathway is associated with a favorable outcome, whereas the silencing of this pathway confers a selective advantage to NB tumors. We observed the same trend when analyzing 3 microarrays performed on other cohorts (L. Shi [36], O. Delattre [37], and R. Versteeg [35]). Finally, a Kaplan-Meier analysis performed on the survival data of the T. Wolf cohort indicated that tumors having a functional TrkC death pathway (Group A: NT-3low, TrkChigh, Hey1high) are significantly associated with a better prognosis than tumors with silenced TrkC proapoptotic pathway (Group B: NT-3low, TrkClow, Hey1low) or other types of tumors (Group C) (S6D Fig). Altogether, these data support the view that TrkC constrains tumor growth via Hey1- and p53-mediated apoptosis in vivo, and this proapoptotic pathway is affected in patients with high-grade tumors. We previously demonstrated that, upon NT-3 deprivation, TrkC-KF is released and shuttled to the mitochondria by its proapoptotic partner, COBRA1. Once at the mitochondria, TrkC-KF and COBRA1 activate BAX and induce the MOMP and the subsequent activation of the apoptosome (S7 Fig and [13]). With this work, we decipher the complex upstream mechanisms involved in TrkC-induced cell death. TrkC-KF is translocated in the nucleus via importins and interacts there with Hey1. Hey1 and TrkC-KF interact and jointly bind to MDM2 promoter E-box, in which TrkC-KF favors Hey1 repressor function on MDM2 transcription. MDM2 transcriptional inhibition promotes p53 stabilization and thus apoptosis. p53 target genes include COBRA1 [29] and BAX [31]. We show here that forced expression of TrkC-KF and Hey1 is associated with enhanced p53 binding to COBRA1- and BAX-respective promoters. Furthermore, we show that COBRA1 expression is enhanced by the activation of the TrkC/Hey1/p53 pathway. When this pathway is altered by Hey1 silencing, the supply of COBRA1 by transient transfection is sufficient to restore TrkC-induced apoptosis. Therefore, the nuclear function of TrkC-KF may not only lead to apoptosis through classic p53 effectors but also through the enhancement of the TrkC mitochondrial pathway by the transcriptional supply of its interactors. Similar fine regulation of proapoptotic protein amounts released in the cytoplasm has indeed already been described for p53 [38]. NB tumors derive from the sympathoadrenal lineage originating from the neural crest cells (NCCs) [39]. NCCs contribute to the formation of the peripheric ganglia, the sympathetic and sensory ganglia, and the medullary region of the adrenal gland. The adequate number of neuronal precursors in the forming ganglia during peripheric nervous development is tightly regulated by peaks of programmed cell death controlled by neurotrophin amounts in the close proximity of the precursors that express at their surface the corresponding neurotrophin receptors (for review [40]). This mechanism of programmed cell death is crucial during gangliogenesis. Indeed, an aberrant number of neuronal precursors in ganglia favors the development of NB, as already observed in mice bearing NB-driving mutations. As a first example, MycN is expressed by migrating NCC [41], and mice having a forced expression of MycN in the sympathetic lineage (TH-MycN) display a hyperplasia of paravertebral ganglia neuroblasts, which are resistant to NGF deprivation–induced apoptosis [42,43]. In parallel, mice bearing a mutation identified in sporadic and familial cases of NB, anaplastic lymphoma kinase F1178L (ALKF1178L), present a higher number of sympathetic neuroblasts per ganglion than wild-type mice [44]. These studies illustrate the crucial need to clearly identify the actors, which define the adequate number of precursors in the peripheric ganglia. It has been well established that NT-3 and TrkC control the adequate number of precursors in the developing sensory ganglia [45]. We have shown in primary sensory neurons and in the chick embryo model that part of the apoptosis occurring upon NT-3 deprivation during neurodevelopment is actively triggered by TrkC itself [9] via COBRA1 [13]. Interestingly, Kessler and collaborators observed that the constitutive up-regulation of Hey1 expression in mutant mice results in a significant loss of TrkC positive sensory neurons. Conversely, Hey1 mutant mice display an increased number of TrkC-positive sensory neurons [46]. These observations are consistent with the fact that TrkC may trigger apoptosis via Hey1 in supernumerary neurons in a setting where the amount of NT-3 is limited. We decipher in this study the mechanisms that may underlie this process. Along this line, Barde and collaborators demonstrated in murine models that TrkA and TrkC constrain the adequate number of peripheric neurons during development by actively triggering apoptosis when deprived of their respective ligands, NGF and NT-3. Conversely, TrkB has no proapoptotic activity in this context [10]. Interestingly, TrkA and TrkC expression have long been associated with regressing NB tumors, whereas TrkB expression is a marker of poor prognosis [2]. We have demonstrated here and in a previous study [12] that TrkC proapoptotic activity controls NB tumor progression. It would be of interest to determine whether TrkA and TrkC proapoptotic activity also controls NB tumor initiation in eliminating supernumerary neuroblasts or neurons in the peripheric ganglia. In their study, Barde and colleagues suggested that p75NTR intracellular domain mediated TrkA and TrkC proapoptotic activity [10]. In our N2A cellular model, p75NTR is not required by TrkC to trigger apoptosis. Moreover, we were able to trigger apoptosis in 3 independent NB cell lines (SHEP, LAN6, and CLB-Ga) displaying various patterns of expression of neurotrophins and their receptors. Further investigations would be needed to investigate the putative interactions between neurotrophin receptors in the control of NB tumor progression. With this work, we have confirmed in vivo that the TrkC/Hey1/p53 proapoptotic pathway indeed limits NB tumor growth. p53 and its inhibitor MDM2 have been particularly studied in NB (for review [47]). However, while mutations in p53 are generally considered to affect half of human adult cancers, pediatric cancers are characterized by the lack of p53 mutations [48–50]. More specifically, in NB, p53 is mutated in less than 1% of the tumors at diagnosis [51]. Tumors with wild-type p53 probably rely on other mechanisms to inactivate p53, and it is thus of interest to note that in pediatric tumors, and more specifically in NB, MDM2 is frequently up-regulated [52]. In the present study, we analyzed the transcriptomic public data sets available and showed that the silencing of TrkC proapoptotic pathway (NT-3low, TrkClow, HEYlow) is also associated with poor patient outcome (S6 Fig). In parallel, TrkC expression has been shown to be epigenetically controlled in various cancers [4,6]. It is thus tempting to investigate whether reactivating TrkC proapoptotic activity in these patients with p53 wild-type tumors may constitute an interesting therapeutic strategy. N2A, SHEP, and HEK293T were described previously [13]. WT HCT116 and p53-KO HCT116 were kindly provided by B. Vogelstein (Ludwig Center at Johns Hopkins, Baltimore, MD) [24]. SHEP, LAN6, and CLB-Ga were kindly provided by V. Combaret (CRCL, Lyon) [12]. N2A cells were grown in DMEM/F-12, GlutaMAX (Life Technologies), supplemented with 10% FBS (Lonza); SHEP, LAN6, and CLB-Ga cells were grown in RPMI1640, GlutaMAX (Life Technologies), supplemented with 10% FBS (Lonza); and HEK293T, WT HCT116, and p53-KO HCT116 cells were grown in DMEM (Life Technologies), supplemented with 10% FBS (Lonza). The plasmid constructs and siRNA were transfected using JetPrime (PolyPlus) for cell death assays and Lipofectamine RNAiMAX (Life Technologies) for RT-QPCR assays following manufacturer’s instructions. Caspase activity was inhibited in SHEP cells by treatment with 20 μM of z-VAD-fmk (Merck-Millipore), a general caspase inhibitor. The pan-importins inhibitor Ivermectin (Sigma I8898) was added to the cells at a final concentration of 10 μM 2 h before cell collection. Pifithrin-α (Sigma P4359), p53 inhibitor was used at a concentration of 20 μM for 30 h. The plasmids encoding full-length TrkC, TrkC-KF, TrkC-642-825, TrkC495-825, TrkC-DM (TrkC-D495N/D641N), Ptc, DCC-IC, and Neogenin IC GFP were described elsewhere [9,16]. The plasmids encoding Hey1, HeyL, and Hey1-RK3 were a kind gift of M. Gessler (University of Wuerzburg, Germany) [26]. TrkC-KF-ΔNLS 1 and 2 were generated by site-directed mutagenesis using QuickChange kit (Stratagene) and the following primers: TrkC-KF-ΔNLS1 forward: 5′-CATATGTTCAACACATCGCCGCCGCCGACATCGTGTTGAAG-3′, reverse: 5′-CTTCAACACGATGTCGGCGGCGGCGATGTGTTGAACATATG-3′. TrkC-KF-ΔNLS1/2 forward: 5′-GAGAGACATCGTGTTGGCCGCCGCCGCCGGTGAGGGAGCCTTT-3′, reverse: 5′-CAAAGGCTCCCTCACCGGCGGCGGCGGCCAACACGATGTCTCT-3′. The plasmid encoding the sgRNA targeting Hey1 (TGACGCGCACGCCCTTGCTA) cloned into the pSPCAS9 BB-2A-GFP (PX458) was generated by GenScript. siRNAs were purchased from Sigma-Aldrich for siRNA control (siRNA Universal Negative Control #2 SIC002_10Nmol), siRNA huHey1 (NM_001040708; SASI_Hs02_00309099), siRNA hup53 (NM_000546; SASI_HS02_00302766), siRNA huCOBRA1 (NM_015456; SASI_hs01_00236976), siRNA muCOBRA1 (SASI_Mm01_00110121), and from Santa-Cruz for siRNA control (sc-37007), siRNA mHey1 (sc-42126), siRNA huNGFR p75 (sc-36057), and siRNA huNT-3 (sc-42125). The 2-hybrid screen was performed by Hybrigenics (Paris, France) using the Mouse Embryo Brain RP2 library as a prey and pB27 (N-LexA-bait-C fusion) and pB66 (N-GAL4-bait-C fusion) vectors. TrkC-KF construct was used as bait. N2A and SHEP cells were transfected with the indicated constructs: After 4 h, the medium was replaced by medium without serum for 24 h to 72 h. Caspase-3 activity was measured as described in [12], using the Ac-DEVD-AFC substrate assay (Biovision, K105-400). Cell death percentages were assessed by trypan blue exclusion, as described in [13]. HEK293T cells were lysed in 50 mM HEPES pH 7. 6,125 mM NaCl, 5 mM EDTA, and 0. 1% NP-40 in the presence of protease inhibitors and were further incubated with an anti-GFP antibody (A11122, Life Technologies) and then with protein G sepharose (Sigma Aldrich) to pull down proteins of interest. Western blots were performed using an anti-Flag antibody (F3165, Sigma Aldrich). HEK293T, SHEP, and N2A cells were lysed in 50 mM HEPES pH 7. 6,125 mM NaCl, 5 mM EDTA, and 0. 1% NP-40 in the presence of proteases inhibitors. For p53 western blots, SHEP cells were lysed in 50 mM Tris–HCl pH 7. 5,100 mM NaCl, 10% glycerol, 0. 1% NP-40, and 0. 2 mM EDTA in the presence of proteases inhibitors. Western blots were quantified using the ImageJ64 software. We used the following antibodies: anti-GFP (A11122, Life technologies), anti-GAPDH (sc-25778, Santa Cruz), anti-Histone (07–449, Millipore), anti-Flag (F3165, Sigma), anti-Actin (MAB1501R, Chemicon), anti-Hey1 (anti-HRT1, sc-16424, Santa Cruz), anti-p53 (SAPU, [53]), anti-Ptc (sc-6149, Santa Cruz), anti-TrkC (AF1404, R&D), anti-MDM2 (VMA00406, BioRad), anti-COBRA1 (F7E4, GeneTex), anti-BAX (sc-526, Tebu Bio), anti-BAK (G-23, Santa Cruz), anti-NT-3 (sc-547, Santa Cruz), and anticleaved PARP (9541T, Cell Signaling). The following antibodies were used for proximity ligation assays (DuoLink): anti-GFP (TP401, Biolabs), anti-Flag (F3165, Sigma), anti-Hey1 (anti-HRT1, sc-16424, Santa Cruz), anti-p53 (sc-126, Santa Cruz), and anti-importins (I1784, Sigma Aldrich). N2A, SHEP, LAN6, and CLB-Ga cells were cultured on coverslips, transfected with indicated plasmids using JetPrime, and then fixed 20 min in 4% paraformaldehyde and permeabilized in PBS/0. 2% Triton. Nuclei were stained using DAPI. Images were obtained by confocal microscopy and analyzed using ImageJ64. For cleaved caspase-3 staining, after permeabilization, SHEP, LAN6, and CLB-Ga cells were incubated in blocking solution (PBS-BSA2%-normal serum 2%) for 1 h before incubation overnight with anticleaved caspase-3 antibody (9661, Cell Signaling) diluted to 1: 1,000 in PBS. After incubation with secondary antibody (Alexa Fluor Donkey anti-Rabbit IgG 555, Invitrogen A31572) diluted to 1: 2,000 in PBS for 1 h, slides were mounted in DAPI-fluoromount G (17984–24, EMS) and imaged using a Zeiss AxioImager microscope. Quantification was performed using ImageJ64. SHEP cells were transfected using JetPrime (PolyPlus). When indicated, cells were treated with Ivermectin (Sigma I8898) 10 μM 2 h before collection. After 4 h, the medium was replaced with medium without serum. 24 h after transfection, cells were harvested, and nuclei were isolated from cytoplasm using the Nuclei Pure Prep Isolation kit (Sigma Aldrich). Input, cytoplasmic, and nuclear fractions were analyzed by western blot, with GAPDH as cytoplasmic marker and Histone H3 as nuclear marker. HEK293T cells were transfected with TrkC-KF-GFP and Hey1 using JetPrime. After 4 h, the medium was replaced with medium without serum. Twenty-four h after transfection, cells were harvested, and cytoplasmic, DNA-bound, and DNA-unbound fractions were separated using the Subcellular Protein Fractionation Kit for Cultured Cells (ThermoFisher Scientific). Fractions were analyzed by WB, using Histone as nuclear marker and actin as loading control. To assay protein interactions in cells by fluorescence, the DuoLink PLA kit was used (Sigma Aldrich). Briefly, cells were cultured on coverslips and then fixed in 4% PFA for 30 min and washed using PBS/7. 5% glycine for 5 min. Cells were then permeabilized in PBS/0. 2% Triton and incubated in a blocking solution for 30 min (PBS/2% BSA). After an overnight incubation with the primary antibodies, cells were incubated with Plus and Minus PLA probes. The probes were ligated and amplified using the Duolink In Situ Detection Reagents Orange (Sigma Aldrich). After several washes with the Duolink In Situ Wash Buffers for Fluorescence (Sigma Aldrich), nuclei were stained using DAPI, and coverslips were mounted in fluoromount. The analysis was made by fluorescence microscopy, and signal quantification was assessed using the ImageJ64 software to count the number of red fluorescence spots compared to total cell number (assessed using DAPI staining). To assay mRNA expression, total RNA was extracted from cells using the Nucleospin RNAII kit (Macherey-Nagel). One microgram of RNA was reverse-transcribed using the iScript cDNA Synthesis Kit (Bio-Rad). RT-QPCR was performed using a Light-Cycler 480 (Roche Applied Science) and the FastStart TaqMan Probe Master Mix (Roche Applied Science). The primers and probes (Universal Probe Library, Roche Applied Science) used are indicated on S1 Table. To assay TrkC-KF transcriptional activity, SHEP cells were transiently transfected with the indicated constructs fused to Gal4 DBD, a plasmid containing the firefly luciferase gene under the UAS-Gal4 control, and a plasmid coding for the Renilla luciferase gene under the CMV promoter as a control. To assess Firefly and Renilla luciferase activities, Dual-Luciferase Reporter Assay System was used following manufacturer’s instructions (Promega). Data represent Firefly value over Renilla value, indexed to control (Gal4). SHEP cells were transiently transfected with the plasmid encoding SpCAS9, Hey1-targeted gRNA, and GFP (Genscript, target sequence: GATAACGCGCAACTTCTGCC) using JetPrime. Two d after transfection, GFP-positive cells were sorted as single cells in 96-well plates for clonal selection. Hey1 mRNA expression level was measured by RT-QPCR for all obtained clones, and 2 clones with significant decrease in Hey1 mRNA level compared to the parental SHEP cell line were selected for further analysis. Editing of the Hey1 gene was confirmed by sequencing for both clones. Control clones were obtained using a plasmid encoding SpCAS9 and GFP (Addgene). GFP-positive sorted clones were analyzed by RT-QPCR to confirm no change in Hey1 mRNA expression level as compared to the parental SHEP cell line, and 2 of them were selected for further analysis. SHEP cells transfected with the indicated plasmids were incubated with 1% formaldehyde for cross-link: Reaction was stopped by the addition of 125 mM glycine. Cells were scraped in swelling buffer (25 mM Hepes pH 7. 9,1. 5 mM MgCl2,10 mM KCl, and 0. 1% NP-40), and nuclei were isolated using dounce homogenizer. After centrifugation, nuclei were resuspended in sonication buffer (50 mM Hepes pH 7. 9,140 mM NaCl, 1 mM EDTA, 1% Triton X-100,0. 1% Nadeoxycholate, and 0. 1% SDS) and sonicated to obtain chromatin fragments of 400 bp to 600 bp size. Chromatin was then incubated with primary antibodies or isotypic IgG (Sigma Aldrich) overnight: anti-Flag (F3165, Sigma), anti-Hey1 (anti-HRT1, sc-16424, Santa Cruz), and anti-p53 (sc-126, Santa Cruz). Complexes were pulled down using protein G sepharose (Sigma Aldrich). After washes, immune complexes were eluted, and cross-linking was reversed at 65°C. Eluates were incubated with RNase A and proteinase K; then, DNA was recovered by phenol-chloroform extraction. DNA fragments were analyzed by RT-QPCR using a Light-Cycler 480 (Roche Applied Science) and the FastStart TaqMan Probe Master Mix (Roche Applied Science). The primers and probes (Universal Probe Library, Roche Applied Science) used are indicated on S2 Table. To assay Hey1 and TrkC-KF ability to bind MDM2 promoter, SHEP cells were transiently transfected with Hey1-Flag and TrkC-KF-GFP-expressing plasmids: Cells were resuspended in hypotonic buffer (10 mM HEPES, pH 7. 9,1. 5 mM MgCl2,10 mM KCl, 0. 5 mM dithiothreitol with protease inhibitors [Roche]) and incubated on ice for 10 min. Nuclei were isolated by centrifugation, resuspended in RIPA buffer supplemented with complete proteases inhibitors, and incubated on a rotating wheel at 4°C for 20 min to obtain a nuclear proteic extract. Meanwhile, biotinylated oligonucleotides corresponding to Hey1 and TrkC-KF binding sites on the MDM2 promoter, and containing either a WT E-box (CACGTG) or a mutated E-box (CCCGGG), were annealed to form double-stranded oligonucleotides of 80 bp size. For WT E-box, the forward oligonucleotide is 5′-biotinylated with the following 5′–3′ sequence: gggggctcggggcgcggggcgcggggcatggggcacgtggctttgcggaggttttgttggactggggctaggcagtcgcc. WT E-box reverse oligonucleotide: ggcgactgcctagccccagtccaacaaaacctccgcaaagccacgtgccccatgccccgcgccccgcgccccgagccccc. For mut E-box, the forward oligonucleotide is 5′-biotinylated with the following 5′–3′ sequence: gggggctcggggcgcggggcgcggggcatggggcccggggctttgcggaggttttgttggactggggctaggcagtcgcc. Mut E-box reverse oligonucleotide: ggcgactgcctagccccagtccaacaaaacctccgcaaagccccgggccccatgccccgcgccccgcgccccgagccccc. Three micrograms of double-stranded biotinylated oligonucleotides were incubated with 300 μg of nuclear protein extract for 2 h at 4°C. Complexes were pulled down using 50 μL of streptavidin-agarose beads (Sigma Aldrich) incubated for 1 h at 4°C. The protein-DNA-streptavidin-agarose complex was washed 3 times with RIPA buffer and loaded onto an SDS gel. Detection of Hey1-Flag and TrkC-KF-GFP proteins was performed by western blot as described in [13]. SHEP cells were transfected with jet prime and siRNA NT-3 (100 nM) and/or siRNA Hey1 or sip53 (50 nM) 24 h before inoculation. Five million cells were suspended in 25 μl complete medium and 25 μl matrigel (Corning 356231) and seeded on 10-d-old (E10) chick CAM. On day 15, tumors were resected and weighted. To monitor apoptosis on primary tumors, they were fixed on 4% PFA, cryoprotected by overnight treatment with 30% sucrose, and embedded in Cryomount (Histolab). TUNEL staining was performed on tumor cryostat sections (Roche Diagnostics), and nuclei were stained with DAPI. Three to eight sections were analyzed at ×20 magnification for at least 3 tumors for each condition. TUNEL and DAPI positive cells were counted by ImageJ64 software. The expression values analyzed here are publically available in GEO database (http: //www. ncbi. nlm. nih. gov/geo/). T. Wolf cohort (GSE45480[54]) analysis was performed with Agilent-020382 Human Custom Microarray 44k (GPL16876); the following data set probes were used: NT-3 (NTF3) UKv4_A_23_P360797, TrkC (NTRK3) UKv4_A_23_P205900, UKv4_A_23_P88538, and Hey1 UKv4_A_32_P83845. In this study, for TrkC and Hey1 expression values, a mean of the values obtained with the various probes was calculated. Kaplan-Meier analysis was performed in R2: Genomics Analysis and Visualization Platform (http: //r2. amc. nl). The p-value is calculated to determine the optimal cutoff and is finally corrected by Bonferoni as described in [55]. A new grouping variable was made on the basis of NT-3, TrkC, and Hey1 as described in the main text. Number of experiments and statistical tests used is indicated in figure legends. Statistical treatment of the data was performed with Prism 6. 0e (GraphPad) and BiostaTGV online statistical software (http: //marne. u707. jussieu. fr/biostatgv/).

Tropomyosin receptor kinase C (TrkC) is a transmembrane receptor at the cell surface and has been described to work paradoxically both as an oncogene and as a tumor suppressor. We partly solved this paradox in a previous study, demonstrating that TrkC is a double-facet receptor: Upon interaction with its ligand neurotrophin-3 (NT-3), TrkC has a tyrosine kinase activity and induces survival and proliferation of the cell; conversely, in the absence of the ligand, TrkC is cleaved and releases a" killer-fragment" that triggers apoptosis. In this study, we analyze the fate of this fragment and show that TrkC killer-fragment is translocated to the nucleus, where it stabilizes the apoptosis inducer p53. We further find that p53 activates the transcription of cytoplasmic molecular partners, which interact with TrkC killer-fragment and induce apoptosis. We also demonstrate that alteration of this mechanism favors tumor growth in neuroblastoma (NB), an avian tumor progression model for a pediatric cancer.

Abstract Introduction Results Discussion Materials and methods

cell death nuclear staining gene regulation cell processes green fluorescent protein cloning luminescent proteins immunoprecipitation mitochondria molecular biology techniques bioenergetics cellular structures and organelles tropomyosin research and analysis methods small interfering rnas specimen preparation and treatment staining proteins gene expression molecular biology precipitation techniques biochemistry rna dapi staining cytoskeletal proteins cell biology nucleic acids apoptosis genetics biology and life sciences energy-producing organelles non-coding rna

2018

Hey1- and p53-dependent TrkC proapoptotic activity controls neuroblastoma growth

Protein translation is the most expensive operation in dividing cells from bacteria to humans. Therefore, managing the speed and allocation of resources is subject to tight control. From bacteria to humans, clusters of relatively rare tRNA codons at the N′-terminal of mRNAs have been implicated in attenuating the process of ribosome allocation, and consequently the translation rate in a broad range of organisms. The current interpretation of “slow” tRNA codons does not distinguish between protein translations mediated by free- or endoplasmic reticulum (ER) -bound ribosomes. We demonstrate that proteins translated by free- or ER-bound ribosomes exhibit different overall properties in terms of their translation efficiency and speed in yeast, fly, plant, worm, bovine and human. We note that only secreted or membranous proteins with a Signal peptide (SP) are specified by segments of “slow” tRNA at the N′-terminal, followed by abundant codons that are considered “fast. ” Such profiles apply to 3100 proteins of the human proteome that are composed of secreted and signal peptide (SP) -assisted membranous proteins. Remarkably, the bulks of the proteins (12,000), or membranous proteins lacking SP (3400), do not have such a pattern. Alternation of “fast” and “slow” codons was found also in proteins that translocate to mitochondria through transit peptides (TP). The differential clusters of tRNA adapted codons is not restricted to the N′-terminal of transcripts. Specifically, Glycosylphosphatidylinositol (GPI) -anchored proteins are unified by clusters of low adapted tRNAs codons at the C′-termini. Furthermore, selection of amino acids types and specific codons was shown as the driving force which establishes the translation demands for the secretory proteome. We postulate that “hard-coded” signals within the secretory proteome assist the steps of protein maturation and folding. Specifically, “speed control” signals for delaying the translation of a nascent protein fulfill the co- and post-translational stages such as membrane translocation, proteins processing and folding. In dividing cells, the process of translation elongation consumes most of the cell energy and resources [1]–[3]. The rate of translation must be tightly controlled for coping with the cell demands and its limited resources. Specifically, translation efficiency is determined by the amount of proteins that are produced from the coding mRNA. In a more mechanistic view, translation efficiency is reflected by the preferable allocation of ribosomes on the mRNA [4]. Sequence-based features such as mRNA folding energy, positioning of individual amino acids (AAs) and codons govern the translation efficiency [5]–[7]. Failure in coordinating the ribosomal flow leads to ribosomal drop-off [3], translation errors [8], frame-shift [9] and protein misfolding [10]. Direct measurements of ribosome density from in vivo studies confirmed that translational rates differ between transcripts [11]. Moreover, the rate may vary by several folds on the same mRNA [2], [12], [13]. Several factors govern protein translation rate and accuracy (see discussion in [3], [14], [15]). A dominant parameter in dictating translation rate is the nature of the codons at the initial segment of the transcripts [16]. Other features include the competition on ribosome binding [17], mRNA folding energy [5], accessibility of specific tRNAs [18] and CG content [5]. A dominating parameter of translation efficiency from E. coli to human is the codon usage [19], [20]. The coding usage of a broad range of organisms positively correlated with cellular proteins' expression levels and thus, indirectly, with translation efficiency [21], [22]. In all eukaryotes, the decoding of mRNAs to proteins obeys the same rules [23]. The genomic tRNA copy number (CN) strongly correlates with the needs for intracellular tRNA levels [24]. This property is best captured by the tRNA adaptation index (tAI) [19] that balances between the decoding rules and the tRNA CN [25]. Indeed, in humans, tAI appropriates the actual abundance of tAI in healthy and diseased cells [26]. In eukaryotes, a distinction should be made between proteins that are translated by the soluble, cytosolic ribosome (CYTO-Rb) and the membrane-bound ribosomes (MEM-Rb). The latter cover the proteins destined to the secretory systems (endoplasmic reticulum (ER), Golgi, endosomes, lysosomes, plasma membrane and the extracellular space) [27]. A common feature of the secretory proteins is the presence of signal peptide (SP) at the N′-terminal [28]. Alternatively, membranous proteins that lack SP (e. g. , many G-protein coupled receptors) use their first TMD as a membrane signal. Translation of the secretory proteins at the ER membranes is a multiphase process that is based on coordinated steps of translation, translocation and folding [13], [29], [30]. In this study, we hypothesized that proteins of CYTO-Rb and MEM-Rb translation differ in their translation elongation management. A local tRNA adaptation pattern at the N′-terminal which starts with segments of lowly adapted tRNAs, followed by segments of highly adapted tRNAs, is characteristic of secreted and membranous SP-proteins but not identified in the bulk of the proteins or in other regions of the transcripts. Such patterns are shared by a large number of eukaryotic proteomes and found also in proteins that are designated to the mitochondria. The impact of “traffic signs” on the management of translating ER-bound ribosomes is discussed in view of recent experimental evidence on translation rates. An estimation of the effect of the tRNA abundance on the efficiency of the translation is captured by the tRNA adaptation index (tAI) (See Materials and Methods). The pairing of tRNA with the mRNAs is not unique in the case of the Wobble pairing (Figure 1A). Each organism differs by the number and the relative appearance of tRNA isoacceptors for decoding the 20 amino acids (AAs, 61 codons). Synonymous codons are associated with a broad range of tAI values (Figure 1B). Some AAs (e. g. , Arginine) are encoded by 6 codons but the range of their tAI values is still very narrow. On the other hand, a broad range of tAI values is associated with AAs that have only two codons each (e. g. , Asparagine and Cysteine) (Figure 1B). The tRNAs copy number (CN) is subjected to evolutionary forces and thus differs substantially throughout the evolutionary tree. For example, there are 287 tRNA genes in the budding yeast S. cerevisiae but as many as 3790 tRNA genes in Bos Taurus. The tAI value that is assigned to each codon varies substantially among different organisms. While the correlations among human, D. melanogaster, C. elegans are moderate, the correlations with B. taurus or A. thaliana (flowering plant) are negligible (Figure 1C). The tAI codon values for each organism is listed in Table S1. The translation of proteins in eukaryotes is executed in two settings: Proteins that are translated by free ribosomes (coined cytoplasmic ribosome, CYTO-Rb) and ER bound ribosomes (coined membranous ribosome, MEM-Rb). We partitioned the entire proteomes into four non-overlapping groups (Table S2): Groups (i–iii) compose the secretory proteome (Figure 2A). The human proteome consists of 18,434 proteins. Among them 26% include at least one TMD and an additional 9. 5% are secreted proteins that contain SP. A similar partition is reported for fly, worm and bovine (Figure 2B) and other model organisms. The tAI of each coding sequence is computed (see Materials and Methods), and the average “global tAI” for the analyzed proteins' group was defined (see Materials and Methods). Each of the three protein groups that together compose the secretory proteome displays a distinct global tAI (Figure 2C). For example, the p-value of the human secreted proteins (marked as “SP-not TMD” group) relative to membranous proteins without SP (TMD not SP) is 2. 58e-11. The calculated p-values of the secreted proteins with respect to membranous proteins with SP (TMD and SP) and the cytosolic group are 1. 08e-14 and 9. 01e-12, respectively. Comparing the average global tAI values for the secretory and cytosolic protein groups in different organisms is shown in Table 1. The main observation (Figure 2C) demonstrates that secreted proteins that have SP tend to have higher global tAI relative to the proteins of the membranous groups (TMD, with or without SP). While the absolute values of the global tAI are different for each organism (based on codon tAI, Table S1), the trend of low tAI for the membranous proteins relative to the secreted proteins is surprisingly robust (Figure 2C). We extended the analysis to include also yeast and plant representatives. The average values of the calculated global tAI values for (i) cytosolic proteins, (ii) SP-no TMD (iii) SP and TMD and (iv) TMD not SP are listed in Table S3. We show the statistical significance among each pair of the protein groups for 6 organisms (Table 1). The statistical difference between the two exclusive sets of membranous proteins (with/without SP) is minimal (with p-value>1. 0e-4, Table 1). For example, the p-values of the global tAI values for the yeast-secreted proteins relative to other groups range from 1. 67e-12 to 6. 48e-23 (Table 1). A striking observation is that secreted proteins and the soluble fraction (i. e. , CYTO-Rb translation) specify high average global tAI values with regard to the membranous proteins. A similar trend was observed in all six tested organisms (included yeast and flowering plant, Table S3). Many determinants govern the protein abundance in eukaryotic cells [11]. The contribution of sequence-dependent determinants to the rates of translation and degradation has been estimated [31]. A positive correlation between the gene tAI and its expression was determined from the signature of gene expression microarrays [32]. We tested whether the average higher global tAI that was associated with the secreted (SP non-TMD) and the cytosolic proteins (Table S3) relative to membranous proteins reflects a difference in the expression levels. We took advantage of the experiments with high coverage of the yeast proteome and compared the protein abundance and the global tAI. We used a resource from mass spectrometry (MS) peptide counts [33] (total of 4012 proteins, Figure 3A) and the quantitative data from GFP-tagged proteins [34] (total of 2279 proteins, Figure 3B). We found substantial agreement between the results from these complementary technologies (compare 3A and 3B). The strongest correlation was noted between the global tAI values and the cytosolic proteins. However, the significance of the correlation between the global tAI and the proteins of the secreted proteome is rather weak (SP not TMD). We suggest that the relatively high global tAI is associated with an overall expression level for the majority of the proteins that are translated by free ribosomes (i. e. , accounts for 78% and 81% of the analyzed proteins, Figure 3A and 3B, respectively). However, a high expression level is not supported for the secreted protein group. Additional parameters such as protein length, AA usage and CG content were also tested. The length of the proteins from the group “SP and TMD” was significantly longer than the rest of the proteins (P value = 1e-4). But the secreted proteins group (SP not TMD) and the “TMD not SP” group that differs in their tAI (Figure 2C) have no difference in protein length (p value = 0. 133). All other correlations show a borderline statistical significance. We concluded that the tAI is strongly associated with protein abundance only for the cytosolic proteins. The same trend was found for the human proteome (data analyzed from [35]). The secreted proteins showed significantly higher global tAI values (Figure 2C, Table S3). We tested the possibility that the tested protein groups may carry segmental information in addition to their global tAI values. To analyze the segmental properties of the proteomes, we discretized the transcripts to segments of 30 codons. The same notations were applied for the C′-terminus, starting from the last codon of the protein (Figure 4A). The results are presented as “Relative tAI, ” which is defined as the current segments' tAI divided by the calculated value of the global tAI of the coding sequence. This measure allows comparing the trends among organisms. Using the Relative tAI values (and not the absolute tAI values) cancels out the inherent difference in expression levels that are associated with the tested proteins groups (Figures 2–3). Among the analyzed model organisms, the annotations for the human proteome are accurate and complete. According to the four groups partition (Figure 2B and the cytosolic fraction), the SP-containing proteins are characterized by an occurrence of lowly adapted tRNAs segment (coined LATS) at the N′-terminal (∼45 codons) followed by highly adapted tRNAs (HATS) (Figure 4B). Notably, proteins that contain SP with or without TMD display a similar profile. All protein groups converged at segment N3 (codons position 60–90, Figure 4B). It is important to note that the “Relative tAI” profile of the entire proteome (combined all 4 groups, marked “All”, Figure 3B) shows no outstanding position-based pattern. Additional segments (e. g. , N4) provided no additional information and will not be discussed further. Figure 4C shows the cumulative distribution of tAI values for each of the analyzed protein groups for N1 and C1 segments from a human proteome. The statistical difference between the N1 and C1 segments is significant (Table 2). Actually, both the N1 and the C1 segments differ significantly from a random selection of a 30-codon segment (Kolmogorov-Smirnov (KS) test, Figure 3C, Table 2). The calculated p-values versus the random sets range between 1. 0e-15 to 1. 0e-22 for N1, and 1. 0e-12 to 1. 0e-27 for C1. More importantly, the statistical tests show significant p-values (7. 6e-6 to 2. 1e-57) for the characteristics of the N1 segment among the four protein groups, while the p-values for the C1 segments are statistically insignificant (Table 2). The tAI segmental analysis was extended to other model organisms including B. taurus (Figure 4D), D. melanogaster (Figure 4E), C. elegans (Figure 4F) and S. cerevisiae (Figure 4G). Assessing the significance of the differences in the “Relative tAI” values for the different segments of the four protein groups is achieved by comparing the maximal range of the computed average relative tAI among the four groups. For example, the “Average Relative tAI' ”of the N1 in H. sapiens spans as much as 0. 053 while the C1 deviates by only 0. 007. We demonstrated these range differences of N1, N3 and C1 segments for all the tested organisms (Figure 4H). A similar pattern is generalized and the range of “Average Relative tAI” of N1 is significantly higher than that of N3 or C1. In this view, the range in values of segment N3 is considered a statistical noise. As many of the secreted proteins (e. g. , hormones, growth factors) are short proteins, we tested the effect of protein length on the observed segmental tAI profile. We confirmed that the impact of the protein length of the segmental local tAI is negligible. Specifically, we partitioned the SP-proteins to very short (90–240 AAs) and very long (>1,000 AAs) protein groups. We found that the trend of the tAI profiles is insensitive to the length. The “very short” and “very long” proteins originated from the same distribution (t-test, p-value = 0. 72). We tested the differential tAI segmental profiles of membranous proteins (composed of the groups of “SP and TMD” and “TMD not SP”) according to the separation to single (marked as types I–IV) and multi-pass proteins (Figure 5A). This type of partition tests whether the membrane topology governs the characteristics of the tAI segmental profile (shown in Figures 4B–4G). It is evident that the existing of SP dominates the profile irrespectively to the number of TMDs or the protein topology within the membrane (Figure 5B). The analysis is limited to yeast and humans due to the poor annotations on membranous protein topologies for the other model organisms. Alignment of the proteins at their N′- and C′-terminal segments was essential to reveal the signal for the SP-proteins, irrespective of the membrane topology of a specific protein (Figure 5). For membranous proteins that lack SP, the first TMD acts as the anchor signal. We further tested whether a codon dependent signal is encoded in the TMD. To this end, we aligned all sequences from the “TMD not SP” group by their first TMD (Figure S1). We found that the segmental tAI values of the first TMD differs from the observation of the SP-proteins. Actually, the “anchored TMD” shares no local tAI characteristics. We concluded that it is not the hydrophobicity per se that dictates the local tAI properties but instead, the SP sequences are characterized by clusters of lower adapted codons followed by clusters of highly adapted segments. The robust phenomena of differential codon usage according to their tAI property along the transcript is not restricted to the N′-terminal segment. The Glycosylphosphatidyl inositol (GPI) anchored proteins reach the ER through an SP dependent process. For these proteins, an additional modification occurs following a proteolytic cleavage at a C′-terminal peptide of the nascent peptide [36]. We tested whether a signal for GPI lipid anchoring is encoded by segmental tAI measurements. We separated the proteins that are predicted as GPI-anchor proteins [37]. Figure 6A shows a histogram for the cleavage site with respect to the last codon (marked as codon 0). In the majority of the cases, the cleavage sites are positioned within the C1 segment (codon marked as -25). The average segmental tAI profile for the 128 human GPI-proteins is shown (Figure 6B). Remarkably, the AAs composition of the GPI-anchor proteins is poorly conserved. Still, the GPI-anchor proteins are characterized by the significance of LATS at their final segment (C1, ∼30 codons, Figure 6B). Thus, GPI-anchor proteins are marked by evolutionary signals at both, the N′- and C′-termini. As opposed to the previously mentioned cases of GPI-anchored and SP-proteins that are modified at the ER on the nascent chain, translocation of mitochondrial proteins occurs as a post-translational stage. Hundreds of proteins reach the different compartments of the mitochondria (and chloroplasts in plants) by sophisticated mechanisms [38], [39]. Many of these mitochondrial targeted proteins have a cleavable Transit Peptide (TP) in their N′-terminals. There are 499 proteins annotated to have TP in humans. Figure 6C shows the cleavage sites with respect to the initiator Methionine. For the majority of the proteins, the cleavage sites are positioned within the N1 or the N1-intermediate segments. The similarity of the local segmental tAI to the profile of the SP-proteins is evident (Figure 6D). TP adopts a more extreme value (“Relative tAI” of 0. 95 in H. sapiens) for an extended segment relative to the SP-proteins (Figure 6D). An overlap in the segmental profiles for the SP and TP protein is striking. Figure 6E demonstrates that when the AA compositions of the SP and the TP are compared, the overlap in the AAs usage is minimal. These results postulate as to the generality of the phenomenon. Notably, the marked difference in codon usage of the SP and TP segments argues for an unrestricted selection that supports a pattern of LATS followed by HATS. Such a design may be used as a general trend for management of protein targeting to sub-cellular compartments and organelles. A key sequence feature of the SP is the central helical region that is dominated by Leu and Ala with some occurrence of Val, Phe and Ile. We show that the SP proteins have a preferable use of some amino acids (e. g. , Leu and Trp), but a limited use of Asn, Asp, Ser, Thr and Arg. There are two possible explanations for the observed profile at the N1-segment of the proteins with SP sequences: (i) The AAs that determine the SP are enriched with “slower” codons (i. e. , lower tAI codon values); (ii) The codons at the initial segment that compose the SP reflect an evolutionary selection process. Both explanations may fulfill the global demands of MEB-Rb translation mode. In order to distinguish between these possibilities, we counted the codon usage in the SP of each of the relevant proteins, and the codon usage in segments of non-SP proteins. For some codons, the deviation between the usage in SP and non-SP is substantial (Figure 7A). For example, the use of Cys is preferable in SP-proteins, while Lys is rarely used in the segment that covers the SP sequence. Additionally, we tested the existence of an evolutionary signal that can account for the preferential selecting of codons in the N′-terminal segments of the SP-proteome. This is performed for any AA, regardless of its actual tendency to be used. Specifically, we questioned whether a selected codon in the SP sequence is randomly chosen from a background of the complete proteome codon usage data. We show the preferred usability of a specific codon in view of its tAI value (Figure 7A, empty frames). For example, the AA valine (Val) is encoded by four codons. Among these codons, the codons that are mostly used for the SP-proteins are the ones with low tAI values (codons GTC) while the ones with maximal tAI value (codon GTG) are rarely used (Figure 7A). In order to assess the statistical power of such observations, we compared the actual local tAI for the SP segment (as in Figure 7A) with that of simulated sequences that are composed of identical amino acids but are encoded by codons that were randomly selected from their synonymous codons, according to the tAI distribution in the entire genome (Figure 7B). While the tAI distributions are quite similar (dKL<0. 001), the mean value of the actual SP local tAI value was lower with respect to the randomized sequences (0. 3143 and 0. 3209 for the original SP and the synonymous codons tAI 1000 randomized tests, respectively). Importantly, the distributions differ significantly from the replaced sequences according to the codon usage distribution (p-value = 1. 3e-07). We concluded that in addition to the preselected AAs for the SP sequences (Figure 7A), an evolutionary signal is attributed to the selection of preferred codons in the SP sequences (Figure 7B). The N′-terminal segmental profile of SP proteins dominated over 3,100 protein sequences in humans (Figures 4B–4G). To ensure an unbiased analysis of the human proteome, we clustered by means of an unsupervised mode all ∼18,400 human proteomes according to their segmental tAI profile (illustrated in Figure 4A). We focused on clusters that are dominated by LATS at the N1 segment (Figure 8, clusters 1–4). Enrichment tests according to the clusters' annotations were performed. The most significantly enriched cluster' s annotation consists of secreted, signal, glycoprotein and disulfide-bridge (p-value of enrichment is 5. 4e-18). An additional set of enriched annotations includes the plasma membrane and membranous proteins. These annotations are fully consistent with MEM–Rb translation (for a detailed analysis, see Table S4). Therefore, the clusters of most significant LATS values followed by HATS are associated with secreted proteins, membranous proteins, extracellular matrix and receptors, all of which belong to SP-containing proteins. Based on a global, unbiased clustering, proteins that are signified by a characteristic pattern are identified. For example, a profile with several consecutive HATS (Figure 8 cluster 6,170 proteins) matches ribosomal proteins. Such a profile is expected for proteins that are expressed at high amounts and a translation speed that reaches maximal efficiency (i. e. , the number of proteins that are produced per transcript). Ribosomal proteins are known by their high expression, efficient translation and the preferable use of abundant codons. A detailed analysis of proteins clusters according to the segmental tAI profile (Figure 8) is beyond the scope of this study. Several models were developed to capture the translation kinetics of the secretory proteome [60]–[62]. Based on this view, the signal that was exposed in this report could also serve to enhance the capacity of the mRNA to engage in a productive ER targeting process. An efficient reuse of the mRNA on MEM-Rb, once the mRNA is “occupied” by an already docked ribosome, is an attractive proposal [13], [63]. Our analysis focused on the MEM-Rb translation. We revisited the mechanistic demands of the secretory proteome [30]. In addition to the need of managing the ribosomal flow for any transcript, special constrains are imposed for the MEM-Rb translation. In mammals, the co-translocation of SP-containing proteins is mediated mostly by the signal recognition particle (SRP) [64]. Once the SRP recognizes the emerging SP from the ribosome [65], a conformational change leads to slowing of translation. Apparently, this attenuation in translation rate is necessary for the nascent chain to diffuse to the ER membrane [47]. The interaction of the SRP with its receptor (SR) and its release serve as an internal “timer” for resuming translation [66], and for production of functional proteins [67]. Recently, the SRP-independent insertion route was systematically assessed in yeast [68] and mammals [69]. The dependency of the hydrophobicity index of the N′-terminal segments of the proteins and the tendency to bind the SRP revealed that a substantial fraction of the yeast secretome is actually SRP-independent and this fraction mainly applies to SP-proteins and to the subset of the GPI proteome [68]. Thus, the notion of a “timer” for translation and translocation may not be limited to SRPs but to the need for a rich network of proteins and chaperones that coordinate their actions to ensure appropriate translocation and targeting. A role for the codons' distribution along the transcripts as a “time delayer” should be considered. With this notion, the generality for transcripts for SP-, TP- and GPI-anchor proteins is striking. We suggest that attenuation of events such as the SP proteolytic cleavage (not necessarily in the end of the LATS), the speed of folding, the cleavage of GPI to promote the locking of the protein at the membrane surface, and recycling of the mRNA to ensure additional rounds of translation are all encoded in the codon organization profile. A similar signature across a range of organisms from yeast to humans indicates a robust, evolutionary refined phenomenon. The list of proteins for each group of each organism was taken from UniProtKB based on a “reviewed” set. For SP proteins we used the UniProtKB (Based on SignalP4. 0 [52]). Only proteins marked with “signal” and “cleaved site” were considered. The SP-anchored proteins were excluded from the SP-proteins group. In addition, the proteins marked as “fragment” were excluded. A similar protocol was applied for GPI-anchored and TP (transit-peptide) and predicted Tail-anchored (TA) Type IV. The canonical variants from UniProtKB were mapped to their matched RefSeq nucleotide sequences. A gene that had no matched sequence, or had a sequence that lacked the ATG initiator codon, was discarded. The corresponding coding sequences were extracted from the RefSeq database. Only proteins that start with an initiator Methionine and end with a stop codon are compiled. Signal peptide sequences were retrieved from the proteins coding sequences according to their position that were marked by UniProtKB. The codon usage for these sequences was counted and defined as SP codon usage. The codon usage of sequence from proteins that are not annotated as SP proteins was counted as non-SP codon usage. Those sequences began at the first position of the coding sequence and terminated at a position that was randomly selected from the signal sequence length distribution. Sequences that were randomly replaced were created by replacing each codon in the sequence with a codon from its synonymous codons by a random choice according to the codon usage of each AA. Randomized tests were performed 1000 times. A high coverage (>70%, 4,500 proteins) mass spectrometry (MS) yeast experiment [33] was used for protein abundance measurements. Protein levels span more than four orders of magnitude. Independent yeast protein quantitation was extracted from the GFP library measurements [34]. Briefly, each protein from the GFP-tagged yeast library was counted by flow cytometry measurement (∼2,500 proteins). For human protein abundance, the MS data resource for the high-coverage of 11 human cell-lines [35] was used. An estimation of the effect of the tRNA abundance on the efficiency of the translation rate of codons is captured by the tRNA adaptation index (tAI) [19]. The tAI value for each codon is composed from two components – the amounts of the relevant tRNA and its codon–anticodon coupling. The latter is not unique - a factorization for each of the wobble pair was used [19]. Global tAI measurement gauges the availability of tRNAs for each codon along the mRNA. Data of genomic tRNA copy numbers were taken from the Genomic tRNA Database (http: //gtrnadb. ucsc. edu/) using human genome hg19 (NCBI Build 37. 1, Feb 2009) [70]. For each tRNA isoacceptor, the number of gene copies (excluding Pseudogenes and Selanocysteine tRNAs) was counted. The codon tAI and global tAI for the model organisms was calculated as above from Genomic tRNA Database (Table S1). A codon–anticodon coupling is not unique - a factorization for each of the wobble pair was used [19]. Formally, let ni be the number of tRNA isoacceptors recognizing codon i. Let tCGNij be the copy number of the jth tRNA that recognizes the ith codon, and let Sij be the selective constraint on the efficiency of the codon-anticodon coupling. We have used the Sij scaling for the Wobble nucleoside-nucleoside pairing as described in [41]. We define the absolute adaptiveness, Wi, for each codon i as: From Wi we obtain wi, which is the relative adaptiveness value of codon i, by normalizing the Wi' s values (dividing them by the maximal of all the 61 Wi). The final tAI of a gene (referred as Global tAI) is the geometric mean of its codons (excluding the stop codon). A geometric mean was calculated in an identical way for calculating the segmental tAI (e. g. , 30-codons, SP-segment, TMD segment). Local tAI is calculated by dividing each coding sequence into several overlapping windows, each containing 30 codons. Relative tAI value is defined as the ratio of the segmental, local tAI (i. e. , 30-codons segment) to the calculated global tAI of the protein (for the entire protein length). A relative tAI value <1. 0 signifies the preference of rarely adapted tRNA codons (“slow” codons) in the analyzed segment relative to the codon composition of the entire coding sequence. Global tAI and C1 segment tAI were computed by excluding the stop codon from their sequences. For sequences that are shorter than 180 amino acids, only local segmental tAI were calculated. This was applied to avoid overlap between N′ and C′ terminal windows. Protein clustering was performed for a matrix of 18,434 rows (each represents a mRNA-mapped coding sequence), and five columns (each represents a window of 30 codons from the N′-terminus segments marked N1 to N3. The functional annotation enrichment of the resulted clusters was according to Fisher Exact Test enrichment scheme with hypergeometric distribution and multiple hypothesis corrections [71]. Different data distributions were compared using the standard Matlab statistical tools such as Kolmogorov–Smirnov (KS) and t-tests. The KS test compared any two samples while quantifying the empirical cumulative distribution functions of the two. The p-value is calculated under the null hypothesis that the samples are drawn from the same distribution. Thus, the lower p values indicate more significant differences between the two examined samples. The difference in the probability distribution between the two datasets was computed using Kullback–Leibler divergence (dKL) (see detailed in [72]). For testing the similarity of the segmental tAI profile to randomly created genes, we created random gene sets with the same codon preference and same length distribution. We selected a set of 1000 genes. The simulation was performed by 1000 repetitions of the protocol.

Measurements of translation by ribosomal profiling and additional large-scale methods support the notion that the elongation speed and ribosomal occupancy are tightly regulated. We revisited the proteomes of a number of organisms, from yeast to human, and focused on the appearance of codons' clusters that impact the speed of translation elongation. Thus, transcripts are analyzed according to their encoded “traffic signs. ” Specifically, translation by free- or endoplasmic reticulum (ER) -bound ribosomes differs substantially with respect to the codon clusters' distribution at the beginning of the coding region. Discretization of all transcripts to consecutive segments exposed the uniqueness of secreted and membranous proteins that have a signal peptide (SP). Similarly, a non-random codon distribution characterized proteins with “targeting peptides” for mitochondria and for GPI-anchor, while the bulk of the proteome carry no significant pattern of their codons. We conclude that translation via an ER co-translocation process imposes unique constraints on translation efficiency that match with the fate of the proteins as secreted, membranous, mitochondrial-targeted or GPI-anchored. Tuning the translation of a nascent protein is essential for coping with the constraints imposed by membrane-bound translation for a successful ER translocation and protein processing for maturation and folding.

Abstract Introduction Results Discussion Materials and Methods

2014

Speed Controls in Translating Secretory Proteins in Eukaryotes - an Evolutionary Perspective

Species abundance distributions (SAD) are probably ecology’s most well-known empirical pattern, and over the last decades many models have been proposed to explain their shape. There is no consensus over which model is correct, because the degree to which different processes can be discerned from SAD patterns has not yet been rigorously quantified. We present a power calculation to quantify our ability to detect deviations from neutrality using species abundance data. We study non-neutral stochastic community models, and show that the presence of non-neutral processes is detectable if sample size is large enough and/or the amplitude of the effect is strong enough. Our framework can be used for any candidate community model that can be simulated on a computer, and determines both the sampling effort required to distinguish between alternative processes, and a range for the strength of non-neutral processes in communities whose patterns are statistically consistent with neutral theory. We find that even data sets of the scale of the 50 Ha forest plot on Barro Colorado Island, Panama, are unlikely to be large enough to detect deviations from neutrality caused by competitive interactions alone, though the presence of multiple non-neutral processes with contrasting effects on abundance distributions may be detectable. The extent to which ecological processes can be inferred from macroecological patterns has long been debated [1–4]. The appearance of common patterns in species abundance distributions (SADs) for different communities suggests that the same ecological mechanisms structure these communities [5,6]. However, it is now thought that many patterns describing communities are rather insensitive to these processes [7–11]. For example, empirical SADs are in many cases found to be statistically consistent with Hubbell’s neutral theory [12–17], but this does not mean that communities are truly neutral because non-neutral models can predict similar [8–10,18,19] or even identical [4] patterns. This raises the question of whether anything can be inferred from fitting it to SAD data [3,4, 20]. Neutral theory has been criticized for the biological processes it omits, but non-neutral models that give qualitatively or even exactly the same predictions can be equally artificial and unrealistic [4,18,21]. Neutral theory has many virtues [22–24] and in many ways it is more complete in scope than competing niche theories [25]. It describes community dynamics at the individual level, treating births, deaths, and dispersal as stochastic processes. It is susceptible to rigorous statistical tests, because unlike many other demographic models the likelihood of obtaining a particular community or sample can be computed exactly [26–30]. Even the controversial neutral assumption that interactions between individuals do not depend on species identity is inspired by biological reality; Hubbell observed that, in tropical forests, all species compete for light—and, therefore, space [31]. This means that neutral theory should be a good starting approximation for communities of sessile species that compete for a common resource, such as space (e. g. tropical trees or coral reefs). More realistic models will include non-neutral processes, such as interactions that depend on species identity [32,33], but neutral theory can act as a null model for assessing the weight of evidence for such processes. Although the SAD may be rather insensitive to the introduction of non-neutrality, this does not mean that it is identical for any kind of non-neutral effects [34]. While there may be patterns or scales for which some processes are undetectable, e. g. due to central limit theorem-like effect [19,35,36], strong interactions between individuals can structure communities and it is in some cases possible to detect their existence from inspection of the SADs [3,33,34]. If a data set is found to be consistent with neutral theory, we should therefore be able to infer that some particular non-neutral processes are not present in that community, or at least are not strong enough to produce detectable deviations from neutrality in a data set of this size. In this paper, we present a power calculation for neutral theory. Our purpose is to estimate an upper bound for the strength of non-neutral processes in tropical forest data sets [37–39] that have been found to be consistent with neutral theory [13]. To do this, we fit the standard neutral model (SNM) to data sets generated by a non-neutral model, and compute the probability of rejecting neutral theory. We test the neutral null hypothesis using a maximum likelihood approach (using an exact expression [26] for the likelihood of a sample from the SNM), where p-values are evaluated by a parametric bootstrap procedure. The power of a statistical test is defined as the probability that the null hypothesis is rejected when it is indeed false. The power therefore depends on which alternative hypothesis is true. In this paper, we focus on two classes of non-neutral processes: interspecific competition, and intrinsic (density independent) fitness differences between species. Interspecific competition is one of the classic mechanisms that promote coexistence [33,40,41], whereas differences in fitness represent the fact that the mean environmental conditions in a particular area of habitat will tend to favour one species over another. These represent opposite ends of a spectrum of possible non-neutral models, because symmetric interspecific competition tends to lead to equal abundances among species, whereas intrinsic fitness differences tend to lead to highly uneven abundances. While these are only two examples out of an infinite set of non-neutral models, our method provides a blueprint for computing the detectability of any type of non-neutral process. Full details of our models are given in the Methods section. Our models are similar in structure to Hubbell’s standard neutral model (SNM), in that we consider stochastic population dynamics in a local community where strong density dependence regulates the total community size to J individuals, coupled by immigration to a much larger metacommunity. We consider two models of interspecific competition: one, which we shall denote HL, is a multi-species stochastic Lotka-Volterra model similar to that studied by Haegemann and Loreau [33]; the other, denoted by PC, has stochastic Ricker-like dynamics as studied by Pigolotti and Cencini [42]. Our model of intrinsic differences in fitness, denoted by IF, assumes that the fecundity of each species is a randomly generated variable. Each of our models has a single parameter that determines how strong the non-neutral processes are. In model HL, parameter γ represents the relative difference in strength between interspecific and intraspecific interactions, so that when γ = 0 the dynamics are neutral whereas when γ > 0 coexistence is promoted. In model PC, parameter c determines the difference between inter- and intra-specific density dependence, so c = 0 corresponds to neutral interactions and non-neutrality becomes stronger as c increases. In model IF, the fitness of each species is generated from a Gamma distribution with shape factor 1/k, so that when k = 0 all fitnesses of the species are the same and the local dynamics are neutral. As in the SNM, local diversity is maintained by a fraction m of all recruits being immigrants from a metacommunity with fixed relative species abundances. The proportion of immigrants of different species follow their relative abundance in the metacommunity. We consider two cases: in case LOGS the metacommunity is described by a logseries with fundamental diversity constant θ, and in case EVEN the metacommunity has ST species which all have equal abundance. A logseries distribution can arise from many processes, including but not restricted to neutral dynamics [43]. We considered the even metacommunity because it represents a metacommunity limit of our local community dynamics, and as a result represents a contrasting, extremely non-neutral, limit to the logseries. When coupled to the LOGS metacommunity, each of our models should be equivalent to the SNM when the local dynamics are neutral (when γ, c, or k equals zero). As the dynamics are made more non-neutral, the deviations from the SNM should become stronger, and we expect the power of the test of the neutral null model to increase. However, when coupled to the EVEN metacommunity, the models are not equivalent to the SNM even when the local dynamics are neutral. In this case the power of the test could be high even if the local dynamics were neutral, though if J is very small the statistical power could still be low. Our study consists of two parts. First, we explore how the parameters of our models affect the probability of detecting non-neutral processes. For a real community we do not know a priori the appropriate parameters to use, so we need to choose the parameters so that the alternative model gives comparable patterns to the empirical data. In the second part of our study, we estimate the power of tests of neutrality for empirical data from three New and Old World tropical forests, including Barro Colorado Island (BCI) in Panama. Our power calculation provides an estimate of the smallest sample size that is needed to detect non-neutrality of known intensity, and of the range strengths of non-neutrality needed to reject neutrality for a given species abundance data set. The strength of non-neutral processes affects the sample size that is required in order to have a good chance of rejecting the neutral hypothesis (see Fig. 1). When interspecific interactions are non-neutral (models HL and PC, top and middle row of Fig. 1), we see a simple pattern: as the strength of non-neutral processes is increased (γ or c increases from zero), the power of the test increases. In addition, for these models the power of the test increases as the local community size J is increased. This shows that any strength of non-neutrality, however weak, can in principle be detected provided the data set is large enough. However, the system sizes needed to give a significant power may be too large to be empirically accessible when local dynamics are nearly neutral (γ or c close to zero). For the LOGS metacommunity, and when the local dynamics are strictly neutral (γ = 0 for model HL or c = 0 for model PC), the models are equivalent to the SNM, and the power is equal to the threshold p-value for statistical significance (0. 05 in our study). However, for the EVEN metacommunity the power can be higher than this threshold even when the local dynamics are neutral, because the immigration process makes the model no longer equivalent to the SNM (though this is not visible for the relatively small values of m used with models HL and PC in Fig. 1) However, the patterns are rather more complicated for model IF (bottom row of Fig. 1). The power is again low when the local dynamics are neutral (k → 0) and the metacommunity follows a logseries (bottom left panel). However, the power does not increase monotonically as the non-neutrality parameter k is increased. This is because strong selection rapidly leads to dominance by a single species [44], especially in small communities, which is a pattern that can also arise from the SNM if the immigration parameter m → 0. Moreover, the power no longer increases monotonically when the local community size J increases; for example the power for J = 2000 in Fig. 1, bottom left, is higher than for J = 200,5000, or 20000. This appears counterintuitive because statistical power should increase monotonically with sample size. However, J represents more than just the amount of data available: it is a parameter which interacts non-linearly with the model dynamics. In the IF model, for instance, it determines whether the dynamics are in the strong or weak selection limit, and it also plays a nonlinear role in the SNM. To illustrate this effect, we can consider the limit k → ∞ of model IF/EVEN. In this special case, there is a single dominant species, relative to which all other species have zero fitness. All local recruits will therefore be of the dominant species, though other species will also be present due to immigration. This case is particularly simple because the species identity of each individual in the local community is the dominant species with probability 1−m (1–1/ST), and each of the other species with probability m/ST. We find that the power of the test of neutrality is low at small J, increases to a maximum at an intermediate value of J, and then decreases as J increases again (Fig. 2). Because, for the non-neutral model in this limit, J is nothing more than a sample size—a community of size 2J can be constructed by adding two communities of size J—this non-monotonic relationship between power and J must be due to the nonlinear role played by J in the community dynamics in the neutral null model. The two other model parameters (immigration rate and diversity of the metacommunity) also strongly affect the chance of rejecting the neutral hypothesis. The parameter m describes the probability that a newly born individual in the local community is an immigrant from the metacommunity, so the local community resembles the metacommunity more closely as m is increased; when m = 1, the local community is effectively a random sample from the metacommunity and local dynamics are irrelevant. Increasing θ in model LOGS, and increasing ST in model EVEN, lead to more diverse metacommunities, and as a result tend to increase diversity in the local community. For models HL and PC, either increasing m (top and middle rows Fig. 3), or increasing the diversity of the metacommunity (either increasing θ or ST as appropriate, top and middle rows Fig. 4), reduces the power of the test. It is clear why increasing m should reduce the power for models HL/LOGS and PC/LOGS, because the local community is then more like a logseries, and hence more like the SNM. The power changes very little between m = 10−4 and m = 10−3, reflecting the much greater importance of local dynamics on the patterns when m is small. It is less clear why increasing θ should reduce the power of the test, though it is worth noting that both increasing θ and increasing m have the effect of increasing the local richness. Increasing m or ST in models HL/EVEN and PC/EVEN also increases the local richness, and while it is not obvious why this should make the model resemble the SNM, we find that it also reduces the power of the test. For model IF, the effect of both m and the metacommunity diversity on the statistical power can be non-monotonic (bottom row, Figs. 3 and 4). For model IF/LOGS when k ≈ 10−3, for instance, for θ = 1000 the power for m = 0. 03 is higher than for m = 0. 1 or 0. 01 (bottom left, Fig. 3); for m = 10−3 the power for θ = 100 is higher than for θ = 10 or 1000. On the whole, however, both m and the community diversity tend to increase the power of the test, which is the opposite effect from what is seen in models HL and PC. This is because the local dynamics tend to lead to monodominant states, which as explained before are indistinguishable from the SNM even if their origin is highly non-neutral. Processes which increase the local diversity allow the non-neutral features of this model to be better detectable. It is interesting to note that the shape of the relationship between power and c for model PC in Fig. 1, middle row seems to be independent of J. This would be a very useful relationship if it were found to hold in general, because power calculations for large J are very expensive computationally and this would enable us to estimate power by varying c as a proxy for J. We find, however, that this behaviour is not preserved for other parameter values. We did not find any simple way to summarise the dependence on model parameters evident in Figs. 1,3, and 4 that would enable us to estimate the power outside of the parameter range we tested explicitly. Our aim is to explore the detectability of non-neutrality in data sets of different sizes. SNM has been found to be statistically consistent with several large tropical forest data sets [12–17], but this does not mean that SNM is an exact description so a power calculation gives us, in principle, an upper bound for the degree of non-neutral processes in these systems. We do not know a priori the appropriate non-neutral parameter values for these forests, but we can choose model parameters so that the model data match a number of features of the empirical data. Specifically, we chose model parameters so that the community size, mean species richness, and mean Shannon diversity of sample data sets from the non-neutral model are the same as in the empirical data set. More details are given in the Methods section. We estimated the power of neutrality for empirical data similar to three data sets collated by the Centre for Tropical Forest Science (CTFS, data available online at http: //www. ctfs. si. edu). The determination of appropriate model parameters, and the power calculation itself, is very computationally expensive, so we performed this for three candidate strengths of non-neutrality, and for models PC and IF only. The results are summarised in tables 1–4. For model PC/LOGS, we find that the power of the test of neutrality is extremely low when the model parameters are chosen to match three statistics of the empirical data, whatever the strength of the non-neutrality (Table 1). This is because the non-neutral process tends to increase the evenness in the community, so when c is increased the fitted immigration rate needs to be increased in order to match the Shannon index in the empirical data. In other words, the parameters of this model need to be close to neutral (either c small, or m close to 1) in order to agree with the empirical data, so the power of the test is low. By contrast, for model IF/EVEN the power of the test is very high for model parameters that reproduce the characteristics of the empirical data (Table 4), even for the smallest strength of non-neutral processes we considered. This is because both the metacommunity and the local dynamics are non-neutral, but the local dynamics tends to lead to very uneven (i. e. monodominant) communities while the metacommunity tends to increase the evenness of the community. These processes need to be in balance in order for the model to match the diversity and richness of real data, and as a result the fitted model is far from neutral. For model PC/EVEN, both the local dynamics and the metacommunity tend to lead to even abundance distributions, and as a result it was only possible to find parameters so the model matches the empirical data when the strength of the non-neutrality was sufficiently weak (Table 2). The largest value of c for which the model is consistent with the empirical forest data was found to be when ST → ∞, in which limit the immigration process behaves effectively as a speciation process. It was found that when the model was fitted to the Pasoh data with c = 0. 7523, the probability of rejecting the neutral null hypothesis was 0. 20. The power of the test was much lower when c = 0. 1, and was always low when the model was fitted to the Lambir data. The model could not reproduce the richness and Shannon diversity of the BCI data set unless c < 0. 1. For related reasons, the probability of rejecting the neutral null model was low for model IF/LOGS when fitted to the empirical data. Here, both the non-neutral local dynamics and the metacommunity tended to lead to highly uneven abundance distributions, and as a result the Shannon index produced by the model was very low unless k was small. It was not possible to fit the model to Pasoh when k ≥ 10−4, or to BCI or Lambir when k ≥ 0. 01. While there were two discrete parameter sets each that matched the richness and evenness of Pasoh and Lambir (one where θ was low and m high, and one where θ was higher and m lower), the power of the test with these parameters was always low. Our power calculation shows that, in principle, non-neutrality would be detectable in large enough SAD data sets or when non-neutral processes are strong enough (see Fig. 1). This contradicts the suggestion that the SAD for large samples will approach the same canonical form, and that larger sampling efforts would consequently be futile [35]. Indeed, the SNM has been rejected using SAD data from very large phytoplankton communities [45], and we found that the SNM could also be rejected for the tropical tree species abundances of Yasuni National Park (see Methods). Our results also show that independent niches can be distinguishable from neutrality, contrary to suggestions by Chisholm and Pacala [36], because the test is most powerful when the species in our model HL undergo independent stochastic logistic dynamics (i. e. when γ approaches 1, see Fig. 1). Our HL model with γ = 1 differs from the independent-niche models of Chisholm and Pacala [36] and Haegeman and Etienne [19] by having strong intra-specific density dependence, so the marginal distributions are very different from the SNM. However, we conclude that tropical forest abundance data sets, on the scale collected by CFTS, might not be large enough to detect even strongly non-neutral interspecific interactions. As shown in Tables 1 and 2, statistical power remains very low as c (the parameter measuring the intensity of inter- versus intra-specific competition) is varied. This is because the model parameters required to give the same richness and evenness in the empirical data are themselves close to neutral, either because c is small or because the metacommunity follows a logseries distribution and m is close to 1 (in which case the local community strongly resembles a neutral-like metacommunity). This result is in agreement with the good fits of some species-independent neutral models to a large number of SADs for very diverse communities [46]. It is interesting to note that Volkov et al. [47] estimated that interspecific species interactions are many times smaller than intraspecific interactions in tropical forests, but we cannot apply their results to our models because they did not include immigration from a metacommnunity. On the other hand, we did find that a combination of non-neutral processes producing opposing effects on the local community led to high statistical power for parameters consistent with empirical CTFS data. In model IF/EVEN, intrinsic local fitness differences tend to decrease richness and evenness, whereas the even local community increases richness and evenness. This means that both processes can be strong while still producing levels of richness and abundance consistent with the empirical data. Du et al. [10] noted that non-neutral processes which have opposing effects on relative abundance distributions can lead to abundance distributions that resemble neutral theory, but our investigation shows that they can still be distinguished from SNM in some cases. We can therefore conclude that such a combination of strongly non-neutral processes is not present in data sets for which the SNM is not rejected, such as the three CTFS forests we studied in this paper. The power of a statistical test generally depends on three factors: first, the sample size; second, statistical significance as measured by the threshold p-value used to assess significance; and third, the effect size, which quantifies departures from the null hypothesis. In our analysis, effect size is encoded in the parameter values of the non-neutral model under consideration (respectively, γ, c, and k for models HL, PC, and IF). When density dependence is non-neutral (models HL and PC), power increases as interactions become more non-neutral (see Fig. 1). However, for non-neutral intrinsic fitness (model IF) the power of the test depends non-monotonically on k (Fig. 1, bottom row). These patterns can be understood from the effect that these parameters have on diversity patterns—strong non-neutrality (k large) in model IF leads to monodominance, which is indistinguishable from the neutral model with strong dispersal limitation (m very small). Our results highlight the fact that the parameter J plays a more complicated role for these models than the sample size in standard power calculations, because the power does not always increase monotonically with J (Fig. 2). In most standard statistical tests, a “sample” consists of a number of statistically independent measurements. In an ecological community (or a model thereof), the individuals are not statistically independent because of their interactions (whether within or between species). This is true even in the SNM: an equilibrium community of size J can be generated as a hypergeometric subsample of a community with larger J [48], but the individuals are not independent because this represents sampling without replacement. This means that the community size J plays a nonlinear role and is not a straight analogue of the sample size in standard statistical tests, so statistical power does not necessarily increase monotonically with J. The dependence on other parameters can also be non-monotonic; for example, for metacommunity model LOGS the local community will resemble a log series (and therefore be indistinguishable from the SNM) in the limit m → 1, but the power does not decrease monotonically with m for model IF/LOGS (see Fig. 3). Increased metacommunity diversity decreases statistical power for models HL and PC (Fig. 4), which echoes the observation that higher local diversity leads to SADs that look more like those created by the SNM even in the presence of niche structuring [19,36]. This suggests that it might be easier to quantify non-neutral interactions in less diverse forests [19]. However, this is not true for all types of non-neutral processes: for model IF the power increases when the metacommunity diversity is increased. Unfortunately, we were unable to find any general rules to allow us to extrapolate the power calculation outside of the parameter range we simulated. The power calculations in this paper are very computationally expensive, and it would be unfeasible for us to repeat them for J much larger than ∼ 30000 individuals. Moreover, to do this we would need to know how parameters γ (for model HL) and c (for model PC) are affected by J. Our notation tacitly assumes that each species is sensitive to mean population densities over the whole community, but in real systems, where individuals of a species are clumped together, an individual will only interact with nearby individuals so the values of γ and c might depend on J. We are therefore unable to estimate the factor by which CTFS data sets would have to be enlarged for us to distinguish model PC from the SNM. In this paper, we have analysed a range of non-neutral scenarios: non-neutral density dependence affecting mortality or recruitment; non-neutral differences in intrinsic fitness; neutral-like or extremely non-neutral metacommunity. These processes have contrasting effects on the SAD, so arguably represent the extremes of a spectrum of possibilities. Nevertheless, there are many types of community processes which are not encompassed by our models. For example, trophic or mutualistic interactions are not present in our models and should lead to very different patterns of abundance, though these are more likely to be relevant in other systems than the tropical forests on which we focus. Similarly, dynamics that lead to multimodal SADs should be relatively easy to distinguish from neutrality [34,49]. We have made a number of simplifying assumptions to keep the number of parameters manageable, but our framework could still be used to perform a power calculation for any type of non-neutrality that can be incorporated in a simulation model. It would also be preferable to perform power calculations for spatially explicit models, which represent a more realistic dispersal process and can readily be simulated [50]; however, for our test we would need a likelihood for the spatially explicit neutral model, which is not currently available. Our results have refined as well as quantified in a statistical sense the suggestion [3,36,51,52] (but see also ref. [53]) that SAD data do not have sufficient resolving power to assess the importance of non-neutral processes in structuring forest communities. Our study shows that even large-scale tropical forest data sets are not large enough, or are too diverse, to detect non-neutral species interactions using the SAD alone. However, we would not expect to see a good fit to the SNM if these forests contained multiple processes with opposing effects on richness and evenness. Patterns that include more information, such as multiple samples [27], spatiotemporal changes [54,55] or phylogenetic data [56,57], are likely to be much more revealing about the processes that generated them. Provided it is possible to compute the likelihood for obtaining such patterns in a neutral model, our approach can be adopted to calculate the sampling effort needed to detect and quantify non-neutral processes, and understand the forces that structure communities. Our local community model “HL” resembles one used by Haegeman and Loreau [33]. It can be thought of as a multispecies stochastic Lotka-Volterra competition model with immigration, where a single parameter controls the relative strength of inter-and intra-specific interactions. Our model differs from that of Haegeman and Loreau [33] in that each death event is immediately followed by a single birth event so that the local community size remains constant. We consider a local community consisting of J individuals, each of which has a species identity which is an integer between 1 and ST. Mortality is affected by inter- and intra-specific density dependent mortality, so that the probability that the next individual that dies has species identiy i is proportional to M i ({ n j }) = n i J + (1 - γ) ∑ j ≠ i n j J n i J = (1 - γ) + γ n i J n i J (1) where ni is the number of individuals of species i in the community, so that J = ∑ j = 1 S T n j. In the neutral case, γ = 0, the mortality rate is the same for all individuals irrespective of species, but when γ > 0 per capita mortality is greater for more abundant species. The dynamics proceed by choosing the individual that dies next, so that the probability that the dead individual has species identity i is proportional to Mi. A recruit is then chosen to be of species i with probability F i ({ n j }) = (1 - m) n i J + m P i, (2) where Pi is the relative frequency of species i in the metacommunity. One time step consists of J of these elemental update steps. If γ = 0, the mortality rate is independent of species identity, so species interactions are neutral; when 0 < γ ≤ 1, niche differentiation tends to promote species coexistence [33]. The model is ill-defined when γ > 1, since that would lead to Mi < 0. At first sight, it might appear that a more general model could be obtained by using the functional forms in Haegeman and Loreau [33], which allow for density independent as well as density dependent mortality. The rates in Equation (13) of that paper correspond in our notation to M i = r - n i + (r + - r -) n i α J + (1 - α) n i K ′ (3) F i = r + n i + μ S T P i, (4) where both Mi and Fi are now rates rather than probabilities (no longer normalised so that they sum to unity) and we have introduced the factor ST Pi to allow the immigration rate to differ between species. Here, r+ and r− denote respectively the rates of density-independent birth and mortality, K′ plays the role of a carrying capacity, and α tunes how neutral the interactions are are (neutral when 1; non-neutral competition for 0 < α < 1; mutualistic for α < 0). A little algebra shows that Equations (3) and (4) are equivalent to Equations (1) and (2) (up to overall prefactors that do not affect the sequence of processes in the simulation) with the choice of parameters γ = 1 1 + r - K ′ + (r + - r -) J α (r + - r -) J (1 - α) m = 1 1 + r + J μ S T. The values of these parameters is within the range for which our model is defined (0 < m ≤ 1,0 ≤ γ ≤ 1) provided r+ > r− (which is assumed to be the case by Haegeman and Loreau [33] in order for there to be a non-trivial equilibrium) and − r − K ′ (r + − r −) J ≤ α ≤ 1. Therefore, our model defined by Equations (1) and (2) captures the apparently more general density dependence defined in Equations (3) and (4), except for the case of strongly mutualistic interactions. Our second model of non-neutral species interactions is a derivative of that of Pigolotti and Cencini [42]. As in model HL, we assume that interspecific interactions are weaker among heterospecifics than among conspecifics, but in model PC we assume a Ricker-like functional form that acts on fecundity so that the number of local propagules of species i ∈ {1, …, ST} is ∝ ni exp (−ani − b∑i ≠ j nj), where nk is the local abundance of species k and a and b (< a) are constants. The fraction of local propagules that are of species i is then L i P C = n i e - a n i - b ∑ j ≠ i n j ∑ k = 1 S T n k e - a n k - b ∑ j ≠ k n j = n i e - c n i / J ∑ k = 1 S T n k e - c n k / J, where c = J (a − b). If the interactions are sensitive to the average local density of the different species, i. e. all species are spread throughout the community, then for the same pool of species we expect c to be independent of J. Spatial effects could lead to a focal species only being sensitive to the dynamics of nearby species, in which case the effective value of c would depend on J, although this can only be modelled correctly using a spatially explicit model. Because a fraction m of recruits are immigrants from the metacommunity, the probability that a new recruit is of species i is R i = m P i + (1 - m) L i P C. (5) In this model, we assume that the generations are discrete and non-overlapping: at each timestep, we compute the Ri from the current configuration, and generate a new configuration using { n i } ∼ multinomial (J, { R i }). We do this for the sake of computational efficiency: using a multinomial pseudo random number generator when the local community size is of the order of J ∼ 10000 a full system update of this synchronous model is orders of magnitude quicker than for the sequential update model. Sequential updating (as in model HL above) is a more faithful biological description of triopical forest dynamics, but it is known that in the neutral limit the sequential model (Moran process) and the synchronous model (Wright-Fisher process) give indistinguishable equilibrium statistics for the large community sizes we are interested in. The syncronous model is therefore well suited to our goal of exploring the detectability of departures from neutrality. The only circumstances where the synchronous model behaves qualitatively differently from the sequential one is where the Ricker-like dynamics tend to lead to limit cycles or chaos, but that does not affect the results in this paper because we always choose c < 2ST (see Metacommunity model EVEN below). Interspecific interactions, as implemented in models HL and PC, could be interpreted as representing environmental variability within the community: each species has its own preferred microhabitat, and as a result competes less with heterospecifics (which occupy neighbouring but different microhabitats) than with conspecifics. By contrast, model IF could be interpreted as considering environmental variability between communities. In any given local community, each (out of a total number ST) species has a different intrinsic fitness fi, different from the other species and different from its intrinsic fitness in other commuities. Our model introduces intrinsic fitness differences in a similar way to Chesson and Warner [58], though our model is otherwise different because our fitness differences do not fluctuate in time and our dynamics are stochastic rather than deterministic. For each realisation, we generate the fi from a Gamma distribution with mean shape factor 1/k, so that k1/2 is the coefficient of variation among the fitnesses, and all the fitnesses are equal when k → 0. The fraction of local propagules that are of species i is then L i I F = f i n i ∑ k = 1 S T f k n k, and the probability that a recruit is of species i is R i = m P i + (1 - m) L i I F. As was the case for model PC, for computational efficiency we assume discrete generations, and simulate the model using a multinomial pseudorandom number generator with probability vector {Ri}. We assume that the relative immigration rates of different species reflect their abundance in a wider metacommunity. Following the SNM, in model LOGS we assume that this abundance distribution follows a Fisher logseries with diversity parameter θ. This is often a good description of empirical data, and can arise from several models of community dynamics including Hubbell’s neutral model [31,59]. Note that the metacommunity represents the pool from which immigrants can be drawn, which could comprise many disparate communities. Therefore, the metacommunity does not necessarily correspond to the large-J limit of a single local community model. This means that one reasonable scenario is that the metacommunity follows a canonical form form due to averaging over very large scales [35], even when the local dynamics are non-neutral. In metacommunity model LOGS we use the distribution introduced by Ewens [60] to give the number fM of species in the metacommunity with relative abundance x within the interval (x; x + dx). f M (x) d x = θ x (1 - x) θ - 1 d x. (6) This distribution is a continuum form of Fisher’s log-series that is appropriate when sampling from an effectively infinite metacommunity [14]. In practice, we use this distribution to simulate from a very large metacommunity containing ST species; our full sampling algorithm is described in S1 Text. The results in this paper are for ST = 2000, which is large enough to be effectively infinite (i. e. a choosing a larger ST did not have a perceptible effect on community statistics or the power of the test of neutrality, but did increase the duration of the simulation). As explained above, the logseries is a reasonable candidate metacommuity model even when local community dynamics are non-neutral. However, it is also the metacommunity model in the SNM, so we want to consider the possibility that non-neutral processes are visible in the metacommunity as well. In our EVEN metacommunity model, there are ST species with equal relative abundance, P i = 1 S T. This distribution has been used in previous modeling studies of neutral and non-neutral community dynamics [33,54,61,62], though it has little empirical support. It is appropriate to use this distribution in our study because, as we shall show, it represents a metacommunity limit of our non-neutral local community models. The HL model is a form of stochastic multi-species Lotka-Volterra model, so its large-J limit is described by differential equations. When 0 ≤ α < 1, this has a stable equilibrium with n i J = 1 S T for all i. The PC model is a form of Ricker map. When J is large, the multinomial distribution becomes sharply peaked around its mean value, so from Equation (5) (for vanishing m) the dynamics of r i = n i J follows r i (t + 1) = r i (t) e - c r i (t) ∑ k r k (t) e - c r k (t). A standard stability analysis shows that the equilibrium r i = 1 S T is stable provided cST < 2; for higher values of c the community displays limit cycles or chaos. In model IF, the community tends to be dominated by the species that has the highest fitness. However, the metacommunity represents an aggregate of many independent local communities, and we expect different species to dominate in different communities. The model assigns fitnesses independently at random to the different species, so we expect each species to have the same overall relative abundance 1 S T in the metacommunity. Therefore, the EVEN metacommunity model represents one metacommunity limit of our local community models. Other metacommunity models could be obtained by taking the limit in different ways. For instance, if parameter c in model PC depends on J and approaches zero sufficiently rapidly in the limit J → ∞, then the metacommunity limit would be neutral and follow the same Ewens distribution as model LOGS. If the fitnesses in model IF were not i. i. d. random variables, but rather different species had different mean fitnesses, then the metacommunity would have a different, uneven distribution. While there is an infinite variety of possible metacommunity distributions, the EVEN metacommunity represents the most contrasting distribution to the logseries, in the sense that it has the maximum Shannon diversity index for a given species richness while the logseries is a very uneven distribution. It also has the advantage of being characterised by a single parameter (ST), whereas other commonly-used distributions (e. g. the lognormal) generally require two parameters. In order to quantify whether a particular data set is consistent with neutral theory, we adopt a maximum likelihood approach together with a parametric bootstrap as used by Walker and Cyr [45] and Rosindell and Etienne [63]. To calculate the p-value of our test, we compare the value of a test statistic for the test data set with values of the test statistic for data sets generated by the null model. We choose the maximized likelihood of the neutral model as our test statistic. The likelihood L (X∣m, θ) that the neutral model would generate a data set X, for parameters (m, θ) is computed using the exact formula derived by Etienne [26]. We use code based on Tetame (http: //chave. ups-tlse. fr/projects/tetame. htm), an efficient implementation in C++ of Etienne’s formula that was developed by Jabot et al. [64]. We have ported the Tetame code to C, and adapted it so it can be loaded as a dynamic library in R using the. C () function. The hypothesis test consists of the following steps: The power of a statistical test is defined as the probability that the null hypothesis is rejected when it is indeed false. The statistical power can only be quantified by specifying an explicit model to represent the alternative hypothesis. The power can be computed by simulating many data sets from the alternative model, and performing a test of the null hypothesis on each data set, as explained above. The power is the fraction of cases for which the null hypothesis is rejected. A Type II error is defined as the failure to reject the null hypothesis when the alternative hypothesis is true, so the power is equal to 1 − β, where β is the probability of Type II errors. The power of a test will depend on the magnitude of the deviation from the null hypothesis—the so-called effect size—and on the quality of the data at hand, typically, sample size. When coupled to the LOGS metacommunity, our models are equivalent to the SNM in the limit where the non-neutral parameter (γ for model HL; c for model PC; k for model IF) is zero, so γ, c, or k is our effect size for these models. When coupled to the EVEN metacommunity, our models are never strictly equivalent to the SNM so the effect size cannot be defined. In general, the power of the test will also depend on other model parameters, so we need to perform power calculations for a wide range of potentially interesting parameter values. To calculate the power of tests of neutrality for a non-neutral model with a particular set of model parameters YT, we use the following procedure: Because each non-neutral and neutral data set is statistically independent, the power is a binomial proportional random variable. Where shown, confidence intervals are 95% Jeffreys intervals [65]. Our focus in this paper is on the detectability of non-neutral processes neutral in empirical situations. The power of the test is a property of the ensemble of data sets that could be produced if the data were generated by a non-neutral alternative model. This ensemble of data sets depends on the model parameters which are chosen, but for a particular data set we do not necessarily know the appropriate parameters to use in the alternative model. Here, we choose parameter sets such that the model best describes a set of summary statistics of empirical data sets, specifically: species richness and Shannon index. Once the strength of the non-neutral process (α, c, or k as appropriate) and the local community size J are chosen, the model is characterised by two further parameters: the immigration rate m and the diversity (θ for model LOGS or ST for model EVEN) of the metacommunity. There will therefore be a discrete set of parameter values where the mean species richness and mean Shannon index of samples from the model match the empirical data sets; in most cases we found only one such parameter set, though in some cases there were two and in others there were none because the model produced a Shannon index that was always higher than, or always lower than, the empirical data. We performed this procedure, for a set of candidate values of the non-neutral parameter, to generate parameter sets resembling three tropical forest data sets belonging to the CTFS network to which neutral theory has successfully been fitted in the past [46]: Barro Colorado Island, Pasoh Forest Reserve, and Lambir Hills National Park [37–39]. For each forest, and separately for each survey year, we tested the null model that the data were generated by the SNM using the parametric bootstrap method described above. In each case we found p > 0. 05, showing that the data were statistically consistent with SNM. The mean total community size, species richness, and Shannon index for these sites, averaged over the census years available at http: //www. ctfs. si. edu, are given in Table 5. These sites were selected because they had higher Shannon index than a logseries distribution with the same size and richness, so we expected that a model with non-neutral interspecific interactions or an EVEN metacommunity would describe the data better than the SNM. Volkov et al. [46] have compared the SNM to three other CTFS forest sites, but the Shannon index in these sites is lower than (Korup National Park and Yasuni National Park) or almost equal to (Sinharaja World Heritage Site) that of a logseries, so we expected it to be more difficult to find suitable model parameters. We also found that the Yasuni National Park data were not consistent with SNM (p = 0. 001 for 1996 and p = 0. 014 for 2003), though we found p > 0. 05 for all Korup and Sinharaja surveys.

In order to predict and mitigate the response of ecological communities to global change, we need to understand the processes that allow multiple species to coexist in close proximity. A classic idea in Ecology is that species coexist because they occupy different “niches”. However, random processes such as dispersal could also explain species coocurrence, without invoking niche differentiation. “Neutral” models embody this idea, omitting niche differentiation and assuming all species are identical. Such models are mostly statistically consistent with the relative abundances of tree species in tropical forests, but statistical procedures always contain an element of uncertainty and many other models could also be consistent with a particular data set. We compute how strong the non-neutral processes would need to be in order for their effect to be detectable in data sets of different sizes. We find that the largest ecological data sets currently available, such as the 50 hectare plot on Barro Colorado Island in Panama, are not large enough to distinguish between neutral and non-neutral models, unless multiple non-neutral processes are at work. This means that other types of pattern need to be studied, or larger data sets collected, in order to understand the mechanisms behind forest biodiversity.

Abstract Introduction Results Discussion Methods

2015

When Can Species Abundance Data Reveal Non-neutrality?

Mammalian Peptidoglycan Recognition Proteins (PGRPs) are a family of evolutionary conserved bactericidal innate immunity proteins, but the mechanism through which they kill bacteria is unclear. We previously proposed that PGRPs are bactericidal due to induction of reactive oxygen species (ROS), a mechanism of killing that was also postulated, and later refuted, for several bactericidal antibiotics. Here, using whole genome expression arrays, qRT-PCR, and biochemical tests we show that in both Escherichia coli and Bacillus subtilis PGRPs induce a transcriptomic signature characteristic of oxidative stress, as well as correlated biochemical changes. However, induction of ROS was required, but not sufficient for PGRP killing. PGRPs also induced depletion of intracellular thiols and increased cytosolic concentrations of zinc and copper, as evidenced by transcriptome changes and supported by direct measurements. Depletion of thiols and elevated concentrations of metals were also required, but by themselves not sufficient, for bacterial killing. Chemical treatment studies demonstrated that efficient bacterial killing can be recapitulated only by the simultaneous addition of agents leading to production of ROS, depletion of thiols, and elevation of intracellular metal concentrations. These results identify a novel mechanism of bacterial killing by innate immunity proteins, which depends on synergistic effect of oxidative, thiol, and metal stress and differs from bacterial killing by antibiotics. These results offer potential targets for developing new antibacterial agents that would kill antibiotic-resistant bacteria. Mammalian Peptidoglycan Recognition Proteins (PGRPs) are a family of four evolutionary conserved antibacterial innate immunity proteins [1]–[3]. Three PGRPs (PGLYRP1, PGLYRP3, and PGLYRP4) are directly bactericidal [4], [5] and one PGRP (PGLYRP2) is a peptidoglycan-lytic amidase [6]. PGRPs kill both Gram-positive and Gram-negative bacteria [4], [5] by a novel mechanism [7]. PGRPs activate envelope stress responses in bacteria, which results in membrane depolarization and intracellular production of toxic hydroxyl radicals (HO•), which leads to energy depletion and inhibition of intracellular synthesis of peptidoglycan, proteins, RNA, and DNA, and cell death [7]. Bactericidal PGRPs do not inhibit extracellular peptidoglycan synthesis, do not hydrolyze the cell wall, and do not kill by permeabilizing bacterial membranes, or by osmotic lysis [4], [5], [7]. The induction of envelope stress by PGRPs in two model Gram-positive and Gram-negative bacteria is to a large extent dependent on the inappropriate over-activation of two-component systems that normally function to detect and dispose of misfolded proteins in bacteria, CssRS in Bacillus subtilis, and CpxRA in Escherichia coli [7]. The exact nature of the signal that activates CssRS and CpxRA is not known, because these two-component systems respond to many other types of stress besides misfolded proteins, including pH, osmolarity, Cu, and Zn [8]. In this study we investigate the down-stream events that are responsible for PGRP-induced bacterial killing. We first focused on the role of oxidative stress and reactive oxygen species (ROS) in PGRP bacterial killing, because we could inhibit bacterial killing by inhibiting PGRP-induced HO• production [7]. Detailed evaluation of the role of ROS in PGRP-induced killing was important, because the previously reported antibiotic-induced killing of E. coli that was also based on CpxRA-dependent induction of HO• [9], [10] was called into question by recent reports showing that antibiotic-mediated killing of E. coli does not depend on ROS, as bactericidal antibiotics did not induce H2O2 production or corresponding oxidative stress responses that would signal the presence of elevated levels of H2O2 [11]–[13]. Our results presented here show remarkably similar responses to PGRPs in both E. coli and B. subtilis. Both model organisms displayed similar transcriptomic signatures upon treatment with PGRPs, including induction of oxidative, thiol, and metal stress responses, along with corresponding increases in intracellular H2O2 and metals and depletion of thiols. We demonstrate that all these three responses are required, but individually are not sufficient for bacterial killing by PGRPs. We further show that bacterial killing can be efficiently reconstituted by the simultaneous treatment with chemicals that lead to production of ROS, depletion of thiols, and elevation of intracellular metal concentrations. These results indicate that killing of bacteria by PGRPs involves synergistic effects of oxidative, thiol, and metal stress and is different than killing by antibiotics. To gain further insights into the mechanism (s) of PGRP-mediated killing of bacteria, we used the unbiased approach of whole genome expression arrays to identify stress response pathways activated in PGRP-treated bacteria. We treated bacteria with human PGRP and after 30 min we isolated RNA (before the numbers of viable bacteria recovered by colony counts began to significantly decrease). We used albumin as a negative control, and we used two well-characterized bactericidal compounds as controls. The first was gentamicin, an antibiotic that activates the same misfolded protein-sensing two-component systems as PGRP [7], [10], but which was also recently shown not to induce H2O2 production or oxidative stress responses in E. coli [12]. The second was CCCP (carbonyl cyanide 3-chlorophenylhydrazone), a membrane potential de-coupler, which, similar to PGRP, induces membrane depolarization in bacteria [7]. Using whole genome expression arrays in three independent experiments we detected expression of 5,531 probes in E. coli and 3,355 probes in B. subtilis, of which 1,510 and 536 probes were expressed significantly higher in PGRP-treated E. coli and B. subtilis, respectively, than in albumin-treated bacteria, and 1,988 and 617 probes were expressed significantly lower in PGRP-treated E. coli and B. subtilis, respectively, than in albumin-treated bacteria (as determined by one-tailed t-test at P≤0. 05). Further calculation of FDR (false discovery rate) q values identified 2,733 and 795 probes in E. coli and B. subtilis, respectively, whose expression was significantly changed in PGRP-treated compared with albumin-treated bacteria at q≤0. 05. In E. coli 2,008 genes and in B. subtilis 1,236 genes were either up-regulated or down-regulated more than 3 times by any of the three treatments (PGRP, gentamicin, and CCCP, Figures S1 and S2). We confirmed increased expression of representative 25 E. coli and 28 B. subtilis up-regulated genes using quantitative real time PCR (qRT-PCR, Tables S1 and S2). The results showed remarkably similar effects of PGRP on gene expression in E. coli and B. subtilis. Virtually all top PGRP-induced genes were involved in defense against oxidative, thiol (disulfide), and metal stress, or in repair of the cellular damage in bacteria caused by these stresses (Figure 1 and Tables 1 and S1, S2, S3, S4). They included: (i) peroxide detoxification genes (oxyS, ahpF, katG in E. coli, and katA, katE, ohrB, ahpF, ahpC in B. subtilis) induced by peroxide-responsive OxyR in E. coli, and PerR in B. subtilis; (ii) genes involved in detoxification of ROS and epoxides (paa operon in E. coli controlled by Crp and Ihf); (iii) genes involved in efflux and detoxification of copper, zinc, arsenite, and other metals induced by metal-responsive or stress-responsive regulators (CueR, ArsR, SoxR, RcnR in E. coli, and CsoR, CzrA, ArsR in B. subtilis); (iv) genes coding for chaperones and protein, RNA, and DNA quality control induced by stress-responsive regulators (σH, Ihf, CpxRA in E. coli, and σB, CtsR, CssRS in B. subtilis); and (v) genes for repair and synthesis of Fe-S clusters (controlled by IscR in E. coli). The remaining groups of highly induced genes also reflect bacterial response to oxidative, thiol, and metal stress and function in energy generation, synthesis or uptake of methionine and histidine, and defense against general stress (Figure 1 and Tables 1, S1 and S2). The majority of genes highly up-regulated by PGRP were not induced or induced less by gentamicin and CCCP (Figures 1, S1 and S2, Tables S1 and S2). Many oxidative stress, energy acquisition, and methionine and histidine biosynthesis genes (in both E. coli and B. subtilis), and some metal detoxification and Fe-S biosynthesis genes (in E. coli) and genes for transporters, envelope remodeling, and general stress response (in B. subtilis) were induced less (or not at all) by gentamicin compared with PGRP. However, both PGRP and gentamicin induced SoxR-regulated soxS and marRAB genes (which control drug resistance in E. coli) and several genes for protein quality control (in both E. coli and B. subtilis). The gene induction patterns by CCCP in E. coli and B. subtilis were also unique and different from the pattern induced by PGRP or gentamicin, with induction of several oxidative stress genes and energy acquisition genes, and some metal detoxification genes (Figure 1 and Table S1). Different patterns of gene activation by PGRP and other antibacterial compounds and also overlapping activation of genes by PGRP for oxidative, thiol, metal, and also envelope stress were further revealed by hierarchical cluster analysis by comparing PGRP-activated genes with previously published gene array data in bacteria exposed to H2O2, diamide (thiol-oxidizing agent), Zn, and vancomycin (inhibitor of peptidoglycan synthesis). This analysis revealed clusters of genes induced primarily by PGRP (e. g. , several OxyR-induced and DNA repair genes), and several clusters of PGRP-induced genes overlapping with genes induced either by H2O2, or diamide, or Zn, or vancomycin (Figure S3). Altogether, our gene expression results suggest simultaneous induction of multiple stress responses by PGRP. Inspection of genes down-regulated after PGRP treatment was also informative. The most down-regulated genes in both E. coli and B. subtilis were for: (i) Fe uptake, controlled by the Fur regulator in both bacteria; (ii) motility, controlled by CpxRA in E. coli; and (iii) phosphate utilization, controlled by PhoPR in B. subtilis (Figures S1 and S2, Tables 1, S3 and S4). Thus, our gene expression results indicate that PGRPs induce oxidative stress, thiol stress, and metal stress in bacteria, and our next experiments were designed to verify these responses biochemically and to determine which of these responses are involved in bacterial killing. In both E. coli and B. subtilis, PGRPs induced expression of genes typical of oxidative stress, including genes regulated by intracellular peroxide sensors, OxyR and PerR (Tables 1, S1 and S2). We therefore tested the hypothesis that PGRPs induce production of H2O2 in bacteria. Oxidative stress can arise from the intracellular production of superoxide anion (O2−), which is then converted into hydrogen peroxide (H2O2) and then into HO•, which are collectively known as ROS [14]. Thus, our hypothesis was also consistent with our previous results showing induction of HO• by PGRPs in bacteria [7]. To directly verify this hypothesis, we measured production of H2O2, because H2O2 is more stable than other ROS (O2− and HO•) and diffuses readily across membranes facilitating its detection. To detect H2O2 production, we used mutants, designated Hpx−, deficient in the major H2O2 degrading enzymes catalase (kat) and alkyl hydroperoxide reductase (ahp) (E. coli ΔkatGΔkatEΔahpCF and B. subtilis ΔkatAΔahpCF) [12], [15]–[17]. Treatment of bacteria with human recombinant PGRP [4], [5], [7] or paraquat (an O2− and H2O2-inducing positive control) [12], [18] strongly induced intracellular H2O2 production in both E. coli and B. subtilis, which was maximal at 15 min (Figure 2A), remained equally high at 30 min, and began to decline after 60 min, likely due to instability of H2O2 (data not shown). H2O2 was not induced by albumin (negative control) or diamide (thiol-oxidizing disulfide stress-inducing agent as another control) (Figure 2A). To determine whether PGRP-induced ROS are required for PGRP-induced killing, we determined the requirement for oxygen for PGRP-induced bacterial killing, as ROS cannot be formed in the absence of oxygen. In the presence of oxygen, PGRP reduced the numbers of E. coli and B. subtilis by nearly 4 logs in 4 hrs. However, in the absence of oxygen (90% N2,5% H2,5% CO2), PGRP did not kill E. coli, and under microaerophilic conditions (1% O2) PGRP did not kill B. subtilis either (Figure 2B). However, under anaerobic or microaerophilic conditions, PGRP was still bacteriostatic for both bacteria. These results show that oxygen is required for PGRP-induced killing, and also indicate additional oxygen-independent antibacterial mechanisms of PGRPs. Oxidative damage of DNA by ROS greatly contributes to their toxicity, and mutants deficient in the excision or recombinational repair of oxidative DNA lesions are especially sensitive to oxidative stress [12], [14], [17]. Accordingly, a ΔrecA E. coli mutant was significantly more sensitive to PGRP than WT bacteria (Figure 2C). These results are consistent with the hypothesis that oxidative DNA damage significantly contributes to the bactericidal effect of PGRPs. To further determine the role of ROS in bacterial killing, we evaluated killing of WT and Hpx− E. coli and B. subtilis by PGRP, paraquat (which directly induces intracellular H2O2 production), and exogenously added H2O2. PGRP readily killed WT and Hpx− E. coli and B. subtilis, and Hpx− mutants were more sensitive to PGRP killing than WT E. coli and B. subtilis (Figure 3A, E). However, the concentrations of paraquat (5–250 µM) that induce comparable amounts of H2O2 production as PGRP (∼1. 5 µM H2O2 induced by 5 µM paraquat, compared with 1. 2–2. 2 µM H2O2 induced by PGRP in Figure 2A) were only bacteriostatic and did not kill WT E. coli and B. subtilis (Figure 3B, E). Although Hpx− mutants were more sensitive to paraquat than WT bacteria, they were still not killed (E. coli) or killed inefficiently (B. subtilis) by 5–250 µM paraquat (Figure 3B, D, E). Only high concentration of paraquat (500 µM) was bactericidal for Hpx− mutants, but still not for WT bacteria (Figure 3D). Similarly, only very high concentrations of exogenously added H2O2 (200–640 µM) were bactericidal for Hpx− mutants, but still not for WT bacteria (Figure 3C, F). These results indicate that the amounts of H2O2 induced by PGRP or by 5–250 µM paraquat (∼2 µM H2O2, Figure 2A) are not sufficient to kill bacteria. Altogether, these results demonstrate that ROS are induced by PGRPs and are required for their bactericidal activity, but that physiologically relevant concentrations of ROS induced by PGRPs are not sufficient for bacterial killing. Therefore, these results suggest that other killing mechanisms work together with O2-dependent generation of ROS in eliciting the bactericidal activity of PGRP. We then tested the hypothesis that PGRPs cause thiol (disulfide) stress by inducing depletion of intracellular thiols, because the pattern of gene induction by PGRP was similar to the previously reported pattern of gene induction by diamide (a thiol-depleting electrophile), including activation of genes for the same metal detoxification systems, chaperones, protein quality control, and thiol stress responses [19], [20]. PGRP, similar to diamide, depleted over 90% of intracellular thiols in E. coli and B. subtilis within 30 min of exposure (Figure 4A), and these low levels of thiols were maintained for at least 2 hrs both in PGRP- and diamide-treated bacteria (data not shown). Paraquat only minimally reduced intracellular thiols (Figure 4A) at a concentration that strongly induced H2O2 production, comparable to PGRP-induced H2O2 production (Figure 2A). Altogether, our results show that PGRPs induce both H2O2 production and thiol depletion, whereas paraquat and diamide selectively induce either H2O2 production or thiol depletion, respectively. We next tested the role of intracellular thiols in PGRP killing. Exogenous thiourea (a membrane-permeable thiol that inhibits depletion of thiols and counteracts the effects of thiol and oxidative stress) significantly diminished bactericidal activity of PGRP for both E. coli and B. subtilis (Figure 4B), consistent with our previous data [7]. These results suggest that depletion of thiols is required for bactericidal activity of PGRPs. We next tested whether depletion of thiols was sufficient for bacterial killing. Diamide, at the concentration that induces similar depletion of thiols as PGRP (Figure 4A), was bacteriostatic, but not bactericidal (Figure 4C). Thus, this level of thiol depletion is not sufficient for bacterial killing. Glutathione and bacillithiol are the major low molecular weight thiols in E. coli and B. subtilis, respectively, that protect against oxidative and thiol stress [21]–[23]. Accordingly, glutathione-deficient ΔgshA E. coli and bacillithiol-deficient ΔbshC B. subtilis mutants had reduced total thiols by ∼45% and ∼30%, respectively (Figure S4). Also, thiol depletion by PGRP or diamide was less efficient in ΔgshA and ΔbshC mutants (79% and 74% depletion) than in WT bacteria (97% and 90% depletion) (Figure S4), suggesting that glutathione and bacillithiol are major targets of PGRP-induced thiol depletion in E. coli and B. subtilis, respectively. However, ΔgshA E. coli and ΔbshC B. subtilis were only somewhat more sensitive to PGRP and diamide than WT strains (Figure 4C), which indicates that these thiols play a modest role in protecting against PGRP and that other cellular thiols in these mutants are still able to maintain nearly sufficient reducing environment in the cytoplasm. Altogether, these results suggest that although thiol depletion likely contributes to bacterial killing, by itself it is not sufficient for strong bactericidal activity. PGRP treatment highly induced genes for detoxification and efflux of Cu, Zn, and other metals (Figure 1 and Tables 1, S1 and S2). We therefore tested whether treatment with PGRP increased intracellular concentrations of free Zn and Cu (also known as “labile” Zn and Cu, because no metal is truly free in cellular context). Indeed, PGRP induced a large increase in intracellular free (labile) Zn2+ in both E. coli and B. subtilis, based on 60- to 100-fold increase in fluorescence of Zn2+-specific membrane permeable Zynpyr-1 probe, measured by flow cytometry (Figures 5A, 5B and S5A). This increase in Zn2+ was significant at 30 and 60 min (data not shown) and was maximal at 2 hrs (Figures 5A and S5A). Detection of intracellular Zn2+ was completely suppressed by the membrane permeable Zn (II) chelator, TPEN [24] (Figure S5A). PGRP also induced a large increase in intracellular free (labile) Cu+ in B. subtilis, but not in E. coli, based on 20-fold increase in fluorescence of Cu+-specific membrane permeable CF4 probe, measured by flow cytometry after 2-hr exposure to PGRP (Figures 5A and S5B). Paraquat and diamide, used at the concentrations that caused similar increase in H2O2 or depletion of thiols as PGRP, did not induce any increases in intracellular free Zn2+ or Cu+ (Figure 5A). These results suggest that PGRP-induced increases in intracellular H2O2 or depletion of thiols are not responsible (or at least not sufficient) for the PGRP-induced increases in intracellular Zn2+ and Cu+. The slower kinetics of increase in Zn2+ and Cu+ than accumulation of H2O2 and depletion of thiols may be related to slower kinetics of transport of exogenous metals into the cell. Thus, these results further suggest that these three effects of PGRPs (oxidative, thiol and metal stress) are independent and do not induce each other. Antibiotics induced different patterns of changes in intracellular metals than PGRP-induced pattern. Gentamicin treatment led to large increases of both intracellular Zn2+ and Cu+ in B. subtilis, and low, but still significant, increase of Zn2+ and a moderate increase of Cu+ in E. coli. Ciprofloxacin, used here as a known positive control for induction of intracellular Cu+ in E. coli [25], caused high increase in Cu+ in both E. coli and B. subtilis, but did not lead to increased Zn2+ levels (Figures 5A and S5). Motivated by the high induction of genes for detoxification of both Cu and Zn (Figure 1 and Tables 1, S1 and S2), and the observed increase in intracellular metal concentrations in both E. coli and B. subtilis (Figures 5A and S5), we next tested whether Zn2+ and Cu+ were required for bactericidal activity of PGRPs. Indeed, chelating Zn2+ with TPEN completely inhibited the bactericidal activity of PGRP in both E. coli and B. subtilis (Figure 5C). Chelating Cu+ with Cu (I) chelator bathocuprione sulfonate (BCS) [26] also completely inhibited bactericidal activity of PGRP in both E. coli and B. subtilis (Figure 5C). These effects were selective for PGRP, because TPEN and BCS did not inhibit killing by a bactericidal antibiotic, gentamicin, and BCS even enhanced gentamicin killing at 1 hr (Figure 5C), consistent with the recent report of Cu+-mediated induction of antibiotic resistance regulator in E. coli [25]. The results with metal chelators, however, need to be interpreted with caution, because chelators are not 100% specific and may chelate to some extent other metals. This could explain the inhibition of PGRP-induced E. coli killing by BCS (Figure 5C), when there was no significant PGRP-induced increase in intracellular Cu+ in E. coli (Figure 5A), because although BCS is a Cu+ chelator [26], it can also form dimers with Cu2+ and possibly with other divalent metals and chelate them [27]. Our current results are also consistent with our previous data showing that chelating Zn2+ with EGTA (whose log stability constant for Zn2+ is 12. 9) inhibits bactericidal activity of PGRPs, and that 5 µM Zn2+ is required for PGRP killing [5]. Our previous results also show that chelating Fe2+ with dipyridyl inhibits bactericidal activity of PGRPs [7]. Cu2+ and Zn2+ at low physiologic concentrations were only bacteriostatic, but not bactericidal (Figure 6), which indicates that at these concentrations Cu2+ and Zn2+ are not sufficient for bacterial killing. To further determine which metal ions are the most critical for PGRP-induced killing, we then compared the sensitivity to PGRP and metal killing of WT bacteria and their mutants deficient in various metal efflux and detoxification systems. We show that both E. coli ΔzntAΔzitB mutant, deficient in two Zn2+ efflux systems [28], and B. subtilis ΔczcD mutants deficient in the Zn2+, Cu2+, Co2+, and Ni2+ efflux system [29] were substantially more sensitive to PGRP-induced killing than WT bacteria (Figure 5D). Similarly, ΔzntAΔzitB mutant was substantially more sensitive to killing by extracellular Zn2+ than WT bacteria (Figure S6A). E. coli and B. subtilis mutants deficient in Cu efflux and detoxification systems (E. coli ΔcopAΔcueOΔcusCFBA and B. subtilis ΔcadA and ΔcopZA) were not more sensitive to PGRP-induced killing than WT bacteria (Figure S6C), and the E. coli ΔcopAΔcueOΔcusCFBA mutant was even more resistant to PGRP killing. Similarly, E. coli ΔcopAΔcueOΔcusCFBA mutant was also more resistant to killing by extracellular Cu2+ than WT bacteria (Figure S6A), perhaps because increased intracellular Cu level protects E. coli from oxidative Fe toxicity [30], and only at higher concentrations Cu becomes bactericidal. B. subtilis ΔcopZA mutant had similar sensitivity to killing by extracellular Cu2+ as WT bacteria, whereas B. subtilis ΔczcDΔcadA mutant was more sensitive to killing by extracellular Cu2+ than WT bacteria (Figure S6B). Higher sensitivity of Zn efflux-deficient than Cu efflux-deficient mutants to PGRP is consistent with high increase of intracellular Zn2+ in both PGRP-treated bacteria. Altogether, these results indicate that Zn2+ and Cu+ are required for bactericidal activity of PGRPs, and that Zn2+ is more important than Cu+ for this bactericidal activity, especially in E. coli. Our results also indicate that these metals are not required for bactericidal activity of antibiotics. Indeed, PGRPs have the same bactericidal activity towards antibiotic-sensitive bacteria and clinical isolates resistant to multiple antibiotics (Figure S7). We next tested the hypothesis that production of ROS, depletion of thiols, and metal toxicity have a synergistic bactericidal effect, because these three stress responses were all induced in PGRP-treated bacteria and each was required, but not individually sufficient, for bacterial killing. To induce intracellular ROS production we used paraquat, which is reduced by Complex I to radical cations, which react with O2 to generate O2−, which then generate H2O2 and then OH• [14], [18]. To induce thiol stress, we used diamide, which directly depletes intracellular thiols by inducing formation of disulfide bonds and S-thiolations (which are disulfide bonds between proteins and low molecular weight thiols, such as glutathione, bacillithiol, and free cysteine) [14], [20], [31]. To induce metal toxicity, we used exogenous Zn2+, or Cu2+ (which is transported into the cell and reduced to more toxic Cu+), or arsenite (AsO2−). Indeed, treatment of E. coli or B. subtilis with the doses of paraquat that induce the amounts of H2O2 comparable with the amounts of H2O2 induced by PGRP were not bactericidal (Figure 6). Also, treatment of E. coli or B. subtilis with the doses of diamide that deplete thiols to a comparable extent as PGRPs were not bactericidal, and low concentrations of Zn2+, Cu2+, or As (AsO2−) by themselves were also not bactericidal (Figure 6). Moreover, the combination of any two of these stresses was also not bactericidal (except for a combination of paraquat plus Zn2+ or Cu2+, which had low killing activity for B. subtilis). However, when all three stress conditions were simultaneously imposed, the resulting combination was strongly bactericidal for both E. coli and B. subtilis, although Zn2+ was less efficient in E. coli than in B. subtilis (Figure 6). These results validate our hypothesis and show that ROS production, thiol depletion, and metal toxicity act synergistically to kill bacteria. To further verify that oxidative, thiol, and metal stress are responsible for the bactericidal activity of PGRPs, we abolished bactericidal activity of PGRP by de-glycosylation, which we previously showed to be required for bactericidal activity of PGRPs for both Gram-positive and Gram-negative bacteria [4], [5]. De-glycosylation abolished 90–95% of the ability of PGRP to induce (i) intracellular production of H2O2 (Figure 7A), (ii) depletion of cellular thiols (Figure 7B), and (iii) increases in intracellular Zn2+ (Figure 7C) in both E. coli and B. subtilis. These results further validate the requirement of oxidative, thiol, and metal stress for the bactericidal activity of PGRPs. Analysis of the global transcriptional responses of both E. coli and B. subtilis to PGRP revealed stress responses involving increased production of H2O2, depletion of thiols, and increases in intracellular Zn2+ and Cu+, which were also verified by direct measurements. Using selective chemical treatments (paraquat to generate ROS, diamide to oxidize thiols, and exogenous metal ions) and specific inhibitors, we demonstrated that ROS production, thiol depletion, and increased intracellular Zn2+ or Cu+ are all required, but individually are not sufficient, for bacterial killing, and that combined action of oxidative, thiol, and metal stress kills bacteria. PGRP treatment induced oxidative stress through rapid induction of H2O2 production. Oxidative stress results from excessive production of ROS (O2−, H2O2, and HO•). Both O2− and H2O2 oxidize solvent-exposed [4Fe-4S] enzyme clusters, causing release of Fe and cluster collapse to inactive [3Fe-4S]+. O2− and H2O2 also inactivate mononuclear iron enzymes by oxidizing Fe-coordinating cysteines or by replacing Fe2+ with Zn2+ [16], [21], [32]–[34]. Moreover, H2O2 reacts with Fe2+ to generate HO• via Fenton reaction. HO• is the most reactive and most toxic ROS and it irreversibly damages DNA, proteins, and other organic molecules [14], [17]. PGRP treatment also depleted over 90% of cellular thiols. Thiol stress results from oxidation of thiols, which maintain the redox state in the cells and protect from oxidative damage. Oxidative and thiol stress not only directly damage cells, but also release Fe from proteins, increase intracellular concentration of Zn and Cu, and increase toxicity of most metals [19], [21], [35]–[37]. Thiols bind free metal ions and protect cells from metal toxicity [38], and for this reason thiol stress induces the same genes for metal detoxification and protein refolding and repair [19], [20], [31] as the genes induced by PGRP (Tables 1, S1, and S2). PGRP treatment also induced a drastic increase in intracellular free (labile) Zn2+ in both E. coli and B. subtilis and intracellular free (labile) Cu+ in B. subtilis (but not E. coli), which is the likely reason for increased expression of metal detoxification and efflux genes. These increases in free metals are required for PGRP toxicity, because chelating intracellular Zn2+ with TPEN (Figure 5C) or extracellular Zn2+ with EGTA [5], or chelating Cu+ with BCS (Figure 5C) or Fe2+ with dipyridyl [7] also inhibits bacterial killing by PGRP. Zn2+ seems the most important for PGRP killing, as revealed by the highest sensitivity of Zn2+ efflux mutants to PGRP killing (Figure 5D). However, the increased concentrations of metals alone that are induced by PGRP are not sufficient for bacterial killing. The origins of metal toxicity are complex. Zn, a redox-inert metal, is more abundant in the cytosol than Cu and at low concentrations it may protect bacteria from oxidative and thiol stress, likely by binding to thiols and preventing their further oxidation [39]. However, high levels of Zn are toxic and up-regulate the expression of genes for Zn efflux (zntA in E. coli and czcD and cadA in B. subtilis, also observed in our arrays). Zn toxicity, similar to Cu, results in part from inactivation of solvent-exposed Fe-S clusters; and although this activity of Zn2+ is lower than Cu+ [40], it is likely compensated by higher concentrations of Zn2+ than Cu+. In oxidative stress, Zn2+ also inactivates mononuclear enzymes by replacing Fe2+ in their active sites [32]. Cu is toxic because it causes loss of Fe from solvent-exposed Fe-S clusters, which inactivates enzymes, and also because this release of Fe makes it available for enhanced production of HO• via Fenton reaction [14], [21], [35]–[37], [41]–[44]. Cu also causes thiol oxidation and sulfhydryl depletion, which contribute to thiol stress and protein damage [21], [35], [37], [42]. Fe toxicity results primarily from generation of HO•, which damages DNA, proteins, and lipids [14], [29]. HO• is induced by PGRPs and chelating intracellular Fe with dipyridyl inhibits both HO• production and PGRP killing [7]. Many of the genes induced by PGRPs reflect direct or indirect bacterial responses to the resulting oxidative, thiol, and metal stress. The genes for repair of damaged proteins and DNA and Ihf-regulated genes (which help to maintain DNA architecture) are induced because ROS oxidize proteins and nucleic acids, because oxidation of thiols damages proteins, and because increased concentrations of intracellular metals also damages proteins [14], [16], [17], [19], [21], [33], [34], [41]–[44]. Genes for transition to fermentation and anaerobic growth (e. g. , members of Fnr regulon in E. coli) are a likely attempt to reduce the use of oxygen to limit further production of ROS. Genes for energy generation are induced because of possible oxidative damage to respiratory chain enzymes and because a decrease in membrane potential [7] may cause a decrease in ATP production by membrane potential-driven ATP synthase [45], [46]. This is also the likely reason why bacteria down-regulate genes for high energy-requiring non-essential processes, such as motility, which are controlled by CpxRA [8], one of the regulators of envelope stress response activated by PGRP [7]. The genes for methionine and histidine synthesis may be induced for several reasons. These amino acids are essential metal-binding components abundant in metal detoxification proteins, e. g. , methionine shuttle is used for Cu efflux and histidine is used for coordination of metals in metal detoxification proteins, such as CusA and CopA Cu efflux and AraA and ArsD As efflux transporters [47]–[49]. Also, histidine shares biosynthetic intermediates with nucleotides, whose synthesis is needed to repair damaged DNA. Moreover, likely oxidation of the thiol group in homocysteine may deplete this methionine biosynthesis intermediate. Methionine is also needed for initiation of translation and DNA replication, and methionine synthase is highly sensitive to thiol stress [50]. The genes for Fe-S cluster assembly (isc in E. coli) are likely induced due to the damage to Fe-S clusters by oxidative, thiol, and metal stress, and most likely Cu+-induced release of Fe2+ from Fe-S clusters. Cu+ also damages Isc proteins, which may further contribute to the induction of isc genes. Damage to DNA could be either direct by Cu+, or more likely by Cu+-induced release of Fe2+ from Fe-S clusters and Fe-driven enhancement of HO• production from H2O2 [21]. This mechanism is supported by the ability to inhibit PGRP killing by chelating either Fe2+ with dipyridyl [7] or Cu+ with BCS (Figure 5C). Concurrently bacteria down-regulate the expression of genes for Fe uptake, which also suggests an increase in cytoplasmic free Fe2+, likely due to release of Fe2+ from Fe-S clusters, caused by oxidative and thiol stress and Cu+. Down-regulation of Fe uptake is controlled by the envelope stress response regulator CpxRA, which is activated by PGRPs [7], and by increased Cu and Zn [8], [51]–[53]. How do PGRPs induce oxidative, thiol, and metal stress in bacteria? PGRPs have a specific peptidoglycan-binding grove that binds disaccharide-pentapeptide fragment of peptidoglycan [2], [3], [54], [55]. However, this PGRP-binding site on peptidoglycan is not easily accessible on the surface of Gram-positive bacteria, because of extensive peptidoglycan cross-linking and its substitution with polysaccharides and proteins. Thus, in Gram-positive bacteria PGRPs preferentially bind to the separation sites of the newly formed daughter cells, created by dedicated peptidoglycan-lytic endopeptidases, which separate daughter cells after cell division. We assume that these cell-separating endopeptidases expose PGRP-binding muramyl peptides, because PGRP bound to bacteria co-localizes with cell-separating endopeptidases and PGRPs do not bind to other regions of the cell wall with highly cross-linked peptidoglycan [7]. This localization is necessary for bacterial killing, because mutants that lack these endopeptidases and do not separate after cell division (ΔlytEΔlytF B. subtilis) do not bind PGRPs and are not killed by PGRPs [7]. In Gram-negative bacteria, PGRPs bind uniformly to the entire outer membrane [7], which is composed of lipopolysaccharide (LPS) and covers a thin peptidoglycan layer. This is possible, because in addition to binding peptidoglycan, PGRPs also bind LPS using binding sites outside the peptidoglycan-binding groove [56], [57]. This binding to bacterial envelope is required for PGRP killing, because exogenous peptidoglycan or LPS inhibit PGRP killing of Gram-positive or Gram-negative bacteria, respectively, by blocking peptidoglycan or LPS binding sites on PGRP [4], [5], [56]. It is not known whether after binding to LPS in Gram-negative bacteria PGRPs also bind to peptidoglycan, located in the periplasmic space beneath the outer membrane. In both Gram-positive and Gram-negative bacteria, after binding to peptidoglycan or LPS, PGRPs do not enter the cytoplasm [7], but probably form oligomeric ribbon-like structures [2], [55] and induce envelope stress by activating stress response two component systems, CpxRA in E. coli and CssRS in B. subtlis, which are typically activated by misfolded or aggregated proteins exported from the cells [3], [7], [58]. This activation ultimately results in membrane depolarization, inhibition of all biosynthetic reactions, and cell death [7]. However, the exact initial mechanism through which PGRPs activate envelope stress response and oxidative, thiol, and metal stress is unknown, as this mechanism is also unknown for other envelope stressors [8], and is currently under investigation. Furthermore, based on induction of multiple stress response regulons by PGRP (Tables 1, and S1, S2, S3, S4) and on incomplete resistance of ΔcpxRA and ΔcssRS mutants to PGRP [7], it is likely that PGRPs activate other stress sensors that induce these multiple stress responses. Other investigators previously proposed that oxidative stress is involved in killing of E. coli by antibiotics [9], [10]. However, recent results do not support this conclusion [12], [13] and are consistent with our results. Our data clearly indicate that the mechanisms of killing by PGRPs and antibiotics are different for the following reasons. (i) PGRPs kill bacteria resistant to multiple antibiotics (Figure S7) [4]. (ii) PGRP killing requires O2 and PGRPs do not kill anaerobically (Figure 2B), whereas many antibiotics kill both aerobically and anaerobically [12], [13]. (iii) PGRPs very strongly induce peroxide-responsive genes (e. g. the OxyR regulon in E. coli) indicating endogenous H2O2 production, but antibiotics do not (Tables S1 and S2) [12]. (iv) PGRPs strongly induce H2O2 production in bacteria (Figure 2A), but antibiotics do not [12]. (v) ΔrecA mutant is more sensitive than wild type strain to PGRPs (Figure 2B), but not to antibiotics [12]. (vi) PGRP-induced killing is inhibited by chelating Zn2+ or Cu+, whereas killing by antibiotics is not affected by chelating Zn2+ and is enhanced by chelating Cu+ (Figure 5C). These results are consistent with induction of the antibiotic resistance regulator MarR by CpxRA [59] and by Cu+ [25], which are induced by both PGRP and antibiotics. However, MarR confers resistance only to antibiotics [25], but not to PGRP. (vii) The patterns of gene expression induced in E. coli and B. subtilis by bactericidal concentrations of PGRP and by gentamicin are different: more than half of the top 100 genes strongly induced by PGRPs are not induced by gentamicin, e. g. , genes for oxidative stress, energy production, Fe-S cluster repair and assembly, Fe-S-containing enzymes (e. g. , edd), amino acid synthesis, and other stress responses. (viii) We could prevent bacterial killing by cell wall synthesis-inhibiting antibiotics, but not by PGRPs, using hyperosmotic medium [7], which should not happen if the main mechanism of killing by these antibiotics was due to oxidative stress and was the same as for PGRPs. (ix) Antibiotics selectively inhibit one biosynthetic reaction and other biosynthetic reactions are not inhibited for several hours until bacteria die, whereas exposure to PGRPs results in simultaneous and rapid inhibition of all biosynthetic reactions in bacteria [7]. PGRPs, bactericidal innate immunity proteins, by combining oxidative stress with thiol depletion and release of intracellular metals, have evolved a powerful antibacterial defense strategy. This strategy is consistent with recent evidence that phagocytic cells, upon phagocytosis of bacteria, in addition to oxidative killing, pump Cu and Zn into phagolysosomes to enhance bacterial killing [41]–[44], [60]. Indeed, the most abundant PGRP, PGLYRP1, is present in neutrophil, eosinophil, and macrophage granules [1], [56], [61]–[64], and other PGRPs (PGLYRP2, PGLYRP3, and PGLYRP4) are produced on the skin and mucous membranes, and in sweat, sebum, and saliva [1], [4], [5]. These body secretions also contain significant amounts of Cu and Zn [5], which is consistent with the requirement for Zn (Figure 4B) [5], Fe [7], and Cu (Figure 4B) for bactericidal activity of PGRPs. In response to PGRPs bacteria up-regulate expression of Cu and Zn exporters (CopA, ZntA, CadA, and CzcD). However, PGRPs defeat this bacterial Cu and Zn defense, because PGRP-induced oxidative, thiol, and metal stress likely damage respiratory chain enzymes and depolarize bacterial membranes [7], which likely reduces ATP production and proton motive force needed to drive bacterial Cu and Zn efflux. Furthermore, because Cu tolerance increases bacterial virulence [41]–[44], targeting Cu tolerance will both increase bacterial killing and decrease bacterial virulence, which should additionally improve host defense against infection. In vivo PGRPs are present at concentrations similar to the concentrations used in our experiments: PGLYRP1 is present in milk at 120 µg/ml [65] and in polymorphonuclear leukocytes' granules at 2. 9 mg/109 cells [64], PGLYRP2 is present in serum at 100 µg/ml [66], [67], and PGLYRP3 and PGLYRP4 are secreted on mucous membranes, likely reaching similar local concentrations [1], [4]. In this study we investigated the mechanism of bactericidal activity of PGRPs in vitro, but the following evidence indicates that PGRPs also have antibacterial activity in vivo: (i) local application of PGRPs into upper respiratory tract protects mice against lung infection [4], [58]; (ii) Pglyrp1−/− mice are more sensitive to some infections than wild type mice [62]; (iii) neutrophils from Pglyrp1−/− mice are less efficient in bacterial killing than neutrophils from wild type mice [62]; (iv) PGRPs protect zebrafish embryos from bacterial infections [68]; (v) PGRPs are required for maintenance of normal intestinal microbiome in mice [69]; and (vi) PGRPs also have several anti-microbial and microbiome-regulating functions in invertebrates [3]. Our results indicate that PGRPs have bactericidal activity in an aerobic environment, which is consistent with the highest expression of PGRPs in phagocytic cells and on the skin and mucous membranes, especially in the mouth, throat, esophagus, and salivary glands [1]–[4], [56], [61]–[64], [69]. Lower PGRP expression in the stomach and small and large intestine is again consistent with their bactericidal activity in an aerobic environment, although anaerobically PGRPs are still bacteriostatic. Bactericidal activity of PGRPs both in vitro and in vivo is enhanced by antimicrobial peptides [5], [58], also expressed in phagocytic cells and on mucous membranes and skin, which likely further strengthens antibacterial defenses of the host. In conclusion, innate immunity proteins, PGRPs, induce oxidative, thiol, and metal stress in E. coli and B. subtilis, which act synergistically to kill bacteria. Because this bactericidal mechanism differs from killing by antibiotics and because PGRPs kill antibiotic-resistant bacteria, synergistic targeting of oxidative, thiol, and metal stress can be used for the development of new approaches to treatment of antibiotic resistant bacteria. Bacterial strains are listed in Table S5. Disruption of Bacillus genes was achieved by transformation with PCR products to amplify DNA fragments flanking each target gene and an intervening antibiotic cassette as previously described [70]. Human PGRPs (PGLYRP3, PGLYRP4, and PGLYRP3: PGLYRP4 heterodimer) were expressed in S2 cells and purified as previously described [4], [5] in a buffer containing 10 mM TRIS (pH 7. 6), with 150 mM NaCl, 10 µM ZnSO4, and 10% glycerol. The experiments were done using PGLYRP3, PGLYRP4, and/or PGLYRP3: PGLYRP4 (as indicated in Figure legends and Table footnotes), and all key experiments were performed with at least two PGRPs with similar results. Note that when expressed individually, PGLYRP3 and PGLYRP4 form disulfide-linked homodimers, and when co-expressed in the same cells, they form disulfide-linked PGLYRP3: PGLYRP4 heterodimers [4]. For some experiments PGRP was de-glycosylated by treatment with 0. 67 units of N-glycosidase/µg PGRP (PNGase F from Elizabethkingia miricola, Sigma) for 2 hr at 37°C, and we verified that this treatment abolished PGRP' s bactericidal activity for E. coli and B. subtilis, as previously described [4], [5]. For non-de-glycosylated PGRP in these experiments, PGRP was similarly incubated in the same buffer, but without PNGase. Purified bovine serum albumin (BSA, Sigma) was used as a negative control, and key experiments were repeated with recombinant mouse serum albumin (rMSA) as an additional control, which was cloned, expressed, and purified by the same methods as PGRPs, as described [7], with similar results, as indicated in figure legends. Paraquat (methyl viologen) was from Acros Organics, Zinpyr-1 and TPEN were from Santa Cruz. Bathocuprione disulfonate (BCS), CCCP (carbonyl cyanide 3-chlorophenyl-hydrazone), ciprofloxacin, diamide, gentamicin, and other reagents were from Sigma-Aldrich, unless otherwise indicated. Arsenite (AsO2−) was prepared fresh from arsenic trioxide at pH 8. 2; CuSO4 was used as Cu2+, and ZnSO4 as Zn2+. Overnight bacterial cultures were diluted 1∶100 in LB, grown aerobically with 250 rpm shaking to OD660 = 0. 1–0. 3, suspended in fresh warm medium (E. coli MG1655 at OD660 = 0. 3 or B. subtilis 168 at OD660 = 0. 1), and incubated aerobically with 100 µg/ml albumin (control), or 100 µg/ml PGRP (human recombinant PGLYRP4), or 5 µg/ml gentamicin for 30 min, or with 800 µM CCCP for 15 min, in 2 ml of 5 mM TRIS (pH 7. 6) with 150 mM NaCl, 5 µM ZnSO4, with addition of 2% of 100% LB (E. coli), or in 1 ml of TRIS-Schaeffer medium with 0. 05% NH4Cl, 5 µM ZnSO4,0. 2% glucose, with addition of 2% of 100% LB (B. subtilis) at 37°C with 250 rpm shaking (these optimum incubation times and concentrations for induction of stress response genes were determined in preliminary experiments using qRT-PCR). Because 5 µM Zn2+ is required for bactericidal activity of PGRP and corresponds to the average concentration of Zn2+ found in saliva, sweat, and other body fluids, where PGRPs are present [5], we confirmed that Zn2+ is not depleted or increased by additions of our proteins and bacteria, by measuring the concentration of free Zn2+ in our incubation mixtures at the initiation of our experiments (time 0) using Zn2+-specific probe, Zinpyr-1, and fluorescence spectroscopy (with Molecular Devices Gemini EM Spectrofluorometer). Our incubation mixtures containing 100 µg/ml of either PGRP (PGLYRP3 or PGLYRP4) or recombinant mouse albumin, and with or without addition of bacteria, all contained similar amounts of free Zn2+ (4. 1–4. 4 µM). Moreover, substituting the addition of 5 µM free Zn2+ with the addition of 25 µM Zn2+ plus 20 µM EDTA (a divalent cation chelator with high affinity for Zn2+, log dissociation constant 16. 6) yielded the same concentrations of free Zn2+ as in our reaction mixture without EDTA, measured by Zinpyr-1 fluorescence. Based on these results we concluded that addition of our control protein or PGRPs and/or bacteria does not substantially change the free Zn2+ concentration in our experiments. To obtain RNA from each culture, bacteria were harvested and RNA was extracted using Ambion RiboPure-bacteria RNA extraction kit according to the manufacturer' s instructions. For B. subtilis, before RNA extraction, bacteria were disrupted by shaking with Zirconia beads. cDNA was synthesized with random hexamer primers, fragmented, labeled with terminal transferase and biotin, and hybridized to whole genome Affymetrix E. coli Genome 2. 0 Array GPL3154 or custom whole genome Affymetrix 900513 GeneChip B. subtilis Genome Array using Affymetrix Hybridization Oven 640 and Affymetrix GeneChip Fluidics Station 450 and protocols provided by Affymetrix GeneChip Technical Manual. Scanning and data extraction were done using Affymetrix GeneChip Scanner 3000 and protocols provided by Affymetrix GeneChip Technical Manual. cDNA synthesis, labeling, hybridization, and scanning were performed at the Genomic and RNA Profiling Core facility, Baylor College of Medicine, Houston, TX. The entire experiment was repeated 3 times both for E. coli and B. subtilis. Hybridization intensity data signals were analyzed, normalized, and corrected for batch effect using Affymetrix GeneChip Command Console Software. Signal average, noise average, scaling factor, % present, and % absent were calculated for each probe. From this analysis, for E. coli, signal intensity of ≥39 was calculated as reliable expression, and using this cutoff, 5,531 probes were classified as present out of total 10,208 probes on the array. For B. subtilis, signal intensity of ≥78 was calculated as reliable expression, and using this cutoff, 3,355 probes were classified as present out of total 5,039 probes on the array. The probes were classified as expressed when at least one experiment in one group showed the signal intensity ≥39 for E. coli or ≥78 for B. subtilis. Signal intensities from 3 experiments were used to calculate fold increases or decreases in gene expression between treated and control groups, with signal intensity of 39 for E. coli or 78 for B. subtilis used as a minimum intensity (i. e. , for these calculations all signal intensities of <39 for E. coli and <78 for B. subtilis were replaced with 39 or 78, respectively). The fold changes in gene expression were calculated using the formula: intensity in treated group/geometric mean of intensity in control (albumin) groups, and reported as means ± SEM in Tables S1, S2, S3, S4. This method yields conservative fold increases or decreases in gene expression and avoids erroneous and unrealistically large fold changes in gene expression, which would have been obtained if signal intensities below the reliable expression thresholds were used for these calculations. Transformed Ln (signal intensity) values were used for direct statistical comparisons of expression signals between treated and control (albumin) groups. We deposited all whole genome expression arrays data in NCBI GEO (accession numbers GSE44211 and GSE44212). We also compared by hierarchical cluster analysis [71] our whole genome expression results with published data on E. coli exposed to H2O2 [72] and Zn (NCBI GEO GSE26187), and on B. subtilis exposed to vancomycin [73], diamide [19], H2O2 [74], and Zn [29]. The functions of genes, gene operons, and gene regulons were annotated using the following web databases: for E. coli: PrFEcT (http: //www. prfect. org/index. php? option=com_content&view=frontpage&Itemid=1), GenExpDB (http: //www. prfect. org/index. php? option=com_wrapper&view=wrapper&Itemid=38), and RegulonDB (http: //regulondb. ccg. unam. mx/index. jsp); and for B. subtilis: SubtilisWiki (http: //subtiliswiki. net/wiki/index. php/Main_Page) and SubtiWiki (http: //subtiwiki. uni-goettingen. de/). E. coli or B. subtilis (300 µl each) were incubated with albumin (control), PGRP, gentamicin, or CCCP, and RNA was extracted as described above for gene expression arrays. The amounts of mRNA were measured using quantitative reverse transcription real-time PCR (qRT-PCR) as previously described [7], [63]. cDNA was synthesized from 100 ng of RNA using RT2 PCR Array First Strand Kit (Qiagen/SA Biosciences). Gene expression was quantified by qRT-PCR using the ABI 7000 Sequence Detection System with 1 cycle 10-min at 95°C and 40 cycles 15 sec at 95°C and 1 min at 60°C using Qiagen/SA Biosciences SYBR Green Master Mix and the gene-specific primers (listed in Table S6) or common primers for 16S rRNA from all Eubacteria (ACTCCTACGGGAGGCAGCAGT and ATTACCGCGGCTGCTGGC) as a housekeeping gene. For each gene, ΔCt was calculated followed by normalization to the housekeeping gene, followed by calculation of ΔΔCt for each gene: ΔΔCt = ΔCt1−ΔCt2, where ΔCt1 is the PGRP- or gentamicin- or CCCP-treated bacteria and ΔCt2 is albumin-treated bacteria. This calculation gives the fold increase in expression of each gene in PGRP- or gentamicin- or CCCP-treated bacteria versus albumin-treated bacteria. The entire experiment was repeated 3 times both for E. coli and B. subtilis. To measure production of H2O2, Hpx− strains ΔkatGΔkatEΔahpCF E. coli and ΔkatAΔahpCF B. subtilis were used, which allow accumulation and measurement of H2O2 [12], [15]–[17]. Bacteria (50 µl) were incubated as for gene expression arrays with albumin (control), PGRP, paraquat, or diamide (at concentrations given in Results), for 15–120 min (15 min was the optimum time for the highest induction of H2O2, determined in preliminary experiments), and total amount of H2O2 was determined using fluorescent Amplex Red Hydrogen Peroxide/Peroxidase Assay Kit (InVitrogen/Molecular Probes) according to the manufacturer' s instructions. To measure depletion of thiols, 50 µl of E. coli MG1655 or B. subtilis 168 were incubated as above for H2O2 production for 30–120 min (30 min was the optimum time for depletion of thiols, determined in preliminary experiments), and the total amount of reduced thiols was determined using fluorescent Measure-iT Thiol Assay Kit (InVitrogen/Molecular Probes) [35] according to the manufacturer' s instructions. For bactericidal assays, overnight bacterial cultures were diluted 1∶100 in LB, grown aerobically at 37°C with 250 rpm shaking to OD660 = 0. 1, suspended at ∼2–4×106 bacteria/ml in 50 µl of fresh warm medium, for E. coli in 5 mM TRIS (pH 7. 6) with 150 mM NaCl, 5 µM ZnSO4,5% glycerol, with addition of 2% of 100% LB [5] or for B. subtilis in TRIS-Schaeffer medium with 0. 05% NH4Cl, 5 µM ZnSO4,5% glycerol, 0. 2% glucose, with addition of 2% of 100% LB [4], [7], incubated at 37°C aerobically with 250 rpm shaking, and the numbers of bacteria were determined by colony counts [4]. Assays on killing under anaerobic conditions were done in the same medium in complete absence of oxygen (90% N2,5% H2,5% CO2) for E. coli or under microaerophilic conditions (1% O2) for B. subtilis (because B. subtilis grows very poorly under strict anaerobic conditions) in Anaerobe Systems AS-580 Anaerobic Chamber for growing the cultures before the assay, during the killing assay, and during incubation of plates for colony counts. Bactericidal activity is defined as an at least 100-fold decrease in the number of inoculated bacteria in 4 hrs. E. coli MG1655 or B. subtilis 168 were prepared and incubated aerobically as for bactericidal assays at ∼2×107 bacteria/ml for 0. 5,1, 2, or 4 hrs with albumin, PGRP, paraquat, diamide, gentamicin, ciprofloxacin, Zn2+, or Cu2+ (at concentrations indicated in Results), and without or with 100 µM TPEN for Zn2+ assay or with 8 µM (E. coli) or 2 µM (B. subtilis) CuSO4 for Cu+ assay. Bacteria were washed, incubated with Zinpyr-1 (dissolved in DMSO with 20% Pluronic F-127,5 µM final concentration) for 15 min at 37°C for free (labile) intracellular Zn2+ determination [24], or with Copperfluor-4 (CF4, dissolved in DMSO, 2 µM final concentration) for 15 min at 37°C for Cu+ determination. CF4 is a 4th generation membrane permeable fluorescent probe that specifically detects free (labile) intracellular Cu+, with improved fluorescent signal, compared with previous CS1–CS3 probes [75]. Bacteria were washed and analyzed by flow cytometry using MACSQuant (Miltenyi) flow cytometer and FITC excitation and emission settings. The maximum increases in Zinpyr-1 and CF4 fluorescence were seen after 2 hrs of incubation and these results are reported as mean fluorescence intensity (MFI) ± SEM. Representative dot plots are also shown in some figures. Quantitative results are presented as means ± SEM, with statistical significance of the differences between groups determined by the two-sample one-tailed Student' s t-test using Microsoft Excel; P≤0. 05 was considered significant. The n and P values are indicated in the figures and tables. Some gene expression results are presented as heat maps generated using Java TreeView. For microarray data statistical significance of differences in gene expression was also analyzed by calculating P values using two-sample two-tailed Student' s t-test, followed by calculation of π0 (λ) and then FDR (false discovery rate) q values, with significance threshold of q≤0. 05, as described [76].

Bacterial infections are still a major cause of morbidity and mortality because of increasing antibiotic resistance. New targets for developing new approaches to antibacterial therapy are needed, because discovering new or improving current antibiotics have become increasingly difficult. One such approach is developing new antibacterial agents based on the antibacterial mechanisms of bactericidal innate immunity proteins, such as human peptidoglycan recognition proteins (PGRPs). Thus, our aim was to determine how PGRPs kill bacteria. We previously proposed that PGRPs kill bacteria by inducing toxic oxygen by-products (“reactive oxygen species”, ROS) in bacteria. It was also previously proposed, but recently refuted, that bactericidal antibiotics kill bacteria by inducing ROS production in bacteria. These findings prompted us to evaluate in greater detail the mechanism of PGRP-induced bacterial killing, including the role of ROS in PGRP killing. We show here that PGRPs kill bacteria through synergistic induction of ROS, depletion of thiols, and increasing intracellular concentration of metals, which are all required, but individually not sufficient for bacterial killing. Our results reveal a novel bactericidal mechanism of innate immunity proteins, which differs from killing by antibiotics and offers alternative targets for developing new antibacterial therapies for antibiotic-resistant bacteria.

Abstract Introduction Results Discussion Materials and Methods

bacteriology innate immune system medical microbiology microbial pathogens biology and life sciences immunology microbiology bacterial biochemistry bacterial pathogens immune system

2014

Peptidoglycan Recognition Proteins Kill Bacteria by Inducing Oxidative, Thiol, and Metal Stress

Heterologous transinfection with the endosymbiotic bacterium Wolbachia has been shown previously to induce pathogen interference phenotypes in mosquito hosts. Here we examine an artificially infected strain of Aedes polynesiensis, the primary vector of Wuchereria bancrofti, which is the causative agent of Lymphatic filariasis (LF) throughout much of the South Pacific. Embryonic microinjection was used to transfer the wAlbB infection from Aedes albopictus into an aposymbiotic strain of Ae. polynesiensis. The resulting strain (designated “MTB”) experiences a stable artificial infection with high maternal inheritance. Reciprocal crosses of MTB with naturally infected wild-type Ae. polynesiensis demonstrate strong bidirectional incompatibility. Levels of reactive oxygen species (ROS) in the MTB strain differ significantly relative to that of the wild-type, indicating an impaired ability to regulate oxidative stress. Following a challenge with Brugia pahangi, the number of filarial worms achieving the infective stage is significantly reduced in MTB as compared to the naturally infected and aposymbiotic strains. Survivorship of MTB differed significantly from that of the wild-type, with an interactive effect between survivorship and blood feeding. The results demonstrate a direct correlation between decreased ROS levels and decreased survival of adult female Aedes polynesiensis. The results are discussed in relation to the interaction of Wolbachia with ROS production and antioxidant expression, iron homeostasis and the insect immune system. We discuss the potential applied use of the MTB strain for impacting Ae. polynesiensis populations and strategies for reducing LF incidence in the South Pacific. Lymphatic filariasis (LF) affects 120 million people globally and has been a leading cause of morbidity in South Pacific regions [1]. An ongoing global campaign to eliminate LF relies upon mass drug administration (MDA) strategies. However, because of inherent issues associated with MDA, such as efficacy of antifilarial drugs and public compliance with drug regimens, an integrated approach that targets the vector has been suggested for the successful control of LF in some regions, such as the South Pacific. Aedes polynesiensis is the primary vector of Wuchereria bancrofti, the filarial nematode that causes LF in the South Pacific [2]. This mosquito is naturally infected with Wolbachia, a maternally inherited endosymbiont that infects a broad range of invertebrates [3], [4]. Wolbachia infection in mosquitoes can induce cytoplasmic incompatibility (CI), a form of conditional sterility that results in early embryonic arrest when a Wolbachia infected male mates with an uninfected female or one harboring a different Wolbachia type [3]. Public health strategies under development are based upon manipulating Wolbachia induced CI in important mosquito species, either to suppress the population through releases of incompatible males or to harness CI as a gene-drive mechanism for spreading useful phenotypes, such as disease resistance, into a targeted population [5]–[7]. A recent proof of concept for this approach comes from a program in Australia that has successfully replaced an existing Ae. aegypti population with an artificially infected mosquito [8], [9]. Wolbachia infections, both natural and artificial, have been shown to interfere with pathogen development within the mosquito host. The presence of a naturally occurring Wolbachia infection in Drosophila protects flies from virus-induced mortality [10]. In mosquitoes, examples include the natural Wolbachia infection within Culex quinquefasciatus, which is associated with a significant reduction in West Nile virus dissemination and transmission rates [11]. Artificial Wolbachia infections have been observed to affect dengue, chikungunya, Plasmodium and filarial worms in Ae. aegypti [12], [13]. Formation of Plasmodium falciparum oocysts was inhibited in Anopheles gambiae that were somatically inoculated with Wolbachia [14]. Although the mechanism underlying pathogen interference is unknown, a possible explanation is the association between artificial Wolbachia infections and increased expression of key mosquito immune factors such as defensins, cecropins and Toll pathway genes [12], [13], [15]. In addition, recent studies have shown that artificial Wolbachia infections are associated with increased oxidative stress in the form of reactive oxygen species (ROS) [16], [17]. ROS are produced as the byproduct of aerobic metabolism [18] and can have detrimental effects on fecundity [19] and survival post blood meal [20]. However, a positive effect of high ROS levels is the inhibition of parasites within the mosquito host [21], [22]. Furthermore, elevated levels of ROS in Ae. aegypti are linked to the activation of the Toll immune pathway [17]. In this study, embryonic microinjection [23], [24] was used to introduce the wAlbB infection from Ae. albopictus into Ae. polynesiensis. Prior transfer of the wAlbB infection into Ae. aegypti induced strong CI in the resulting strain [25] and increased host viral resistance to dengue by increasing ROS levels and elevating expression of immune genes [15], [17]. In addition, the introduction of an artificial Wolbachia infection decreased filarial competence in Ae. aegypti [12]. We hypothesized that artificial introduction of the wAlbB infection into Ae. polynesiensis might facilitate a similar immunological response and reduce the intensity of filarial worm infection. Unlike other mosquito vector species, relatively little genomic information and molecular tools are available for Ae. polynesiensis, making examination of immune gene expression difficult. However, ROS measurement methods are a relatively robust indicator of immune system activation and have been applied to numerous species, including mosquito vectors [17], [22], [26], [27]. Here, we have compared ROS levels between the Ae. polynesiensis strains infected with different Wolbachia types. The results show an association between Wolbachia type and ROS levels. Comparisons of the Ae. polynesiensis strains show significant differences in their ability to support Brugia pahangi development. We discuss the results in relation to a possible interaction between Wolbachia infection type, ROS levels and filarial competency and the potential application to public health strategies targeting decreased LF incidence. The MTB strain of Ae. polynesiensis was generated by microinjecting embryos of the aposymbiotic APMT strain with cytoplasm from naturally superinfected Ae. albopictus embryos (HOU strain). The mosquito strains used in the injection experiment are listed in Table 1, and the outcome of one injection experiment is shown in Table 2. Of 134 injected APMT embryos, 17 G0 females and 12 G0 males survived to adulthood, seven and six of which were infected with Wolbachia, respectively. The seven PCR positive G0 females were screened for specific infection type (Table 3). A G0 female infected with both wAlbA and wAlbB was chosen for further selection. Her daughters (G1) had a 60% infection rate, with the majority of females single-infected with wAlbB only. PCR testing and selection of the subsequent generations were unable to sustain the superinfection. Thus, the resulting MTB strain is infected with wAlbB only. Using PCR-guided selection, infected females were continuously outcrossed with APMT males until G6. Beginning at G7, MTB females were mated with MTB males. Subsequent to G7, periodic testing of the MTB strain confirmed that the infection is stable and maternal inheritance rates remain at 100% (data not shown). Crosses were performed to examine for CI. The results demonstrate bidirectional incompatibly between APM and MTB. High egg hatch was observed in crosses between similar males and females (Table 4). In contrast, few eggs hatched in reciprocal crosses between the strain types. Incompatible crosses were significantly different from the controls (H = 12. 5, df = 3, p = 0. 0059). ROS levels can be significantly affected in mosquitoes that are artificially infected with Wolbachia [17]. To examine for a Wolbachia-mediated effect in Ae. polynesiensis, we compared ROS levels in young adult females of MTB, APM and APMT. In addition to examining females fed sucrose only, we provided females with a blood meal to examine for an effect of blood feeding on ROS levels, which has been observed in prior studies [22], [28]. The results (Figure 1) show that blood meal status had a significant effect on ROS levels, but only for the strains in which Wolbachia had been manipulated (i. e. , not in the naturally infected strain). Specifically, a model with strain and blood meal status as factors and ROS level as the variable was significant (GLM: χ2 = 38. 2, df = 5, p<0. 0001). Blood meal status was significant (GLM: χ2 = 27. 2, df = 1, p<0. 0001), while the overall strain effect was not significant (GLM: χ2 = 5. 12, df = 2, p = 0. 07). A significant interactive effect was observed for strain×blood meal status (GLM: χ2 = 19. 3, df = 2, p<0. 0001). Following a blood meal, ROS levels in the APM strain remained similar to those observed in sucrose fed females (ANOVA, F1,8 = 0. 11, p = 0. 75). However, significant decreases in ROS levels are observed for blood fed females of APMT (ANOVA, F1,8 = 19. 09, p<0. 05) and MTB (ANOVA, F1,8 = 35. 08, p<0. 001). Post hoc Tukey HSD tests determined that after blood feeding, APM had significantly higher levels of ROS than APMT (p<0. 05) and MTB (p<0. 05), the latter of which were equivalent (p = 0. 9). Prior studies have shown that changes in ROS levels can be detrimental to Plasmodium development in Anopheles gambiae [21], [22], [27] and that artificial Wolbachia infection can affect filarial worm development in Ae. aegypti [12]. Therefore, the number of infective stage filarial worms were compared following a Brugia-infected blood meal (Figure 2). MTB had significantly lower worm loads relative to both APM (χ2 = 53. 3, df = 1, p<0. 0001) and APMT (χ2 = 44. 2, df = 1, p<0. 0001). Equivalent worm loads were observed with APM and APMT (χ2 = 0. 52, df = 1, p = 0. 47). Observations during competency experiments suggested a difference in strain survivorship after feeding on Brugia-infected blood. Therefore, a formal experiment was conducted to compare survivorship. Significant differences were observed in survivorship between strains fed Brugia-infected blood (GLM: χ2 = 119. 6, df = 11, p<0. 0001, Figure 3). There were significant differences between replications (GLM: χ2 = 28. 5, df = 3, p<0. 0001) and strain (GLM: χ2 = 73. 79, df = 2, p<0. 0001). Despite the variation between replicates, the pattern between strains remained consistent (GLM: strain×replicate, χ2 = 6. 27, df = 6, p = 0. 39; Figure 3C). To examine for a role of the Brugia parasite in the different survivorship, the strains were compared when fed on blood either with or without Brugia (Figure 4). The results show that, in general, each of the strains experienced lower survivorship when fed Brugia-infected blood, relative to uninfected blood (GLM: χ2 = 14. 3, df = 1, p = 0. 0002). Similar to the pattern observed in the preceding experiment, APM was longer lived than APMT (GLM: χ2 = 9. 85, df = 1, p<0. 05), which was longer lived than MTB (GLM: χ2 = 6. 05, df = 1, p<0. 05). There was no significant interactive effect for blood meal type×strain (GLM: χ2 = 4. 5, df = 2, p = 0. 1). Comparing females fed either blood or sucrose only (Figure 5), the pattern was similar to that observed for ROS levels (Figure 1). The two artificially infected strains experienced significantly reduced survival following a blood meal. Specifically, APMT (GLM: χ2 = 6. 72, df = 1, p<0. 05) and MTB (GLM: χ2 = 16. 98, df = 1, p<0. 05) were longer lived when fed sucrose only. In contrast, APM females were longer lived following a blood meal (GLM: χ2 = 11. 92, df = 1, p<0. 05). Embryonic microinjection was used to transfer the wAlbB infection from Ae. albopictus to Ae. polynesiensis. PCR assays show that the infection is stable, with high maternal transmission. Overall mosquito survival after microinjection was high as compared to previous studies, which observed survival rates of less than 5% [24], [29]. Prior crossing experiments found that wPolA [30] and wAlbB [31] cause cytoplasmic incompatibility when crossed with different Wolbachia types. Consistent with expectations, crosses between APM and MTB were bidirectionally incompatible. Although superinfected cytoplasm was injected, only the wAlbB Wolbachia infection was established. The separation of superinfected Wolbachia types after microinjection is consistent with prior reports [31]–[34] and may result from different infection levels of wAlbB versus wAlbA within superinfected Ae. albopictus [35]. The differing ROS levels observed in the artificially infected Ae. polynesiensis strains is similar to that of previous reports, which have shown differing ROS levels resulting from transinfection with wAlbB Wolbachia infection in both adult female Ae. aegypti [17] and in an Ae. albopictus cell line [16]. Multiple cellular pathways, including iron metabolism and immunity, influence ROS levels in mosquitoes [17], [36]. For example, H2O2 can be catalyzed via the Fenton reaction, along with excess labile iron [36]. Prior studies have demonstrated Wolbachia-produced bacterioferritin can scavenge labile iron, with the potential for iron competition between Wolbachia and host [16], [37]–[44]. Unfortunately, testing of specific hypotheses is limited at present, since many of the genomic and molecular tools available for better studied mosquito species (e. g. , Ae. aegypti) are not yet available with Ae. polynesiensis. Our results provide additional motivation for developing such tools and methods. Particularly intriguing is the previously reported ability of Wolbachia to upregulate dual oxidase (DUOX), which may influence the observed variation in ROS levels. A key component of innate immunity, the DUOX transmembrane protein is involved in ROS generation [26], [45]–[47] and can be increased if an artificial Wolbachia infection is recognized as foreign by its mosquito host [17]. Imbibing a blood meal can cause significant oxidative stress due to heme released from the degradation of hemoglobin, which can have pro-oxidant and cytotoxic effects when not bound to regulatory proteins [48]. An ability to maintain homeostasis despite this massive influx of iron resulting from blood feeding is important to the evolutionary success of hematophagous insects [36]. In addition, the Wolbachia genome retains genes for heme biosynthesis [49]–[51], and naturally occurring Wolbachia infections have been shown to buffer iron flux in insects, allowing iron homeostasis despite large influxes and limiting the deleterious effect of labile iron [37], [38]. In the wild type APM strain, no overall variation in ROS levels were observed following a blood meal. In contrast, a significant decrease in ROS levels was observed in the artificially infected MTB and aposymbiotic APMT strains following a blood meal. The reintroduction of wAlbB did not restore MTB to the homeostasis phenotype observed in the wild-type APM strain, indicating that not all Wolbachia types are equivalent and that the wPolA infection in Ae. polynesiensis represents an evolved symbiosis. Whole mosquitoes were examined in this study, which may mask a tissue-specific effect. For example, mitochondrial generation of H2O2 was reduced in flight muscles after blood feeding in Ae. aegypti [52], and ROS levels were elevated in An. gambiae hemolymph following a blood meal [22]. In Ae. aegypti, blood feeding was associated with a significant decrease in ROS levels in midgut tissue through activation of a heme-mediated protein kinase C pathway [28]. While beyond the scope of our experimental design, our results encourage additional experiments designed explicitly to examine ROS in specific tissues in an effort to better understand the role of Wolbachia in influencing iron metabolism and the mosquito immune system. The number of infective stage filarial worms that developed within Ae. polynesiensis differed significantly between the strains. Specifically, the wild-type Wolbachia infected APM and its aposymbiotic counterpart (APMT) had similar numbers of successfully developing, infective L3 worms. This is comparable to the result of previous studies in which the removal of naturally occurring Wolbachia in Ae. pseudoscutellaris had no effect on the mean number of worms [53]. In contrast, significantly lower worm loads were observed in the artificially infected MTB compared to both the naturally infected and aposymbiotic strains. The latter is consistent with a prior experiment in which Wolbachia was artificially introduced into Ae. aegypti [12]. In the prior study, significantly lower B. pahangi numbers were observed in the artificially infected strain, relative to the naturally uninfected Ae. aegypti. Using the substantial genomic information available for Ae. aegypti, the authors speculated upon an association between an observed constitutive up-regulation of immune genes and an observed inhibition of filarial worm development. With the future development of additional genetic tools for Ae. polynesiensis, a similar approach can be used downstream to examine for an impact of artificial Wolbachia infection. Artificial Wolbachia infections can detrimentally affect the fitness of their hosts [54]–[56]. In this study, reduced survival was observed for females with artificial Wolbachia infection types when fed on infected or uninfected blood. The decreased number of L3 filarial worms within MTB may be due, at least in part, to a reduced ability of MTB females to tolerate filarial worm infections and their premature deaths prior to dissection assays. The observed ROS variation is an additional potential explanation for the observed variation in filaria development. Recent studies show that changes in ROS levels can affect pathogen development. In An. gambiae, high levels of ROS were associated with increased melanotic encapsulation of Plasmodium parasites [22], [27]. In Ae. aegypti, increased ROS expression is associated with induction of the Toll pathway, which mediates the expression of antimicrobial peptides and antioxidants to balance oxidative stress and is associated with reduced dengue virus titer [17]. In addition, ROS generated independent of the mosquito immune system by the native microflora have also been found to negatively affect development of Plasmodium parasites [21]. It is also relevant to highlight the importance of iron to filarial worm development [57], [58]. If the artificial Wolbachia infection in Ae. polynesiensis were to affect the regulation of iron, as previously discussed, then filarial worm development and survival may be affected in the MTB strain. However, a simple direct association with overall ROS levels cannot explain the pattern of differential filarial worm development that was observed here, since the overall ROS levels were lower in MTB, relative to wild type mosquitoes. Furthermore, the lower ROS levels observed in the aposymbiotic APMT strain was not observed to be associated with reduced filarial development. The variation observed between experimental replicates is not unexpected and is similar to prior reports [59]–[61]. Possible reasons for this variability include differences in sausage casing thickness, blood quality, and additional factors that affect mosquito feeding and microfilariae. Importantly and directly related to the study design, regardless of the variation between replications, the observed differences between strains remained the same (Figure 3C). Filarial worm infections in mosquitoes are not benign. They can cause damage to the midgut and flight muscles [62], [63], sometimes affecting flight behavior [64]. Increasing the number of worms in an infected blood meal decreases mosquito survival rates [65]. Our results confirm the detrimental nature of filarial worms present in a blood meal, where a general reduction in survival was observed for all of the examined strains (Figure 4). Blood contains an important nutritional component for adult mosquitoes [66]–[68] so it is not surprising that reduced survival was associated with blood deprivation of wild-type APM mosquitoes. However, this pattern was reversed in the aposymbiotic APMT and artificially infected MTB strains, where we observed a significant reduction in survivorship after blood feeding as compared to sucrose fed mosquitoes (Figure 5). As described above, this may reflect an evolved mutualism between the wPolA infection and Ae. polynesiensis, since increased survival of blood fed females is adaptive for both the anautogenous mosquito and the maternally inherited Wolbachia infection. The observed pattern of decreased survivorship was similar to the pattern of H2O2 levels (Figure 1), suggesting an association with ROS homeostasis. Similar to our results, a recent study observed a link between decreased ROS levels and the proliferation of gut bacteria, which can be detrimental to the survival of the mosquito [27], [28]. In An. gambiae that were artificially infected with Wolbachia, no fitness effect was observed until a blood meal was taken. The authors proposed that this virulence could be due to modulated ROS levels and proliferation of gut bacteria within the mosquito following a blood meal [14]. In this study we were able to successfully infect Ae. polynesiensis with an artificial Wolbachia type that is stably maintained and causes bidirectional cytoplasmic incompatibility when crossed with the wild-type strain. We observed that the removal of Wolbachia and subsequent introduction of a novel Wolbachia type into Ae. polynesiensis affected host physiology and filarial worm development. We observed significant effects on ROS production both before and after blood feeding. The artificially infected mosquitoes varied also in their ability to support filarial worm development. Decreased survival was observed for blood fed mosquito strains that were cleared of their natural Wolbachia infection. The results encourage additional investigation into the specific physiological mechanisms affected by the artificial Wolbachia infection in Ae. polynesiensis. In addition, the findings presented here lend support for additional experiments with the human parasite, W. bancrofti. Although B. pahangi is a commonly used model system, it is important to determine if similar inhibitory effects can be observed against W. bancrofti. However, as there is no animal model, the latter will require transporting the transinfected mosquitoes to an endemic area or importing infected blood. The observed experimental outcomes suggest applied strategies to impact filarial worm transmission by Ae. polynesiensis in the South Pacific. Specifically, the bidirectional incompatibility occurring in crosses of MTB and the wild-type provides a potential means to reduce the population of this important vector [2]. Another avenue of control is to purposely replace the existing population with MTB. The latter strategy would be similar to ongoing work in Australia against dengue transmission by Ae. aegypti [8], [9]. However, since MTB is bidirectionally incompatible with the wild type population, one would not expect CI to drive the spread of Wolbachia. Instead, the strategy would be suppression followed by female releases to establish the new infection type [69]. Furthermore, the decreased MTB survivorship observed in blood fed females and their reduced filarial worm development are consistent with the phenotype desired for a strategy in which the indigenous Ae. polynesiensis population is replaced with a strain less able to transmit filarial worms. This study was performed in strict accordance with the recommendations in the Guide for Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Kentucky (Protocol number: 00905A2005). Aedes albopictus (HOU) and an aposymbiotic Ae. polynesiensis strain (APMT) were used as Wolbachia donor and recipient, respectively. The HOU donor strain is naturally super-infected with two Wolbachia types, wAlbA and wAlbB [70]. The recipient strain, APMT, was generated by tetracycline treatment of the APM strain [2] and has been maintained for >30 generations in the laboratory. The wild-type APM strain is single infected with wPolA and exhibits a 100% infection rates in wild populations [30], [71] (Table 1). APM was used in crossing tests to characterize the newly generated strain, MTB. Unless otherwise specified, mosquitoes were maintained using standard insectary conditions at 28±2°C, 75±10%RH, and a photoperiod of 18∶6 h (L∶D). Larvae were reared in optimal conditions, at low density in excess of 6% liver powder solution (MP Biomedicals, LLC, Solon, OH), until pupation. Adult mosquitoes were provided with a 10% sucrose solution ad libitum, and a blood meal was given once a week with anesthetized mice. Collection, preparation and microinjection of embryos were based upon successful techniques used for previous mosquito transfections [23], [31]. Injection needles were prepared using quartz glass capillaries with an outer diameter (OD) of 1. 00 mm, an inner diameter (ID) of 0. 70 mm, and a length of 7. 5 cm (QF100-70-7. 5; Sutter Instrument Co. , Novato, CA). Needles were beveled at a 15° angle using a micropipette beveler, model BV-10 (Sutter Instrument Co. , Novato, CA). Microinjection was done using an Olympus IX70 inverted microscope (Olympus Co. , Tokyo, Japan) at ×200 magnification. Blood-fed APMT females were held in Drosophila vials (Fisher Scientific) containing wet germination paper (Anchor Paper Co. , Saint Paul, MN) and allowed to oviposit. Recipient embryos (APMT) to be injected were collected, aligned on wet germination paper, briefly desiccated and covered with water-saturated halocarbon 700 oil (Sigma-Aldrich Co.). Donor HOU embryos were treated similarly, but not desiccated. Cytoplasm was withdrawn from the posterior of donor HOU embryos and injected using an IM 300 microinjector (Narishige Scientific, Tokyo, Japan) into the posterior of the recipient APMT embryos. Recipient embryos were injected up to 90 minutes post-oviposition. After injection, the embryos were incubated under standard conditions for approximately 40 minutes. Injected embryos were removed from oil and transferred to wet germination paper, where they were allowed to develop for 5 days. The eggs were hatched (G0) and reared using standard maintenance conditions. Females of the parent generation (G0) were isolated as virgins and mated with APMT males, yielding a new strain named MTB. After oviposition, G0 females and males were assayed for both presence of Wolbachia infection and type using PCR (see below) (Table 3). Females that were negative for Wolbachia were discarded along with their progeny. Daughters (G1) from infected G0 females were isolated as virgins and outcrossed with APMT males. All G1 females that oviposited were tested for Wolbachia infection by PCR. PCR-guided selection was performed for 6 generations (G1–G6) (Table 3). At G7 the MTB strain was closed (i. e. not outcrossed with APMT males, but crossed with MTB males), and PCR was used to monitor the frequency of infection periodically through the following generations. All infection types were confirmed using Wolbachia specific primers and PCR. Adults were homogenized in 100 µl of buffer containing 10 mM Tris-HCl, 1 mM EDTA and 50 mM NaCl using a Mini-beadbeater (BioSpec Products, Inc. , Bartlesville, OK), boiled for 5 minutes and centrifuged at 14,000 rpm for 5 minutes. Two µl of supernatant were used for each PCR reaction. PCR reactions were amplified in 50 mM KCl, 20 mM Tris-HCl (pH 8. 4), 1. 5 mM MgCl2,0. 25 mM dNTPs, 0. 5 mM primers and 1 U Taq DNA polymerase in a total volume of 25 µl. Wolbachia infection in all strains was confirmed using general Wolbachia primers 438F (5′CAT ACC TAT TCG AAG GGA TAG-3′) and 438R (5′AGC TTC GAG TGA AAC CAA TTC-3′) and PCR cycling conditions of 94°C 2 minutes, 39 cycles of 94°C for 30 seconds, 55°C for 45 seconds and 72°C for 1 minute 30 seconds, followed by a final extension temperature of 72°C for 10 minutes. Infection type of all strains was confirmed using A-clade (136F and 691R) or B-clade (81F and 522R) specific primers [72]. PCR cycling conditions were 94°C for 4 minutes, followed by 35 cycles of 94°C for 1 minute, 48°C (A) or 55°C (B) for 1 minute and 72°C for 1 minute and a final extension temperature of 72°C for 10 minutes. Similarly aged egg papers from APM and MTB were hatched concurrently in dilute liver powder solution (∼0. 6 g/L). One hundred first instar larvae were moved into a rearing container and fed optimally until pupation. Pupae were isolated in individual test tubes to ensure virginity. After eclosion, 20 virgin adults were introduced into a crossing cage at a 1∶1 sex ratio and allowed to mate. A full factorial crossing design between APM and MTB was implemented, and four replicates were performed for each cross (Table 4). An oviposition cup was available to females in each crossing cage for one week, after which the cup was removed. Eggs remained hydrated and were allowed to mature for 10 days. Egg papers were removed from the oviposition cups and hatched by submersion for two days in dilute liver powder solution. All eggs were examined by microscope to determine the total number of eggs and the proportion hatched as indicated by the position of the operculum. The normality of the hatch rate data was analyzed using a Shapiro-Wilkes test (JMP, SAS Institute, Cary, NC). A Kruskal-Wallis test was used to determine overall significance and post-hoc Wilcoxon tests were used for pairwise comparisons of hatch rates between crosses. To determine ROS levels in mosquitoes fed sucrose only, whole bodies of seven-day-old APM, APMT and MTB were collected in 150 µl of 1× PBS containing 2 mg/ml of the catalase inhibitor 3-amino-1,2, 4-trizole. To determine ROS levels in blood fed mosquitoes, six-day-old APM, APMT and MTB were provided with a blood meal from an anesthetized mouse. Twenty-four hours after blood feeding, the midgut was dissected from the mosquito and the blood bolus was flushed from the midgut using 1× PBS with catalase inhibitor. Mosquito carcasses and midgut tissues were collected in 1× PBS with catalase inhibitor. For both treatments, samples were homogenized then centrifuged for 5 minutes at 10,000 g. The supernatant was filtrated through a 10 K molecular weight cutoff spin filter (Corning SpinXUF; Corning Incorporated Life Sciences). The elution was collected and tested using a Hydrogen Peroxide Assay kit (BioVision) following manufacturer' s instructions. The fluorescence intensity was detected with Excitation/Emission 544/590 using a fluorescence microplate reader (Fluoroskan Ascent FL, Thermolabsystems). Five biological replicates, with three females for each strain were used for each treatment. A general linearized model with a normal distribution was used to determine if ROS levels differed between strain, feeding status or strain×feeding status. The sucrose treatment and the blood treatment were analyzed using separate ANOVAs with post hoc Tukey HSD comparisons. Three replicates were performed to test for relative filarial susceptibility between strains. Brugia pahangi-infected dog blood was provided from the NIH/NIAD Filariasis Research Reagent Resource Center at the University of Georgia. Egg papers for APM, APMT and MTB were hatched concurrently and reared under standard maintenance conditions. Adult female mosquitoes were anesthetized using chloroform, and 75–90 mosquitoes were placed into cages. They were provided with a 10% sucrose solution and given 3 days to acclimate to the cage. Females aged 3–5 days were sucrose starved for 6 hours prior to blood feeding. They were given a Brugia-infected blood meal (10 microfilariae/µl) using sausage casing and a Hemotek membrane feeding system (Discovery Workshops, Accrington, UK) that maintained the blood at 37°C. All mosquito strains were allowed access to blood for 2 hours. After feeding, females were allowed to rest for one hour before sorting. All mosquitoes were briefly anesthetized using chloroform and observed under a microscope for presence of a blood bolus. Blood fed and non-blood fed females were placed into separate cages. Ten days after feeding, surviving blood fed females were anesthetized on ice and dissected in sterilized Hank' s balanced salt solution (Sigma-Aldrich). Individual mosquitoes were examined for L3 parasites by microscopy. The total number of filarial worms in each mosquito was recorded. To determine whether worm load data were normal, a Shapiro-Wilkes test was used (JMP, SAS Institute, Cary, NC). We built a general linearized model with a Poisson distribution to determine if mean worm load differed across replicates or between strains. Post-hoc contrasts were used to compare worm loads between strains. To correct for multiple comparisons we used the Benjamini-Hochberg correction with an α value of 0. 05 [73]. Mosquito rearing and blood feeding methods were the same as those described in “filarial susceptibility testing. ” We recorded the number of mosquitoes alive and dead ten days after feeding on different blood meal types to compare differences in survival between APM, APMT and MTB. Three separate experiments were performed: 1) comparisons between strains fed on Brugia-infected blood only, 2) comparisons between mosquitoes fed uninfected and Brugia-infected blood meals, and 3) comparisons between mosquitoes that were blood fed or fed sucrose only. For each of the above experiments, we built a general linearized model with a binomial distribution to determine if survivorship at day 10 differed between replicate, strain and blood meal type (JMP, SAS Institute, Cary, NC). Post-hoc contrasts were used to compare survivorship between strains. To correct for multiple comparisons we used the Benjamini-Hochberg correction with an α value of 0. 05 [73].

Lymphatic filariasis (LF), the leading cause of morbidity in South Pacific regions, is caused by a filarial worm, Wuchereria bancrofti. Elimination of LF in the South Pacific requires an approach integrating both mass drug administration and strategies targeting the primary mosquito vector, Aedes polynesiensis. Ae. polynesiensis is naturally infected with Wolbachia, an endosymbiotic bacterium that is a focus of novel control strategies, due to its ability to affect mosquito reproduction and interfere with pathogen development. Artificial Wolbachia infections are associated with increased levels of reactive oxygen species (ROS), which can alter immune gene expression and inhibit dengue proliferation. Here, we describe the generation of an Ae. polynesiensis strain that has been artificially infected with Wolbachia from Ae. albopictus. The infection is stably maintained and causes conditional sterility when crossed with the wild-type. The artificially infected strain exhibits different ROS levels than the wild-type, indicating a decreased ability to regulate oxidative stress. The number of successfully developing infective stage filarial worms was reduced in the artificially infected strain. In addition, survival of the artificially infected strain was significantly lower than the wild-type. The artificially infected Ae. polynesiensis strain is discussed in relation to ongoing mosquito-borne disease control efforts.

Abstract Introduction Results Discussion Materials and Methods

biotechnology applied microbiology medicine veterinary diseases zoonotic diseases immunology biology veterinary science immune response

2012

Reactive Oxygen Species Production and Brugia pahangi Survivorship in Aedes polynesiensis with Artificial Wolbachia Infection Types

Inflammatory bowel disease (IBD) is a chronic condition driven by loss of homeostasis between the mucosal immune system, the commensal gut microbiota, and the intestinal epithelium. Our goal is to understand how these components of the intestinal ecosystem cooperate to control homeostasis. By combining quantitative measures of epithelial hyperplasia and immune infiltration with multivariate analysis of inter- and intracellular signaling, we identified epithelial mammalian target of rapamycin (mTOR) signaling as a potential driver of inflammation in a mouse model of colitis. A kinetic analysis of mTOR inhibition revealed that the pathway regulates epithelial differentiation, which in turn controls the cytokine milieu of the colon. Consistent with our in vivo analysis, we found that cytokine expression of organoids grown ex vivo, in the absence of bacteria and immune cells, was dependent on differentiation state. Our study suggests that proper differentiation of epithelial cells is an important feature of colonic homeostasis because of its effect on the secretion of inflammatory cytokines. IBD, comprised of Crohn’s disease (CD) and ulcerative colitis (UC), is characterized by chronic inflammation of the gastrointestinal tract. The symptoms of IBD, which include diarrhea, abdominal pain, and intestinal blockage, are chronic and debilitating and, in extreme cases, can result in death. Treatment options include steroids, aminosalicylates, and targeted therapies such as tumor necrosis factor alpha (TNF-α) neutralizing antibodies, but many patients become refractory to all of these therapies and require surgery [1]. As such, there is great need for medical therapies that exhibit a more durable response. Generally, IBDs are understood to result from a loss of homeostasis between the intestinal epithelium, the mucosal immune system, and the gut microbiome. Genetic approaches, such as genome-wide association studies (GWAS), have identified numerous single nucleotide polymorphisms (SNPs) associated with IBD risk, many of which are involved in adaptive and innate immune function [2,3]. While the immune system is clearly one of the key drivers of IBD, the intestinal epithelium also plays a central role in preventing inflammation by maintaining homeostasis of the gut. It represents a critical physical barrier between the commensal flora and the immune system that resides in the lamina propria and also plays a central role in antigen presentation [4,5]. Without the epithelium, it would be impossible to maintain proper homeostatic control of the mucosal immune system. Indeed, IBD results from hyperactivation of the immune system in response to commensal or pathogenic bacteria in the context of epithelial damage. Although experimental approaches indicate that changes to any of the 3 main components of the intestinal ecosystem can trigger IBD, the resulting clinical presentation of patients with CD or UC is indistinguishable, regardless of initiating event. Because of this, we have hypothesized that disparate initiating events converge on a shared, self-sustaining disease network composed of pathologic changes to all 3 main components of the gut ecosystem. Therefore, therapeutic approaches that target genetic initiating events might fail because inhibition of the triggering event would be ineffective in a self-sustaining disease state. An approach that focuses directly on these convergent downstream physiological processes and signaling pathways may identify novel targets that provide sustained therapeutic value for a larger number of patients. In this paper, we have applied a protein-centric systems biology approach to characterize, at the tissue level, the key molecular and phenotypic features that comprise this convergent chronic inflammatory disease state. Through this work, we developed quantitative phenotypic readouts of inflammation and measured more than 50 inter- and intracellular signaling molecules that are associated with inflammation. Using dimensionality reduction algorithms, we predicted mammalian target of rapamycin (mTOR) signaling as a driver of colitis. While mTOR plays numerous roles that may be linked to IBD pathogenesis, including control of immune differentiation and activation and autophagy [6,7], in this context we found that mTOR’s regulation of the differentiation state of the intestinal epithelium plays a key role in sustaining chronic inflammation. Using mouse models and in vitro 3D systems, we found that undifferentiated colonic epithelium produces high levels of innate immune cytokines and chemokines that drive inflammation, comparable to those proinflammatory molecules regulated by mTOR during colitis. Altogether, this work has resulted in a systems-scale model of colitis and has identified defective epithelial differentiation as a central mediator of chronic bowel inflammation. By extension, IBD patients, and particularly those with UC, who have hyperproliferative and undifferentiated colonic epithelium, may benefit from therapeutic approaches that induce differentiation to break the cycle of chronic inflammation. To create a tissue-level systems model of chronic colitis, we utilized the T-cell transfer (TCT) model of colitis, which allowed us to control the timing and severity of inflammation in mice [8]. In this model, naïve T cells isolated from wild-type (WT) mice are injected into Rag1 null mice, which lack an adaptive immune system. In the absence of regulatory T cells (Tregs), the naïve T cells produce an inflammatory response to resident microflora, resulting in colitis with high penetrance (80%–90%) and a latency of 3–15 weeks. Our initial experimental population included 20 Rag1 null mice injected with naïve T cells and, as a negative control, 8 Rag1 null mice injected with Tregs (S11 Fig). Mice were weighed biweekly and assessed for symptoms of inflammation, such as diarrhea and rectal prolapse. After animals began to show severe weight loss and diarrhea, they were killed randomly over a range of time points to capture a spectrum of inflammatory states, an important feature for subsequent mathematical modeling. After humane killing, colons from individual mice were subdivided into matched sections for histology, immunophenotyping by flow cytometry, and multiplexed, Luminex-based protein measurement (S1 Fig). Naïve T-injected animals exhibited significant changes to the intestinal mucosa, including epithelial hyperplasia, expansion of the proliferative zone, loss of differentiation, and increased immune infiltrate (Fig 1A and S2 Fig). The majority (18/20) of the naïve T cell-injected animals showed some degree of crypt hyperplasia and immune cell infiltration, which were highly correlated to one another and with the amount of weight lost (Fig 1B–1D). The increased height of the colonic crypts was due to the dramatic increase in the number of epithelial cells per crypt (S2 Fig). Immunohistochemistry (S2 Fig) and immunophenotyping (S3 Fig) revealed that many immune cell types were present in increased numbers in the colons of animals with colitis. For example, in Rag1 null control animals, CD45+ cells made up only a small proportion of total colonic cells (0. 1%–0. 3%), and the majority of these cells were plasmacytoid dendritic cells (pDCs). In naïve T-injected animals, the percentage of CD45+ cells showed a strong correlation with crypt hyperplasia (Fig 1B), and this infiltrate was composed primarily of CD4+ T cells, macrophages, and neutrophils (S3 Fig), although the relative proportions of these cell types varied from animal to animal. In the course of phenotyping animals with colitis, we noted that the histological manifestations of colitis were not uniform throughout the colon but rather restricted to specific colonic regions. In order to quantitatively measure inflammation as a function of the geographical position in the colon, we measured the thickness of the mucosa every 5 crypts along the entire length of the colon for every mouse. These measurements were used to generate scatter plots that graph crypt height as a function of longitudinal location along the colon (Fig 1E). In addition to crypt height measurements, an immune infiltration score was used to identify and locate areas of inflammation. Foci of infiltrating immune cells were rarely found in control animals but were found throughout the colons of naïve T-injected animals. These infiltrates were more prevalent in continuously inflamed colons and were closely associated with areas of crypt hyperplasia (Fig 1E). After analyzing our entire experimental cohort, we found that we could divide experimental animals into 2 discrete classes based on the extent of colitis (Fig 1F). Continuously inflamed animals exhibited significant crypt hyperplasia (300–600 μm/crypt) throughout the distal colon, while focally inflamed animals were defined by variable crypt hyperplasia beginning at the proximal/distal junction and spreading distally towards the rectum. While animals that were killed at later time points had a higher prevalence of colitis, the presence and severity of colitis were most closely associated with weight loss and the time since initial weight loss (Fig 1C and 1D). In follow-up experiments, by using this weight loss metric to select the timing of drug treatments, we were able to ensure that we treated only animals with established continuous inflammation. Together, these data establish the quantitative and qualitative histological features of chronic colitis, defining crypt hyperplasia as a tissue-level feature of colitis in this model and identifying expansion of the proliferative zone and loss of differentiation as key cellular features of disease. Having established quantitative phenotypic measures of colitis, we sought to identify the inter- and intracellular signals underlying inflammation. Fresh tissue from the distal colon was lysed for Luminex-based analysis of 37 cytokines, chemokines, and growth factors and 17 phosphoproteins marking key signaling pathways (S4–S6 Figs). These 54 “signals” represented all of the commercially available Luminex-based assays for which we were able to detect signal above background in the mouse colon. Many of the signals were significantly different in control animals and those receiving adoptive transfer of CD45RBhi T cells, and these changes were not simply due to the adoptive transfer, since animals that received CD45RBhi T cells did not exhibit increased signaling prior to the onset of colitis (S7 Fig). We used principal component analysis (PCA) to cluster animals based on protein expression profiles and to identify key proteins driving this clustering. PCA showed a high degree of covariance between the proteins in the data set, and the majority of this variance was explained by the first principal component (PC1) (Fig 2A). Additionally, PC1 clustered each animal based on its degree of inflammation, with animals exhibiting continuous inflammation scoring positively on PC1, the focal inflammation animals scoring in the middle, and the control animals scoring negatively. Two of the naïve T cell-injected animals clustered with the negative control animals; however, these showed almost no hyperplasia or immune infiltration, representing the incomplete penetrance of this model (Fig 2A). Scores on PC1 were plotted against average crypt height and showed a strong linear correlation (R2 = 0. 80), confirming that this principal component is strongly associated with inflammation (Fig 2B). PC1 showed strong positive loading for chemokines involved in macrophage, neutrophil, and T-cell chemotaxis (macrophage inflammatory proteins [MIPs], KC [keratinocyte chemoattractant], monocyte chemoattractant protein 1 [MCP-1], and RANTES [regulated on activation, normal T cell expressed and secreted]), inflammatory cytokines (interleukin [IL]-1s and IL-6), and growth factors (leukemia inhibitory factor [LIF] and vascular endothelial growth factor [VEGF]) (Fig 2C). T helper cell 2 (Th2) cytokines (IL-31 and IL-33) did not show positive loading, suggesting polarization of the T-cell response towards a Th1/Th17-type response, which is typical of CD and the TCT model of IBD [9]. Intracellular signaling molecules, such as Mek and the insulin receptor, showed strong negative loading, suggesting that these pathways are down-regulated in inflamed colons. The main intracellular signaling pathways up-regulated in inflamed animals included NF-κB and mTOR. We were particularly interested in the mTOR pathway because it was activated at multiple levels, including Akt, p70 S6K, and S6RP. The activation of mTOR in the context of colitis was not specific to the TCT model; we found that mTOR was activated in multiple genetic models of IBD (S8 Fig) and in a subset of human IBD patients (S9 Fig). To determine whether mTOR signaling was generally up-regulated in the context of colitis or whether up-regulation in a specific cellular compartment accounted for the increased signal detected via Luminex, we performed immunohistochemistry for phosphorylated ribosomal protein S6 (p-S6) on colons from control and adoptive transfer mice. In control animals, we found high p-S6 signal only in intraepithelial lymphocytes, which is consistent with the well-characterized role for mTOR signaling in immune function (Fig 2D) [10]. In animals with colitis, by contrast, we detected strong p-S6 throughout the colonic epithelium (Fig 2D), suggesting that the increased mTOR signal that we detected derives primarily from ectopic activation in epithelial cells. In order to investigate a potential role for mTOR in driving colonic inflammation by affecting epithelial homeostasis, we tested the effects of the mTOR inhibitor rapamycin on colitis induced in the TCT model. To ensure experimental animals had severe inflammation, we utilized animals at late time points after adoptive transfer. We also used weight loss as a marker for onset of inflammation, beginning treatment on a mouse-by-mouse basis when an animal lost more than 1 gram of weight and did not regain the weight within 1 week. Inflamed animals were treated with rapamycin (5 mg/kg/day) for 2 weeks and then processed and analyzed as described above. Colons from rapamycin-treated animals exhibited a clear reduction in immune infiltrate, normalization of the proliferative zone at the bottom of the crypts, and restoration of goblet cell differentiation (Fig 3A and 3B). Quantification of crypt heights revealed a partial normalization of mucosal thickness and a significant decrease in average crypt height in rapamycin-treated animals (Fig 3C). Because rapamycin promotes expansion of Tregs in other systems [11], we explored whether the general reduction in inflammation following rapamycin treatment was associated with an increase in Tregs in the colonic lamina propria. While animals with colitis did have significant numbers of Tregs in the colon, those treated with rapamycin did not have greater numbers (Fig 3D) —in fact, they had fewer—indicating that rapamycin treatment did not suppress colitis in the TCT model by expanding the number of Tregs. Analysis of phosphoproteins confirmed that signaling downstream of mTOR activity was strongly inhibited by rapamycin (Fig 3E), although phosphorylation of the upstream activator Akt appeared to be increased by rapamycin, presumably by relieving inhibition of the insulin-like growth factor receptor/insulin receptor substrate 1 (IRS1) axis [12]. We next sought to determine whether mTOR inhibition altered the local intracellular signaling network or whether it had a broad effect on the tissue-level inflammatory network (S4–S6 Figs). We first applied partial least squares regression (PLSR) to signaling data from our initial specimens in order to create a mathematical model linking dysregulated signaling to colitis, regressing to epithelial thickness. As with our original PCA model, the PLSR model was able to segregate animals with continuous inflammation from control animals and those with focal inflammation based on the accumulated signals that were measured in colonic lysates (Fig 3F). When the signaling data from animals treated with rapamycin were projected onto this PLSR model, 6 out of 8 rapamycin-treated animals clustered with the focally inflamed animals (Fig 3F). This reflects the fact that rapamycin treatment represses signaling directly downstream of mTOR and also the majority of the cytokines, chemokines, and growth factors associated with inflammation. Together, these results suggest that epithelial mTOR is a central regulator of intestinal inflammation in mouse models of IBD. While long-term rapamycin treatment revealed a key role for mTOR in the maintenance of colitis, chronic inhibition does not provide mechanistic insight into the potential role that epithelial mTOR plays in this complex, tissue-level phenotype. In order to address this limitation, we treated inflamed animals acutely with rapamycin for 1,4, 8,24, or 48 hours. Within 1 hour of rapamycin treatment, goblet cells began to appear, and the proliferative zone began to normalize (Fig 4A and 4B). Within 24 hours, these cellular phenotypes were returned to nearly normal levels. By 48 hours, the inflammation-induced hyperplasia was reduced, reflecting the induced differentiation and decreased proliferation induced by mTOR inhibition (Fig 4C). Immune cell infiltrate changed more slowly following rapamycin treatment than did the epithelium, with a decrease in the proportion of macrophages and neutrophils only by 48 hours (Fig 4D). At that time point, the proportion of CD45+ cells that were T cells was significantly increased (Fig 4D). Cytokine and chemokine expression showed a variable kinetic response to rapamycin (S10 Fig). Some proteins decreased rapidly, within 1 hour of rapamycin treatment, to control levels, while others showed a graded decrease correlating to the goblet cell phenotype (Fig 4E). Some of the cytokines, like IL-6, even increased following rapamycin treatment (Fig 4E). These data are consistent with a kinetic model in which hyperactivated mTOR suppresses differentiation to promote colitis. Upon mTOR inhibition with rapamycin, colonic epithelial cells rapidly differentiate and cease to proliferate. Subsequently, the cytokine profile of the tissue changes, and, as a consequence, the complexion of the innate immune landscape returns to a state that resembles the normal colon. We next considered the possibility that rapamycin indirectly affected intestinal homeostasis by causing shifts in the composition of the commensal flora. To address this possibility, we profiled the fecal microbiome of control and inflamed animals, both before and after treatment with rapamycin (5 mg/kg/day for 2 weeks) (S11 Fig). Principal coordinate analysis (PCoA) of the beta diversity relationships between samples demonstrated that samples separated based on their inflammatory phenotype, but that rapamycin treatment did not shift the overall status of the microbiome, whether an animal had inflammation or not (S11 Fig). We also plotted the Bray-Curtis dissimilarity scores between all pairs of mice to see how individual mouse pre- and post-treatment scores compared with differences between animals. This analysis showed that the microbiome of control animals was most similar to other control animals, regardless of whether they received rapamycin (S11 Fig). The same was true for animals with inflammation. Although we did note that resolution of inflammation altered microbiome diversity and that the representation of certain bacteria, for example Lactobacilli, increased in inflamed mice after treatment, overall our data reveal that rapamycin itself does not alter microbial composition in this model. In our initial hypothesis-generating experiment and in the subsequent perturbation experiments, we found an inverse correlation between epithelial differentiation and colonic inflammation. This observation led us to ask whether induction of differentiation independently of mTOR inhibition would have a similar effect on tissue homeostasis. To this end, we used the gamma-secretase inhibitor dibenzazepine (DBZ), which induces goblet cell differentiation by inhibiting Notch signaling [13]. Animals with colitis were treated with 10 μmol/kg DBZ or vehicle by daily intraperitoneal (IP) injection for 1 week. DBZ-treated animals exhibited complete conversion of the colonic epithelium to goblet cells, with high mucous levels in almost every cell in the crypt (Fig 5A). Consistent with rapamycin treatment experiments, induction of differentiation via Notch inhibition was accompanied by a reduction in macrophages and neutrophils (Fig 5B). Interestingly, there was no change in the proportion of T cells within the colon, suggesting that their recruitment and maintenance within the tissue is not affected by systemic inhibition of Notch. This result is consistent with our prior observation that rapamycin treatment rapidly induced differentiation prior to having any effect on the number of T cells in the colon (Fig 4D). The rescue of differentiation was also associated with a decrease in the inflammation-associated chemokine profile (Fig 5C and S12 and S13 Figs), particularly those associated with innate immune cell chemotaxis. When the DBZ- and vehicle-treated samples were projected onto a cytokine-specific PLSR model built from the baseline samples, DBZ-treated samples clustered with focally inflamed and negative control animals, just like the rapamycin-treated animals in our previous experiment (Fig 5D). These experiments demonstrate that 2 mechanistically independent methods of inducing differentiation in the colonic epithelium lead to a reduction in the expression of inflammatory cytokines and chemokines and subsequent reduction in inflammatory immune infiltrate. These observations provide strong evidence that the epithelium is a key player, beyond its role as a barrier, in the inflammatory process and that this connection is tightly linked to differentiation state. Our studies indicate a clear role for the differentiation state of the epithelium in mediating cytokine and chemokine expression and innate immune recruitment in murine colitis. Nevertheless, the stage at which the epithelial cells were arrested remained unclear. In order to address this, we performed microarray analysis on 4 control and 6 inflamed colons (GEO accession GSE87317). Analysis of individual markers for different cell types suggested that there was a pronounced decrease of goblet cells and enteroendocrine cells and a moderate decrease of enterocytes and stem cells (Fig 6A and 6B), cellular phenotypes that we have confirmed by histologic analysis (Fig 1A and S2 Fig). Because transit amplifying cells (TACs) are defined functionally, the existing monovariate markers are not sufficient to assess their enrichment. In order to determine whether there was a relative increase in TACs in inflamed colon, we employed Gene Set Enrichment Analysis (GSEA) using published gene sets derived from epithelium enriched for various cell types. A stem cell gene set was derived from sorted Lgr5+ epithelial cells [14], while enterocyte and secretory progenitor gene sets were derived from various pharmacologic and genetic treatments [15]. Finally, a TAC signature was derived from regenerative epithelium composed of 97% TACs [16]. Using GSEA, we found de-enrichment for gene sets associated with secretory cells, enterocytes, and stem cells (Fig 6C and 6D). Conversely, we saw enrichment for genes associated with Atoh1 knockout (which lacks goblet cells) and the TAC signature. Importantly, several of the genes found in the TAC signature are chemokines, all of which were strongly induced transcriptionally in our microarray data set (Fig 6E). By contrast, there were no cytokines or chemokines present in any of the gene sets for stem or differentiated cell types. This observation confirms that cytokine expression is present within the epithelium itself and that this is enhanced in TACs, but not in stem cells or differentiated cells. Together, these analyses suggest that the suppressed differentiation and enhanced proliferation that is seen in colitis is the result of expansion of the TAC compartment and not the result of stem cell expansion. Although our in vivo results were consistent with a model in which undifferentiated epithelium supplies many of the cytokines needed to recruit inflammatory cells to the colonic lamina propria, these experiments did not rule out the possibility that these cytokine and chemokine changes are secondary effects originating from immune cells. In order to test whether the expression of proinflammatory signaling molecules is intrinsic to the epithelium, we measured the production of cytokines and chemokines as a function of epithelial differentiation state in an in vitro 3D organoid model. Crypts were isolated from WT mouse colons and developed into 3D organoids in matrigel following well-established protocols [17]. We then shifted the growth conditions to maintain a mixed cell population to enrich for proliferative stem cells or to differentiate into the enterocyte or goblet cell lineages (S14 Fig) [15]. Following confirmation of appropriate differentiation based on expression of cell type markers (Fig 7A), purified organoids were analyzed for expression of the suite of cytokines, chemokines, and growth factors that were measured in intact colons (S15 and S16 Figs). Relative to the stem cell-enriched and mixed population organoids, the majority of cytokines and chemokines were suppressed in organoids enriched for secretory and/or absorptive cells. For example, MCP-1 was reduced during differentiation in a lineage-independent manner, while VEGF and mitokine induced by gamma interferon (MIG) were reduced only when cells were differentiated into the goblet lineage (Fig 7B). Interestingly, MIP-1α, which was strongly induced in animals with inflammation (S5 Fig), was increased in expression when organoids were differentiated into the goblet lineage. Altogether, these data suggest that differentiation alters the cytokine and chemokine expression pattern of the colonic epithelium independent of immune cells. To put the in vitro analysis of cytokine/chemokine expression into the context of our prior in vivo studies, we projected the organoid expression data onto the same PLSR model from Fig 5D, where expression of the cytokines and chemokines highlighted in Fig 5C specifies latent variable 1 (LV1). In this analysis, the more stem-like organoids were projected positively on LV1, clustering with the inflamed mouse colons (Fig 7C). Conversely, the absorptive- and secretory-differentiated organoids clustered with the focal and noninflamed colons. This analysis demonstrates that the patterns of cytokine and chemokine expression that correlate with loss of differentiation and immune cell infiltration in inflamed colons in vivo are largely recapitulated in an in vitro system that lacks both bacteria and immune cells. Together with the in vivo mouse experiments, this result indicates that the colonic epithelium plays an active role in inflammation that extends beyond its role in maintaining barrier integrity. The goals of this study were to define the tissue-level disease state in a mouse model of IBD and to use this information to understand how the components of the intestinal ecosystem interact to maintain homeostasis. We focused on the epithelium, the immune system, and the molecules that mediate interaction between them, using computational approaches to identify correlations between molecular, cellular, and histological characteristics of the inflammatory state in the TCT model. In doing so, we established a hypothesis-generating platform to identify pathways important for homeostasis and developed a composite phenotype that was used in subsequent experiments to measure the effects of pathway perturbation at a systems level (Fig 1). Using our systems approach, we identified activation of the mTOR pathway as being one of the strongest correlates of colonic inflammation in this model (Fig 2). To determine whether this pathway plays an active role in the disease, we treated sick animals with rapamycin. Rapamycin (or sirolimus) is an inhibitor of the target of rapamycin complex 1 (TORC1) complex, but it should be noted that it can also exert mTOR-independent effects through its ability to bind FK506 binding protein 12 (FKBP12) [18]. Rapamycin treatment resulted in global changes in the inflammatory phenotype, including reduction of crypt hyperplasia, increased differentiation and decreased proliferation in the intestinal epithelium, decreased immune infiltration, and globally decreased expression of molecular correlates of inflammation (Fig 3). Notably, and contrary to prior observations [11,19], rapamycin did not increase the number of Tregs in the colon of the TCT model (Fig 3D), nor did it affect the overall composition of the colonic microbiome (S11 Fig). While the role of mTOR signaling has not been studied in the TCT model of CD, several studies have demonstrated efficacy for rapamycin and related molecules (rapalogs) in mouse models of colitis, including chronically dextran sodium sulfate (DSS) -treated and IL-10 knockout mice [7,20,21], suggesting that mTOR might be a therapeutic target for IBD. In addition to our mouse studies, we also found that a subset of human IBD patients—those with ileal CD—had increased mTOR pathway activity in inflamed tissue (S9 Fig). Nevertheless, mTOR inhibitors have shown limited efficacy in unselected clinical trials for adult IBD patients [22], although rapamycin has shown clinical efficacy in individual adult cases and in a larger study of pediatric IBD patients [23,24]. A clinical trial on patients selected for high mTOR signaling, perhaps using p-S6 as a biomarker, will be required to determine whether mTOR inhibition has potential as an effective therapeutic strategy for some IBD patients. The identification of mTOR signaling as a mediator of colitis in the TCT model provided an entryway to study the cellular mechanisms controlling intestinal homeostasis. While mTOR has been linked to mouse colitis through the effect of rapamycin, the pathway was presumed to promote inflammation by affecting T helper cell proliferation and polarization, and the role of epithelial mTOR was not explored [7,21]. By contrast, one key facet of the inflammatory phenotype that we found to be controlled by mTOR was loss of differentiation and expansion of the epithelial proliferative zone. Defective epithelial differentiation is a feature of IBD in human patients; goblet cell depletion has been reported in both CD and UC and has been linked to reduced cytokine secretion and antigen presentation [25,26]. In our acute treatment experiments, we found that epithelial differentiation was rapidly induced by rapamycin and that this preceded changes in cytokines and chemokines, which preceded changes in innate immune infiltration (Fig 4). These data support a mechanism whereby induction of differentiation results in reduced chemokine signaling and reduced inflammatory infiltrate. The role that epithelial differentiation plays in protection from colitis has been explored in other mouse models. For example, mice in which differentiation of goblet cells is suppressed, such as those lacking Jak3 or overexpressing Claudin-1, are predisposed to spontaneous and DSS-induced colitis [27,28]. Although goblet cells are important for establishing the barrier, these studies failed to recognize the potential for signaling interaction between the epithelial and immune components of the intestinal environment. To test directly whether cellular differentiation state, rather than mTOR activation state, controlled the cytokine profile of the colon, we used the Notch inhibitor DBZ to drive goblet cell differentiation in inflamed epithelium, and this was also associated with decreased cytokine and chemokine expression and immune infiltrate (Fig 5). Pharmacologic inhibition of Notch is known to protect against DSS-induced colitis, and genetic inhibition of Notch was found to decrease the expression of proinflammatory cytokines in cultured Caco-2 cells [29,30]. Notch signaling is the primary regulator of goblet cell differentiation, and most of the studies linking differentiation to colitis have focused on Notch. Interestingly, although Tsc2, an upstream regulator of mTOR signaling, has also been shown to regulate Notch [31], we could not find any evidence in our protein or gene expression data indicating that Notch signaling, in addition to mTOR signaling, is activated in the TCT model of colitis. This observation suggests that mTOR is affecting differentiation independently of Notch in the TCT model. Our transcriptional profiling analysis suggested that inflammation-associated mTOR activation is associated with an expansion of the transit amplifying pool (Fig 6), consistent with prior observations that mTOR signaling is important for repair of the colonic epithelium following damage, a process that is driven by TACs [32,33]. Together, these observations are consistent with our model linking differentiation, and not mTOR or Notch signaling directly, to proinflammatory signaling by the epithelium. Inhibition of both mTOR and Notch can affect the function of lymphocytes directly [34,35], so our in vivo studies could not rule out a contribution of perturbation of immune cell function to the overall suppression of inflammation in animals treated with inhibitors. To address this limitation, we used an in vitro organoid system to investigate the relationship between differentiation state and cytokine secretion in epithelium that is isolated from the microbiome and the immune system. We found that proliferative organoids expressed higher levels of inflammation-associated cytokines, chemokines, and growth factors than either absorptive- or secretory-differentiated organoids (Fig 7). This result demonstrates that, even in the absence of gut bacteria and immune cells, the differentiation state of the epithelium, and not necessarily the activation state of mTOR or Notch, specifies the expression of immunomodulatory molecules. Altogether, our work clarifies the important role that epithelial differentiation and epithelium-immune cross talk plays in maintaining overall colonic homeostasis. We demonstrate that undifferentiated epithelium plays an active role in maintaining inflammation by secreting chemokines that recruit innate immune cells such as macrophages and neutrophils. By extension, altered epithelial homeostasis is a central feature of self-perpetuating inflammation in the colon; since expansion of TACs is required for epithelial repair, the repair that serves to restore barrier function during acute inflammation will also function to recruit additional inflammatory cells to the site of damage. Notably, modulation of the proliferative state of the epithelium had minimal impact on T cell numbers within the tissue. In both mouse and human IBD, dysregulated T-cell activity is thought to be the key triggering event in inflammatory pathology. Our work suggests that T-cell activation is upstream of epithelial differentiation defects in the chain of events that leads to chronic inflammation. We propose that therapeutic approaches that promote epithelial differentiation could show efficacy in reducing chronic inflammation in IBD. TCT was performed according to established methods [8]. WT and Rag1-deficient mice on the C57BL/6 background were used for TCT. These animals were purchased from the Jackson Laboratory (Bar Harbor, Maine, United States). Naïve T cells and Tregs were collected on the BD Aria sorter and injected at 400,000 and 200,000 cells/animal, respectively. All animal work was approved by the Institutional Care and Use Committees of Massachusetts General Hospital and Beth Israel Deaconess Medical Center under protocol numbers 2007N000058,078–2014, and 080–2017. Approved protocols conformed to the USDA Animal Welfare Act, the PHS Policy on Humane Care and Use of Laboratory Animals, and the “ILAR Guide for the Care and Use of Laboratory Animals. ” Rapamycin was purchased from LC Laboratories (R-5000), and treatment was carried out for 1,4, 8,24, or 48 hours for the acute experiment and for 2 weeks for the extended experiment. Rapamycin was injected IP at 5 mg/kg in a vehicle of 5. 2% Tween 80 and 5. 2% PEG400 in water. DBZ was purchased from Sigma (SML0649-25MG) and used at 10 μmol/kg in water with 0. 5% HPMC and 0. 1% Tween 80. Colons were removed and flushed with PBS, and one-fifth lengthwise was taken for flow cytometry. A matched fifth was cut into proximal and distal regions and lysed in 250 μl Bio-Plex lysis buffer with protease inhibitor, factors 1 and 2, and PMSF. The final piece was rolled and fixed overnight in 10% formalin for histology. For microarray tissue collection, one-fifth was snap frozen and processed as described below. This replaced flow cytometric measurement for 4 Treg and 6 inflamed animals from the initial study. For crypt height measurements and gross histology, 5 μm sections of rolls were stained with hematoxylin and eosin (HE) according to standard protocols. Alcian Blue/Periodic Acid/Schiff was used from Leica (38016SS3A, 38016SS4A, and 38016SS4B), and goblet cells were visualized according to reagent instructions. Immunohistochemistry for phospho-Histone H3 S10 (Cell Signaling Technology 9701), liver fatty acid binding protein (Abcam ab7366), Ki-67 (Cell Signaling Technology 12202), chromogranin A (Abcam ab15160), and phospho-S6 ribosomal protein S235/236 (Cell Signaling Technology 4858) was carried out with citrate antigen retrieval and visualized using horseradish peroxidase reaction. For Foxp3 immunohistochemistry, antigen retrieval was performed in DAKO Target Retrieval Solution (#S1699) in a pressure cooker. Tissue sections were blocked with DAKO Serum-Free Protein Block (#X0909) and primary anti-E-cadherin (BD Medical Technology #BDB610181) and anti-FoxP3 (Abcam ab54501) antibodies were incubated overnight at 4°C in DAKO Antibody Diluent (#S3022). Following PBS washes, Alexa Fluor-labeled secondary antibodies (ThermoFisher) were incubated in DAKO Antibody Diluent for 1 hour at room temperature. Slides were mounted in ProLong mounting medium (ThermoFisher #P36962) and imaged on a Zeiss Axio Imager Z2. Crypt height measurements were performed using an Olympus microscope or slide scanner software. Briefly, an arbitrary line was drawn from the crypt base to the top of the crypt every 5 crypts across the length of the colon. Crypt measurements in which multiple crypts were associated with a single line were deemed off axis and removed from subsequent analysis. For scatter plots, all measurements were evenly spaced across the distal and proximal regions separately using polyline measurements of the lengths of those sections. Averages were taken of all measurements in the distal and proximal regions separately. To calculate the proliferative zone, the distance from the crypt base to the highest PH3-positive cell or Ki-67-positive cell was divided by the total crypt height. Tissue was homogenized in serum-free DMEM with 2 mg/ml collagenase type I C (VWR 234153-100MG) and incubated for 1 hour at 37°C. Following incubation, the sample was strained through a 45-μm filter and spun down for 5 minutes at 700 g. Cells were then stained with the following antibodies from BioLegend (1: 300 in fluorescence-activated cell sorting [FACS] buffer) for 10 minutes: Alexa-488 CD326 (118210), BV-421 F4/80 (123131), BV-605 CD4 (100547), BV-510 cd11b (101245), Alexa-700 Ly6G (127622), APC CD25 (102012), PE/Cy7 cd11c (117317), APC/Cy7 CD45 (103116), and PE CD45RB (103308). Cells were analyzed on a 5-laser LSR II (Becton Dickson). Details of markers used to quantify individual cells types are presented in S2 Fig. Luminex-based protein measurements were carried out according to manufacturer’s instructions, as we have reported previously [34,36,37]. The following kits from Bio-Rad were used: Group I Cytokine Assay (M60-009RDPD), Group II 9-plex (MD0-00000EL), and Group III Th17 panel (171-FA001M). BioplexPro phosphoprotein measurements were carried out with the following groupings: 10-plex: Akt Ser473 (171-V50001M), c-Jun Ser63 (171-V50003M), CREB Ser133 (171-V50028M), Erk1/2 Thr202/Tyr204, Thr185/Tyr187 (171-V50006M), GSK-3 Ser21/Ser9 (171-V50007M), Jnk Thr183/Tyr185 (171-V50011M), Mek1 Ser217/Ser221 (171-V50012M), p38 MAPK Thr180/Tyr182 (171-V50014M), Stat3 Ser727 (171-V50021M), p90Rsk Ser380 (171-V50035M); single-plex: IR-1 Tyr1146 (171-V50031M), IRS-1 Ser636/Ser639 (171-V50030M), IκB Ser32/Ser36 (171-V50010M), p70S6K Thr389 (171-V50016M), S6RP Ser235/Ser236 (171-V50038M), and Atf-2 Thr71 (171-V50024M). Crypt epithelial cells were isolated from mouse colons by incubating the tissue in 2 mg/ml type I collagenase (Invitrogen). Organoids were grown in matrigel in the presence of EGF, Noggin, R-spondin, and Wnt3a as previously described [17,38]. During the first 4 days of culture, GSK3β inhibitor, CHIR99021 (3 μM, Stemgent), and histone deacetylase inhibitor and Notch agonist, VPA (1 mM, Sigma-Aldrich), were added to the medium to enrich for stem cells. Following this treatment, organoids were either maintained in this stem cell media, returned to standard growth media, or differentiated as previously described with the following minor modifications [17]. To increase enterocyte differentiation, organoids were grown in Wnt3a-deficient medium supplemented with VPA (1 mM) and the Wnt inhibitor IWP-2 (5μm, Stemgent). To promote goblet cell differentiation, Wnt3a-deficient medium was supplemented with the Notch inhibitor DAPT (10 μm, Stemgent) and IWP2. Following 4 additional days of treatment, organoids were harvested for RNA collection or lysed for Luminex analysis. RNA was harvested from snap-frozen colon tissue using the Qiagen RNAeasy microarray tissue mini kit (cat. 73304). Microarray was performed using Affymetrix mouse 430 2. 0 GeneChip at the Dana Farber microarray core. Microarray data were deposited in the NCBI gene expression omnibus (accession number GSE87317). The following gene sets were pulled and generated from published manuscripts: Atoh1 null, secretory progenitors (Sec-pro), and enterocytes [15]; Lgr5-positive cells [14]; and TACs [16]. All gene sets were used as published except the enterocyte gene set, which was generated using the raw data from Kim et al. Enterocyte-specific genes were defined as the transcripts that showed at least 2-fold increased expression compared to all other measured cell types (Lgr5, Atoh1 null, and Sec-pro) with an FDR < 0. 05. Gene set enrichment analysis was run using the Broad Institute’s GSEA software [39]. Mice were treated with rapamycin or vehicle daily for 2 weeks as described above. Fecal samples were harvested pretreatment and post-treatment, frozen until the end of the treatment period, and then homogenized in DNA/RNA Shield reagent (Zymo R1100-50) at 10% (v/v). DNA was extracted using the ZymoBIOMICS DNA Miniprep Kit (Zymo R2002). Generation of the 16s library, sequencing, read calling, and taxonomy assignment were performed by Zymo Research through their ZymoBIOMICS service. ZymoBIOMICS Microbial Community Standards (Zymo D6300) were used as positive controls. The library was sequenced on an Illumina MiSeq with >10% PhiX spike-in. Amplicon sequences were deduced from raw reads, and chimeric sequences were removed using the Dada2 pipeline [40]. Taxonomy assignment was performed using Qiime v1. 9. 1 [41]. PCoA and correlation analysis of Bray-Curtis dissimilarity scores were performed using MATLAB.

Chronic inflammation of the gastrointestinal track is the common defect shared by inflammatory bowel diseases (IBDs), such as Crohn’s disease and ulcerative colitis, which affect many people around the world. However, the genetic and physiologic complexities of IBDs have made it difficult to identify therapeutically tractable drivers of disease that can alleviate the symptoms. We reasoned that this complexity is probably originated by a smaller number of dysregulated signaling pathways, and therefore, a “protein-centric” approach would be more suited to identify new therapeutic targets. To this end, in this study we profiled the expression and phosphorylation status of proteins that mediate signaling between and within cells in a mouse model of colitis. We found that hyperactivated mammalian target of rapamycin (mTOR) signaling interferes with the proper differentiation of epithelial cells, which promotes colitis by altering the epithelial inflammatory cytokine secretion in the colon.

Abstract Introduction Results Discussion Materials and methods

cell motility innate immune system medicine and health sciences organoids immune physiology cytokines pathology and laboratory medicine immunology biological cultures cell differentiation colitis developmental biology signs and symptoms gastroenterology and hepatology molecular development inflammatory bowel disease organ cultures digestive system research and analysis methods inflammation biological tissue immune response chemotaxis immune system gastrointestinal tract diagnostic medicine anatomy cell biology physiology chemokines epithelium biology and life sciences colon

2018

The colonic epithelium plays an active role in promoting colitis by shaping the tissue cytokine profile

Many protein interactions are conserved among organisms despite changes in the amino acid sequences that comprise their contact sites, a property that has been used to infer the location of these sites from protein homology. In an inter-species complementation experiment, a sequence present in a homologue is substituted into a protein and tested for its ability to support function. Therefore, substitutions that inhibit function can identify interaction sites that changed over evolution. However, most of the sequence differences within a protein family remain unexplored because of the small-scale nature of these complementation approaches. Here we use existing high throughput mutational data on the in vivo function of the RRM2 domain of the Saccharomyces cerevisiae poly (A) -binding protein, Pab1, to analyze its sites of interaction. Of 197 single amino acid differences in 52 Pab1 homologues, 17 reduce the function of Pab1 when substituted into the yeast protein. The majority of these deleterious mutations interfere with the binding of the RRM2 domain to eIF4G1 and eIF4G2, isoforms of a translation initiation factor. A large-scale mutational analysis of the RRM2 domain in a two-hybrid assay for eIF4G1 binding supports these findings and identifies peripheral residues that make a smaller contribution to eIF4G1 binding. Three single amino acid substitutions in yeast Pab1 corresponding to residues from the human orthologue are deleterious and eliminate binding to the yeast eIF4G isoforms. We create a triple mutant that carries these substitutions and other humanizing substitutions that collectively support a switch in binding specificity of RRM2 from the yeast eIF4G1 to its human orthologue. Finally, we map other deleterious substitutions in Pab1 to inter-domain (RRM2–RRM1) or protein-RNA (RRM2–poly (A) ) interaction sites. Thus, the combined approach of large-scale mutational data and evolutionary conservation can be used to characterize interaction sites at single amino acid resolution. Protein activity, folding and stability are regulated by the interactions of proteins with other macromolecules. Thus, the identification of sites on a protein where these interactions occur is a critical but difficult undertaking. In some cases, structural analyses provide these sites at high resolution. In other cases, combinations of biochemical, biophysical and genetic methods with mutagenesis strategies have delineated specific residues that contribute to physical interactions. However, the vast number of protein-protein interactions and the low throughput and robustness of approaches to identify interaction sites have led to the limited and often imprecise characterization of only a tiny fraction of the contact sites. Sequence-based computational methods offer an alternative and cost-effective approach that can predict interacting positions by making use of homologous sequences. For example, the evolutionary trace method [1] assumes that the locations of interaction sites are conserved over evolution, and that sequence variation within these sites occurs in response to changes in evolutionary constraints to allow the protein to maintain its activity. Other computational methods are based on the idea that physical interaction between two proteins leads to linked evolutionary changes between their contact sites [2,3, 4]. Thus, the correlated changes between pairs of positions in multiple sequence alignments of two interacting proteins can identify binding sites [2]. However, despite improvements in the construction of multiple sequence alignments and phylogenetic trees, and the huge increase in the number of homologous sequences, the accuracy of these methods remains challenged by fundamental problems [5,6]. For example, transient interactions often yield poor evolutionary signals due to increased rates of substitutions at contact sites [7]. In consequence, these contact sites resemble other, less critical residues in the protein that also tolerate multiple substitutions. We begin with the idea that substitutions tolerated in nature usually cause only minor changes in structure [8]. Thus, if a position in a protein is substituted with an amino acid that is found at that position in homologous proteins, the resulting protein is likely still to function in its native organism. However, when such a substitution has a detrimental effect, it may have affected a functional site that has changed over evolution [9]. For a protein contact site, such a detrimental effect is likely due to the lack of other compensating substitutions also present in the homologous protein that have co-evolved to support its binding to a partner protein. Alternatively, compensatory substitutions might be present in the homologue of the protein partner. Complementation assays using a protein with such natural substitutions have been used to characterize binding site residues [10,11,12,13]. However, the utility of this approach has been limited by the lack of large-scale assays that can test a protein’s activity when it carries all the possible substitutions that occur in homologous sequences. Recently, a method known as deep mutational scanning was developed to assess the functional consequences of up to hundreds of thousands of variants of a protein in a single experiment [14,15]. This method combines next generation sequencing with a functional selection, using the change in frequency for each variant over the course of the selection as a proxy for the variant’s activity. We previously applied this method to study the in vivo function of an RNA recognition motif (RRM) of the Saccharomyces cerevisiae poly (A) -binding protein, Pab1 [16]. The eukaryotic poly (A) -binding protein regulates mRNA translation and decay [17,18,19] by binding to the poly (A) tail of an mRNA via its four RRMs [20,21]. This binding leads to an interaction between RRM2 and the translation initiation factor eIF4G, a constituent of the mRNA cap-binding complex, eIF4F [22], which is assumed to enhance the rate of translation by supporting the establishment of a closed loop structure of the mRNA [23,24,25]. Yeast encode two eIF4G paralogues, eIF4G1 and eIF4G2 [26], which both interact with Pab1 [12]. Complementation assays by Otero et al. [12] with yeast Pab1 containing residues from the human orthologue mapped the binding site for the two eIF4G isoforms to five amino acids on the surface of Pab1 RRM2 [12]. However, this study addressed only the 25 Pab1 residues in the RRM2 domain that vary between human and yeast, and thus the contribution of the other 50 RRM2 residues and the precise Pab1 contact sites for the two isoforms of eIF4G were not determined. We analyzed deep mutational scanning data for the RRM2 domain of yeast Pab1 to examine the functional consequences in yeast of single amino acid substitutions that differentiate the yeast domain from its homologues. This large-scale inter-species complementation data allowed us to characterize the eIF4G1 and eIF4G2 binding sites on the RRM2 surface at single amino acid resolution and to identify residues associated with the RRM2–poly (A) and RRM2–RRM1 interactions. By combining epistasis data for double mutants with natural variation data, we identify a humanizing substitution that promotes a change in binding specificity of the yeast Pab1 RRM2 from the yeast to the human eIF4G1 protein. Taken together, in vivo deep mutational scanning data integrated with evolutionary variation can be used to characterize interaction sites with high resolution and to predict epistatically interacting residues in natural homologues of a protein. We recently scored the in vivo function of more than 100,000 variants of the RRM2 domain of the yeast Pab1 [16]. The assay was based on turning off the expression of a wild-type copy of the PAB1 coding sequence and assaying growth of yeast dependent on mutated versions of a C-terminally truncated form (Pab1-343) that includes the first three RRM domains and a small portion of RRM4. For each variant, we assigned an enrichment score that represents the ratio between the fractions of its sequence read counts after and before selection, normalized to the wild-type enrichment score. Hence, enrichment scores serve as indirect readouts for the effects of mutations on growth rate. We obtained scores for 1246 single amino acid substitutions, including 1190 missense mutations and 56 nonsense mutations (∼83% of all possible single amino acid substitutions in the 75 amino-acid long sequence that covers most of this domain) [16]. We realized that the scores of variants with amino acid substitutions present in Pab1 homologues might provide insight into functional sites that diverged in sequence throughout the evolution of this protein. To this end, we collected sequences of 52 poly (A) -binding proteins that represent all Pab1 homologues in the UniProtKB/Swiss-Prot database. The 52 homologous sequences include both orthologues and paralogues of the poly (A) -binding protein and are derived from eukaryotic species including fungi, plants and mammals. All 52 proteins carry four tandem RRM domains, allowing us to align the Pab1 RRM2 against all its corresponding domains. The multiple sequence alignment showed conservation between the homologous RRM2 sequences and the yeast Pab1 RRM2 ranging from 88% identity for Candida glabrata to 55% identity for Encephalitozoon cuniculi. The alignment revealed 210 single amino acid differences (“natural substitutions”) with respect to the yeast Pab1 RRM2 sequence. The in vivo deep mutational scanning data from our previous study [16] provide functional scores for 197 of these 210 substitutions (Fig. 1A). Most of these natural substitutions resulted in small effects (Fig. 1B), with a median score of −0. 07 relative to the wild-type (the score, in log2 scale, is comparable to ∼5% reduction from the wild-type score) and narrow upper and lower quartiles. On the contrary, substitutions that do not appear in Pab1 homologues (“non-natural substitutions”) displayed a much larger range and more negative effects, with a median score of −0. 53 (comparable to ∼30% reduction from the wild-type score). That most natural changes result in small effects suggests that the functional constraints on the poly (A) -binding protein remained largely constant throughout its evolution. However, a few natural substitutions showed low enrichment scores that correspond to poor Pab1 performance in S. cerevisiae. In particular, enrichment scores of 45 natural substitutions ranged between −0. 15 and −0. 5 (a range that we term mildly deleterious, comparable to ∼10–30% reduction from the wild-type score) and enrichment scores of 17 other natural substitutions were lower than −0. 5 (a range that we term strongly deleterious, comparable to more than 30% reduction from the wild-type score) (Fig. 1A). We further compared the score distribution of natural variants to the score distribution of synonymous variants which serve as a proxy for non-deleterious variants, as previously described [16]. This comparison allowed us to assess the contamination of the mildly and the strongly deleterious groups by variants that carry non-deleterious mutations (S1 Fig.). Based on this analysis, we estimated that the natural substitutions in the mildly deleterious group are contaminated by 35% non-deleterious variants, while the natural substitutions in the strongly deleterious group are contaminated by only 8% non-deleterious variants. Given these results, we further analyzed only mutations classified as strongly deleterious. The solvent accessibility of residues in the structure of a human orthologue of Pab1 reveals that both natural non-deleterious and natural strongly deleterious substitutions, compared to all other non-natural substitutions, occur preferentially at solvent-exposed areas (Fig. 1C). However, an evaluation of the conservation of each substitution using its Blosum62 score revealed a significant difference between the natural non-deleterious and the natural strongly deleterious groups (Fig. 1D). Though both groups showed high conservation compared to non-natural substitutions, the natural strongly deleterious substitutions displayed a lower conservation score (median of −1) than the natural non-deleterious substitutions (median of 0). The differences in Blosum62 score distributions of the two groups suggests that natural deleterious effects in general are due to substitutions to amino acids that display physicochemical properties that are neither as disruptive as non-natural substitutions nor as subtle as natural non-deleterious ones. Nonetheless, a few natural-deleterious substitutions resulted from replacements by highly similar amino acids (e. g. L186M and L153V), indicating that sometimes the exact identity of the Pab1 residue is of crucial importance. Of the 25 single amino acid substitutions that differentiate the yeast Pab1 RRM2 domain from its human orthologue (Fig. 2A), 24 have enrichment scores in our dataset. Three of these mutations (E181R, A185K and L186M) are strongly deleterious (Fig. 2B). These three substitutions occur in two short stretches of the yeast Pab1,180-KE-181 and 184-DAL-186, that when replaced with the corresponding human stretches to create 180-ER-181 and 184-EKM-186 interfere with in vitro binding to ∼100 amino acid fragments of yeast eIF4G1 and eIF4G2 [12,22]. The large-scale mutational data indicate that the other two mutations in these short stretches, K180E and D184E, cause no measurable effect on function (Fig. 2B). To test whether the in vivo effects on Pab1 performance correlate with eIF4G1 and eIF4G2 binding, we established a two-hybrid assay between yeast Pab1 and the N-terminal 341 amino acids of yeast eIF4G1 or eIF4G2, which contain the binding sites for Pab1 [12,22]. The full-length Pab1 tested with the eIF4G1 or eIF4G2 fragment did not activate HIS3 reporter gene expression (Fig. 2C). However, as some protein-protein interactions can be detected by the yeast two-hybrid system only when parts of the proteins are removed [27], we tested various truncation products of Pab1 for eIF4G1 and eIF4G2 association. Indeed, RRM2 alone produced a positive interaction signal with both isoforms (Fig. 2C). In agreement with Otero et al. [12], the replacement of residues 184–186 with those from human resulted in complete loss of binding to both eIF4G1 and eIF4G2 (Fig. 2D). When tested individually, A185K and L186M did not bind eIF4G1 or eIF4G2, while D184E showed wild-type binding. The replacement of residues 180–181 with those from human abolished eIF4G1 binding and reduced eIF4G2 binding. This residual binding to eIF4G2 may reflect the greater sensitivity of the two-hybrid assay compared to the in vitro assay [12]. When tested individually, E181R resulted in loss of eIF4G1 and eIF4G2 binding, while K180E had no effect (Fig. 2D). Since the E181R effect on eIF4G2 binding was more severe in the absence of the K180E substitution, K180E might suppress the negative effect of the E181R mutation on eIF4G2 binding by decreasing the local positive charge. Overall, the in vivo function of Pab1 carrying any of the five single amino acid substitutions correlates with the ability of Pab1 to support eIF4G1 and eIF4G2 binding. We hypothesized that the deleterious effects of some of the other natural substitutions might be due to a loss of eIF4G1 and eIF4G2 binding. We therefore tested in the two-hybrid assay the 17 substitutions that cause a strongly deleterious effect, as well as A185K and D184W, which score similarly but had lower sequence read coverage in the original experiments [16]. Of these 19 mutations, 10 (occurring in 8 different residues) impaired the ability of RRM2 to bind eIF4G1, with I137F, T145H, T145L, V148K, E181R, A185H, A185K and L186M showing the most severe effects (Fig. 3A, left). D138T and A141D resulted in modest effects on eIF4G1 binding (Fig. 3A, left). The same Pab1 variants assayed against eIF4G2 revealed similar effects (Fig. 3A, right), suggesting that eIF4G1 and eIF4G2 use the same set of Pab1 RRM2 residues for binding. However, eIF4G1 binding was more sensitive to A141D and T145L compared to eIF4G2. Based on the effects of the natural amino acid substitutions on binding, we set the boundaries of eIF4G recognition site to the upper surface of RRM2 (Fig. 3B), a region much wider than previously identified [12]. While combining natural variation with in vivo deep mutational scanning highlights the contribution to protein-protein interactions of residues that change over evolutionary time, it overlooks highly conserved residues and ignores the effects of substitutions to amino acids that do not appear in homologues. We therefore sought to study the effects of mutations on Pab1 RRM2–eIF4G1 association by an alternative approach. To this end, we performed a large-scale two-hybrid analysis. We expressed each of three libraries of RRM2 as a DNA-binding domain hybrid, with mutations covering Pab1 positions 131–150,151–175 or 176–197, and tested for the binding of these hybrids to the yeast eIF4G1 expressed as an activation domain hybrid. Samples were collected before (input) and after (selected) two-hybrid selection, and the library segments were recovered and sequenced. For each variant, the change in its frequency from input to selected pool (i. e. its enrichment score) was determined as previously described [16]. We were able to extract enrichment scores for 802 single amino acid substitutions across the three library segments, which comprise 60% of all possible substitutions (S1 Table). While mutations that disrupt RRM2 structure caused fortuitous activation of the yeast two-hybrid reporter gene, positions that were shown to be sensitive to natural substitutions when tested individually showed similar sensitivities to mutation in this large-scale assay, suggesting that the enrichment scores for mutations that specifically affect the contact site for eIF4G1 were valid (S2 Fig.). In particular, of the 44 mutations that reduced the enrichment score by more than 50% (log2 enrichment score < −1), 22 mutations occur at the eight positions that were found by our natural variation analysis to be involved in eIF4G1 binding (I137, D138, A141, T145, V148, E181, A185 and L186); eight mutations occur at the immediate sequence neighbors of these positions (D136, S147, F149 and D184); and 11 mutations occur at residues that show physical but not immediate sequence proximity to these contact site residues (G150, G188, M189, L190 and N192). Overall, in addition to identifying eIF4G1 contact site residues that were elucidated by the combined approach of the in vivo mutational data and the natural variation data, the large-scale two-hybrid results highlighted the contribution of residues at the periphery of this site (Fig. 4A). To understand why mutations at these positions were not discovered using our combined approach, we examined the level of natural variation at these sites. While F149 and G150 are fully conserved, the other residues show some degree of variation in Pab1 homologues. Though some of these natural changes interfered with eIF4G1 binding in the two-hybrid assay, none of them showed a strongly deleterious effect in vivo (Fig. 4B), suggesting that the central residues of the eIF4G1 binding site are more sensitive to natural variation substitutions in vivo than the peripheral ones. To understand how incompatible Pab1 variants have evolved in different lineages, we constructed a maximum likelihood tree from the 52 Pab1 homologues. In agreement with theoretical expectations [28], we found that the number of substitutions in Pab1 that were strongly deleterious in S. cerevisiae increases with evolutionary distance (Fig. 5A). Specifically, while closely related fungi provide zero or one strongly deleterious substitution, the microsporidian Encephalitozoon cuniculi, which carries the most diverse PABP sequence, contributes six deleterious substitutions. The deep divergence of E. cuniculi PABP, likely due to rapid evolution of microsporidia after branching off the fungal lineage [29], provides a unique set of mutations (I137F, D138T and A141D) that interfered with eIF4G1 binding. However, unlike the metazoan substitutions that interfered with this binding, the E. cuniculi substitutions localize to helix α1 (Fig. 3B), which suggests two alternative paths of eIF4G-binding site evolution. In addition, the deleterious effects of substitutions T145L and T145H, from the non-yeast paralogues of the poly (A) binding protein (PABP5 and PABP4L), reveal the critical function of T145 in eIF4G binding. Taken together, these results highlight the need to analyze evolutionarily remote sequences in order to obtain a detailed map of functional sites in proteins. The functional scores of the natural substitutions that occurred throughout evolution suggest ancestral states that were likely to promote the divergence of the eIF4G1-binding site. In particular, for position 185, we observe a stepwise decrease in charge in the S. cerevisiae lineage, from lysine through histidine and asparagine to alanine (Fig. 5B, middle). Both A185K and A185H were strongly deleterious in yeast, suggesting that the lack of positive charge in yeast was accompanied by other changes in eIF4G or in Pab1 orthologues that are no longer compatible with the ancestral state of this position. At positions 181 and 186, substitutions matching variation within the S. cerevisiae lineage were mildly deleterious or non-deleterious, while substitutions matching variation that occurred after the fungal–metazoan divergence were strongly deleterious. Therefore, changes in eIF4G or in Pab1 orthologues that compensate for the otherwise detrimental effects of these mutations are likely to be conserved along the metazoan branch of the tree. We asked whether the yeast Pab1 and eIF4G protein sequences might enable us to infer the compensatory changes that allowed the establishment of the strongly deleterious substitutions E181R, A185K and L186M in the human orthologue of Pab1. For instance, a pair of mutations comprising one humanizing substitution in yeast Pab1 that interferes with yeast eIF4G1 association and a compensating, second humanizing mutation in the yeast eIF4G1 might restore binding. However, the identification of candidate humanizing substitutions in the yeast eIF4G1 that may form deleterious–compensatory clusters with humanizing mutations in Pab1 is challenging due to the extreme diversification of eIF4G1 and its contact site residues throughout evolution (Fig. 6A). Thus, we decided to explore the inter-protein interactions in Pab1 that underpin the binding of the RRM2 domain to either the yeast or human eIF4G1. While the human and yeast RRM2 domains interacted with their cognate eIF4G1 fragment, neither bound to its non-cognate eIF4G1 fragment (Fig. 6B), suggesting that eIF4G1 binding specificity is dependent on the 25 positions that differ between the yeast and the human RRM2 domains. We tested a few humanizing mutations in Pab1 RRM2 for their ability to change the binding specificity towards human eIF4G1. Though there are many possible combinations of humanizing substitutions, we used the deep mutational scanning results to narrow down the list of candidate residues. We first evaluated the ability of Pab1 RRM2 fragments that carry each of the three humanizing substitutions (E181R, A185K and L186M) that abolished binding to the yeast eIF4G1 to bind the human eIF4G1 fragment. The E181R variant activated the two-hybrid reporter gene (Fig. 6B), indicating that despite other sequence differences, elements within the yeast Pab1 RRM2 domain support this change in binding specificity. Unlike E181R, A185K and L186M did not bind to human eIF4G1, suggesting that these two substitutions require other humanizing changes in Pab1 RRM2 to function. Combining A185K and L186M with E181R to form a triple mutant did not enable binding of yeast Pab1 to human eIF4G1 (Fig. 6B). Because this triple mutant carries all of the strongly deleterious substitutions that differ between the human and the yeast Pab1 RRM2 domain, this finding suggests that some of the remaining mildly deleterious or non-deleterious substitutions are necessary to compensate for the detrimental effects of A185K and L186M on eIF4G1 binding. Because the deep mutational scanning of Pab1 RRM2 provided functional scores for multiple variants that change two amino acids [16], we realized that the contribution of other humanizing substitutions to the function of contact site residues might be inferred from the epistasis scores of such variants. We calculated epistasis by taking the enrichment score of a double mutant and subtracting the product of the scores of the component single mutants. Humanizing substitutions that compensate for the deleterious effects of E181R, A185K or L186M are likely to show positive epistasis (i. e. the double variant functions better than predicted) while humanizing substitutions that do not should display no epistasis. We extracted the epistasis scores for 866 double mutants ([16], S2 Table), each carrying two substitutions that are found in one of the 52 homologues of Pab1. Comparing the epistasis score distribution of these variants to that of 38,742 double mutants that carry pairs of mutations that do not occur in any of the individual homologues of Pab1 that were sampled in our analysis revealed a small yet significant increase (Wilcoxon rank sum test p-value = 3. 712e-10) in epistatic interactions between substitutions that are present together in natural variants (Fig. 6C). Thus, two mutations found in a natural protein variant are more likely to interact positively, either by synergistic or compensatory mechanisms. Of the 866 double mutants with two substitutions found in Pab1 homologues, eight carry one of the strongly deleterious humanizing substitutions together with a second humanizing mutation (S2 Table). Of these, a double mutant carrying the deleterious substitution L186M together with the non-deleterious substitution G177E had a high epistasis score (Fig. 6C). Specifically, while L186M alone resulted in ∼30% loss of in vivo function, addition of the non-deleterious G177E substitution restored Pab1 function to the wild-type level (S2 Table). G177E was able to partly restore eIF4G2 binding of an RRM2 mutant that carries the L186M substitution (Fig. 6D), suggesting that the positive epistasis of G177E and L186M is at least in part due to an improved association of the double mutant with eIF4G2. While adding G177E to the triple mutant did not shift the binding specificity towards the human eIF4G1, humanizing its adjacent residue by E176Q substitution supported this switch (Fig. 6D), suggesting that the local humanized environment of G177E is important for its function. The contribution of E176Q and G177E to human eIF4G1 binding is specific, as other groups of humanizing substitutions, found either at a distance or in close physical proximity to the three deleterious substitutions, were not able to promote this shift in binding specificity (S3A Fig.). Thus, despite the lack of measurable effects of single amino acid substitutions at position 177 of yeast Pab1 (Fig. 1A), the amino acid at this position is important for Pab1 binding to the human eIF4G1. The ancestral state of position 177 in the Pab1 lineage was glutamic acid, which was replaced by glycine in the recent ancestor of S. cerevisiae (S3B Fig.). Therefore, it is likely that the pre-establishment of glutamic acid at position 177 compensated in human for the detrimental effects of at least one of the three deleterious substitutions, while becoming dispensable in the evolutionary path that was taken by S. cerevisiae. Of the other nine natural and strongly deleterious substitutions in Pab1, five (K140A, L153V, S155V, K156N and A158E) map to the interface between RRM1 and RRM2. In particular, L153 and K156, present in the human orthologue, are key residues in the interaction between RRM1 and RRM2 that allow for efficient poly (A) binding [30]. In addition, an allosteric change in the RRM1 and RRM2 interface upon poly (A) binding is suggested to facilitate the association of RRM2 with eIF4G [31]. Therefore, deleterious substitutions in the RRM1–RRM2 contact site are likely to result from loss of either poly (A) or eIF4G binding activity, or both. Three other substitutions (Y197N, Y197V and A199E) map to the poly (A) -binding site [30]. Residue 197 is the only RNA-binding residue that is highly divergent, as all the other residues that bind RNA are either identical across the 52 homologues or display a small variation that is highly tolerated by the yeast protein. It is likely that the structure of the poly (A) forces extreme conservation on the RNA-binding residues, similar to enzyme-substrate binding sites [32], in a way that prevents useful characterization by natural substitutions. Elucidating contact sites with high resolution is important to clarify how proteins exert their functions. With respect to Pab1, we found that the binding sites for eIF4G1 and eIF4G2 extend beyond the helix α2 element [12] to include part of helix α1. The inclusion of this helix provides a plausible explanation for the molecular mechanism that couples poly (A) and eIF4G binding by Pab1. In yeast, binding of eIF4G to Pab1 requires the prior association of Pab1 with poly (A) in order to promote translation [22]. In human, these sequential steps are separated by inter-domain allostery of RRM2 and RRM1, allowing PABP1 to adopt a more extended conformation in the presence of RNA [31]. Since the association of RRM2 and RRM1 involves direct interactions between helix α1 of RRM2 and helix α2 of RRM1 [30,31], conformational changes of the two domains might make helix α1 of RRM2 and its surrounding residues available for eIF4G association upon poly (A) binding. Our finding that a Pab1 fragment consisting only of RRM1–RRM2 was unable to bind eIF4G supports the regulatory role of this inter-domain interaction in this function. eIF4G1 and eIF4G2 are functionally interchangeable under optimal growth conditions [36]. However, differences in eIF4E co-purification and in vitro translation efficiencies suggest that each of the two isoforms possesses unique roles in translation under non-optimal conditions [37,38]. Despite the overlap in location and similar mutational sensitivity of the binding sites for eIF4G1 and eIF4G2, a few Pab1 RRM2 substitutions resulted in differential sensitivities to binding. Whether this difference in Pab1 RRM2 binding points to altered mechanisms of action is a matter of further studies. T145L, which bound only to eIF4G2, might be useful in clarifying specific roles for the isoforms in translation. We identified three substitutions (E181R, A185K and L186M), corresponding to the residues present in the human PABP1, each of which when introduced into the yeast Pab1 eliminated its binding to yeast eIF4G1. We tested whether these substitutions might switch the specificity of Pab1 to bind human eIF4G1. The single humanizing substitution, E181R, allowed the yeast Pab1 RRM2 to bind to human eIF4G1, demonstrating that in spite of sequence diversification, the human and yeast orthologues of eIF4G1 and Pab1 share similarities with respect to their physical association. However, Pab1 carrying A185K and L186M did not bind to the human eIF4G1, even after humanizing the contact site by other substitutions. Thus, this shared similarity in binding activity is likely to be maintained by other intra-protein interactions in Pab1 that compensate for the otherwise deleterious effect of A185K and L186M. Our finding that two substitutions that are both present in an individual homologue are more likely to display positive epistatic interactions suggests that compensating mutations reconstruct functional modules that are conserved between organisms despite changes in the amino acid sequence that comprise these modules. Indeed, that the addition of the G177E substitution repairs the binding of an RRM2 L186M mutant to the yeast eIF4G2 suggests that the two humanizing substitutions restore a functional binding site for the yeast eIF4G2. Additional studies will be required to determine whether the tendency for positive epistasis of two substitutions present in a homologous sequence is a universal property of proteins or a specific feature of Pab1. Nonetheless, it is likely that substitutions from paralogues of closely related species are more prone to this type of epistasis than substitutions from other homologous sequences, given the functional conservation and the small number of amino acid changes in these paralogues. Additionally, G177E together with E176Q combined with the three deleterious substitutions E181R, A185K and L186M to allow yeast Pab1 binding to the human eIF4G1. This finding supports the use of epistatic interactions between two natural substitutions tested in a model organism to infer similar interactions between those residues in their natural context. We suggest that systematic integration of large-scale epistasis data with bioinformatic tools that use sequence homology might improve prediction accuracies of co-evolutionary relationships and functional association between residues. Approximately 20% of S. cerevisiae genes are essential for growth on rich glucose medium [39], with many of the remaining genes required upon environmental or genetic perturbations. Therefore, growth selections compatible with deep mutational scanning can be used to study the in vivo function of a large fraction of yeast proteins. This experimental strategy can also be applied to cross-species complementation assays to analyze human proteins in yeast [9,40,41]. However, the score assigned to each protein variant reflects the consequence of mutation only on growth rate. Therefore, inferring the direct impact of mutations on an in vivo activity such as ligand binding remains challenging. Here we show that integrating deep mutational scanning results with natural variation data provides a high throughput inter-species complementation assay that can be used to identify and characterize functional regions in proteins, including protein-protein contact sites. In addition, the large-scale analysis of natural amino acid substitutions can provide an experimental platform to evaluate the performance of computational tools that use protein homology to predict function and co-evolutionary relationships. The procedures for Pab1 RRM2 deep mutational scanning, including establishment of the experimental platform, construction of mutant libraries, sequencing of RRM2 DNA fragments and data analysis were previously described [16]. Unless otherwise indicated, only variants with input-read counts greater than 40 were used for the analysis. pOBD2 and pOAD were used to test the interactions between Pab1 and eIF4G isoforms in the yeast two-hybrid system. Full length PAB1 encoding amino acids 1–578 (DMP87) was PCR amplified from pCM188-Pab1 [16] and cloned into the NcoI and SalI sites of pOBD2. The following PAB1 truncations, encoding amino acids 1–343 (DMP88), 1–204 (DMP183), 123–204 (DMP180) and 1–120 (DMP179) were PCR amplified from p415GPD-Pab1-343BX [16] and cloned into the NcoI and SalI sites of pOBD2. PAB1 fragments encoding amino acids 123–204 (RRM2) with the point mutations I137F (DMP201), D138T (DMP202), K140A (DMP203), A141D (DMP230), T145H (DMP204), T145L (DMP189), V148K (DMP205), L153V (DMP206), S155V (DMP207), K156N (DMP208), A158E (DMP209), K180E (DMP197), E181R (DMP193), D184E (DMP188), D184W (DMP210), A185H (DMP211), A185K (DMP186), L186M (DMP185), Y197N (DMP191), Y197V (DMP194), A199E (DMP192), [K180E, E181R] (DMP198), [D184E, A185K, L186M] (DMP190), [E181R, A185K, L186M] (DMP235), [E181R, A185K, L186M, A158V, T159C] (DMP286), [E181R, A185K, L186M, E176Q, G177E] (DMP287), [E181R, A185K, L186M, V148A, K180E, D184E] (DMP291), [E181R, A185K, L186M, P135K, Q194R] (DMP292), G177E (DMP293), [G177E, L186M] (DMP297) and [E181R, A185K, L186M, G177E] (DMP298) were created by PCR using the same p415GPD-Pab1-343BX plasmid as a template and cloned into the NcoI and SalI sites of pOBD2, C-terminal and in-frame with the Gal4 DNA binding domain. eIF4G1 and eIF4G2 fragments encoding amino acids 1–341 were amplified from yeast genomic DNA (strain W-303) and cloned into the EcoRI and SalI sites of pOAD, C-terminal and in-frame with the Gal4 activation domain (DMP92 and DMP212, respectively). The human PABP1 fragment encoding amino acids 95–176 was amplified from HsCD00042197 (PlasmidID) and cloned into the NcoI and SalI sites of pOBD2 (DMP264). The human eIF4G1 fragment encoding amino acids 1–260 was amplified from HsCD00342900 (PlasmidID) and cloned into the NcoI and SalI sites of pOAD (DMP265) Yeast strain PJ694a (MATa trp1-901 leu2-3,112 ura3-52 his3-200 gal4Δ gal80Δ LYS2: : GAL1-HIS3 GAL2-ADE2 met2: : GAL7-lacZ) carrying pOBD2- and pOAD-based vectors were grown overnight, at 30°C in synthetic complete media lacking leucine and tryptophan. To test for activation of the HIS3 reporter gene, cells were spotted in a dilution series on synthetic complete plates lacking leucine and tryptophan, with or without histidine and grown at 30°C for three days. We collected 52 Pab1 homologues (see S1 File for sequences and accession numbers), representing sequences of all poly (A) -binding proteins that carry four consecutive RRM domains that can be found in the UniProtKB/SwissProt database [42], which contains high quality annotations of protein sequences. Multiple sequence alignment (MSA) was performed using Clustal Omega [43] with default parameters (S2 File). Enrichment scores for natural and non-natural single amino acid substitutions were obtained from Supplementary Table 2 of Melamed et al [16]. To assess the fraction of natural substitutions that result in impaired function, enrichment score distributions of 160 natural single amino acid substitutions, 539 non-natural single amino acid substitutions and 229 synonymous variants with input read counts greater than 500 were determined. The stringent input read count threshold was set to minimize fluctuations of enrichment scores due to low representation of variants in the library pools. The enrichment scores distribution of the synonymous variants was used as a proxy for the enrichment scores distribution of non-deleterious variants in the dataset. To estimate the fraction of deleterious substitutions within the natural substitutions, for each enrichment score bin shown in S1B Fig. , we subtracted the estimated fraction of non-deleterious variants from the fraction of the natural variants. For each single amino acid substitution, the fractional accessible surface area (ASA) was obtained for the side chains of the wild-type residue in the human PABP1 RRM2 structure (PDB ID 2K8G) using VADAR server, version 1. 8 [44]. Data for K164 residue was omitted, as this residue is absent from the human RRM2 (see Fig. 2A). The Blosum62 matrix was used to score each substitution to determine the degree of conservation. Box plots were generated using R-studio software. RRM2 sequences containing one of the three library segments were PCR amplified from the library plasmids that were previously described [16] and cloned into the NcoI and SalI sites of pOBD2. Yeast expressing the RRM2 hybrid containing one of the three libraries were grown to log phase in SC medium lacking leucine and tryptophan, supplemented with 2% glucose, and diluted into fresh medium lacking leucine, tryptophan and histidine to a final concentration of 4 × 104 cells/mL. Selection was carried out for 21 h with the culture growing to a density of 5 × 106–1 × 107 cells/mL. 2. 5 × 108 cells from each culture were collected before (“input”) and after selection (“selected”). Library preparation for high throughput sequencing was carried as previously described [16]. Amplicons were created with internal primers that flanked each library segment and carried at their 5’ end common sequencing targets for Illumina read1, read2 and index primers (11 PCR cycles) and with external primers that added Illumina adapter sequences (8 PCR cycles). Amplicons were sequenced by an Illumina NextSeq500 using paired-end reads. We used the Enrich software package (Fowler et al. 2011) to filter for high quality reads (base Q score >20). Based on the variance of enrichment scores of 2423 synonymous variants (i. e. variants that encode the wild-type Pab1 RRM2 protein sequence and carry at least one synonymous mutation), we selected variants with at least 20 input read counts for further analysis (synonymous variance <0. 4 for all three libraries). Enrichment scores of single amino acid substitutions were log2 transformed and visualized using Java TreeView 1. 1. 6r2 [45]. Average linkage hierarchical clustering with a Euclidean distance similarity metric for both RRM2 residues and substituting amino acids was performed using Gene Cluster 3. 0 [46]. A maximum likelihood tree was constructed using the Phylogeny. fr tool [47] using default parameters. Probabilities for ancestral states were calculated using the JTT model of substitution by the FastML tool [48]. Ancestral amino acids were considered “true” if their reconstruction probabilities were greater than 0. 7 (the sum of probabilities for all amino acids equals 1. 0). Otherwise, the two most probable amino acids with a minimal probability of 0. 3 for each, and sum of probabilities greater than 0. 75 were considered. The human PABP1 RRM1-RRM2 structure (PDB ID 1CVJ) was visualized using PyMol software (version 1. 5. 0. 5).

The interactions of proteins with each other are essential for almost all biological processes. Many of the sites of protein contact have evolved to maintain these interactions, but use different sets of amino acid residues. As a result, the residues at a contact site in a protein from one species might not allow a protein interaction when they are tested in a second species. This property underlies the idea of inter-species complementation assays, which test the effect of replacing protein segments from one species by their equivalents from another species. However, this approach has been highly limited in the number of changes that could be analyzed in a single study. Here, we present a novel approach that combines a high-throughput analysis of mutations in a single protein with the set of natural sequences corresponding to evolutionarily divergent variants of this protein. This integration step allows us to map at high resolution both sites of inter-protein interaction as well as intra-protein interaction. Our approach can be used with proteins that have limited functional and structural data, and it can be applied to improve the performance of computational tools that use sequence homology to predict function.

Abstract Introduction Results Discussion Materials and Methods

2015

Combining Natural Sequence Variation with High Throughput Mutational Data to Reveal Protein Interaction Sites

Tsetse flies of the Palpalis group are the main vectors of sleeping sickness in Africa. Insecticide impregnated targets are one of the most effective tools for control. However, the cost of these devices still represents a constraint to their wider use. The objective was therefore to improve the cost effectiveness of currently used devices. Experiments were performed on three tsetse species, namely Glossina palpalis gambiensis and G. tachinoides in Burkina Faso and G. p. palpalis in Côte d' Ivoire. The 1×1 m2 black blue black target commonly used in W. Africa was used as the standard, and effects of changes in target size, shape, and the use of netting instead of black cloth were measured. Regarding overall target shape, we observed that horizontal targets (i. e. wider than they were high) killed 1. 6-5x more G. p. gambiensis and G. tachinoides than vertical ones (i. e. higher than they were wide) (P<0. 001). For the three tsetse species including G. p. palpalis, catches were highly correlated with the size of the target. However, beyond the size of 0. 75 m, there was no increase in catches. Replacing the black cloth of the target by netting was the most cost efficient for all three species. Reducing the size of the current 1*1 m black-blue-black target to horizontal designs of around 50 cm and replacing black cloth by netting will improve cost effectiveness six-fold for both G. p. gambiensis and G. tachinoides. Studying the visual responses of tsetse to different designs of target has allowed us to design more cost-effective devices for the effective control of sleeping sickness and animal trypanosomiasis in Africa. Tsetse flies (Diptera: Glossinidae) infest about10 million km2 of sub-Saharan Africa where they transmit trypanosomes which cause Human African Trypanosomiasis (HAT; also known as sleeping sickness) and African Animal Trypanosomiasis (AAT; also known as Nagana). This complex of diseases has an important impact on health and economic development in sub-Saharan Africa [1], [2]. Tsetse are commonly divided into three, ecologically distinct groups: savannah tsetse (= Morsitans group) which are largely responsible for transmitting the trypanosomes that cause nagana; riverine tsetse (= Palpalis group) which play a major role the transmission of Trypanosoma brucei spp. , the causative agents of sleeping sickness; and forest tsetse (= Fusca group) which, generally speaking, do not play an important epidemiological role. Tsetse traps or their simplified two-dimensional derivative targets, when impregnated with insecticides, have constituted a central component of tsetse control campaigns in many countries in Africa [3]–[6], albeit such baits have been more used against AAT than HAT, except for a few notable exceptions [7], [8]. The reasons it has not been used more widely against HAT are several, but one of the most important is the financial and logistical cost of using baits [9]. Hence, if the method is to be more widely used, especially by communities directly afflicted by HAT, then these costs must be reduced [10]. The type of target used to control tsetse varies according to the geographical location of the operation and the target species of tsetse. However, in general targets are coloured blue and/or black [11], [12]. The use of blue in combination with contrasting colours such as white or black significantly improves landing behaviour of tsetse on targets [11], [13], [14], [15]. The shape of the target is also important for both the overall shape (horizontal versus vertical) and the patterns (e. g. banding) on the target. For example, vertical banding seems to be more effective than horizontal for some Palpalis group tsetse (e. g. G. p. palpalis, see [16]). For Morsitans group tsetse, horizontal oblongs elicit a stronger landing response than vertical ones, whereas vertical and horizontal oblongs seem equally attractive [17], [18]. Interactions between size and efficacy are also variable. While it is generally acknowledged that for the Morsitans group tsetse of East and South Africa, “the bigger the target the better” [19], [20], this is not the case for Palpalis group tsetse [11], [21], [22]. For example, recent studies in Kenya showed that a 90% reduction in target size only reduced the catch of G. f. fuscipes by 50% [22]. In this latter study, as well as other recent works on tsetse in West Africa (I. Tirados et al. , In Press; J. Esterhuizen et al. , In Press), size and shape comparisons always involved black targets. However in West Africa, the most used target to control tsetse has been the 1 m2 black-blue-black target developed by Laveissière et al. [11]. Consequently in this study, we investigated if this target could be improved in terms of size, shape and overall design, focusing on three major vectors of human and animal trypanosomiases in West Africa: G. palpalis gambiensis, G. p. palpalis, and G. tachinoides. To achieve this, taking into account previous observations on tsetse behaviour in particular their circling behaviour around targets [14], [19] and the poor ratio reported of the numbers landing compared to the numbers attracted [23], we designed several experiments to examine how (i) adding netting panels and/or varying (ii) the size and/or (iii) overall shape could improve the number of tsetse that might be attracted to the vicinity of a target, and then the proportion that subsequently contact it, either by landing on the target or colliding with it. Actually we show that smaller designs made of overall horizontal oblongs incorporating netting offer promising, more cost effective, alternatives to control tsetse of the Palpalis group and that these improved designs may lead to more sustainable control efforts because of their better cost efficiency than in the past. G. p. gambiensis and G. tachinoides were studied in southern Burkina Faso between November 2008 and April 2009, near the village of Folonzo (9. 9°N, 4. 6°W) along the Comoe river (January – April) where G. tachinoides is predominant. We also worked in western Burkina Faso on the Mouhoun river near the village of Solenzo (12. 20°N, 4. 4°W) in November, where only G. p. gambiensis is found. The detailed description of these study areas can be found in [24], [25]. Work was also undertaken on G. p. gambiensis in Orodara (11°18N, 5°27W) (South West of Burkina Faso) on the Pindia river. In Côte d' Ivoire the study was undertaken near Azaguié (05. 67°N, 04. 11°W) where G. p. palpalis is abundant. The numbers of tsetse attracted to the vicinity of a target was assessed by covering the targets with a grid of fine electrocuting wires which killed or stunned tsetse as they landed [23]. A proportion of tsetse approaching targets do not land and to provide a relative estimate of these circling tsetse, an electrocuting net (E-net) was placed adjacent to the target. The E-net is effectively invisible to tsetse and hence circling flies collide with it. The electric targets and E- net were mounted on a tray. Tsetse contacting the electrocuting grid fall vertically into the water tray below. For instance, in the target represented in figure 1, the tray is divided in three parts to allow the separation of flies killed by contacting the blue or the two flanking nets. The combined catch from the target+E-net provided a relative measure of the number of tsetse attracted to a target. Many different designs of target were studied in different experiments. To facilitate inter-experiment comparisons, a “standard” target was included in all experiments. The standard design, which is used as a control, is derived from the Laveissière et al. (11] black-blue-black target, and consisted of a 1 m2 target (1 m wide×1 m high) with three, vertical stripes of black, blue and black in widths of relative proportion 1∶2∶1, respectively (Fig. 2). Henceforth, this design will be called “the standard” in the rest of the manuscript. This design has been widely used in tsetse control operations across West Africa (see [7] for instance). The sequence of these stripes of black, blue, and black that will be found in other treatments tested here will be called “BkBlBk” in the rest of the manuscript. When the black cloth sections of a target were replaced by black netting, the letter ‘N’ was used (e. g. NBlN). Flies contact a target either by landing on the cloth or when they collide with netting panels which are invisible to them; we wanted to explore which was the best strategy to pursue in target design. All experiments were carried out for 4 h between 08∶00 hours and 12∶00 hours local time when Palpalis group species are most active [26]. The different targets were compared with the standard target in a series of replicated Latin squares of days×sites×treatments where sites were always >100 m apart. The daily catches (n) were normalized and variances homogenized using a log10 (n+1) transformation and subjected to analysis of variance using GenStat 11 edition (version 11. 1. 0. 1504). When the ANOVA showed a significant difference after this first analysis, a Bonferroni pairwise comparison test was undertaken to detect significant differences between the different treatments. To provide a common index of the effect of shape, size or colour on catches, the detransformed mean catch of tsetse from different targets was expressed as the proportion of that from the standard target and this value was termed the catch index. For example, a target that doubles the catch from the standard target would have a catch index of 2 while one that halved the catch would have a catch index of 0. 5. Differences in the proportion of tsetse caught on different sections (e. g. , blue, black or netting) of a target were analysed by logistic regression. The total catch from all sections of the target was specified as the binomial denominator and the catches from a particular section (e. g. , blue section) was specified as the y-variable. The significance of changes in deviance was assessed by either chi2 or, if the data were overdispersed, an F-test following re-scaling [27]. The standard error is asymmetric about the mean and thus mean percentages are accompanied by the larger standard error. Unless stated otherwise the term ‘significant’ denotes that the means differ at P<0. 05. Details of the individual experiments are given below. Seven targets were compared, including the standard (treatment A), to determine the effect of the overall shape of the target (i. e. horizontal (wider than high) versus vertical (higher than wide) ) on fly captures. The 6 targets were compared by pairs as following, with the first number indicating the width and the second one, the height (in meters): 0. 5×0. 75 net-blue-net (NBlN) vertical (Treatment B) versus 0. 75×0. 5 NBlN horizontal (C); 0. 25×0. 5 black-blue-net (BkBlN) vertical (D) versus 0. 5×0. 25 BkBlN horizontal (E); 0. 25×0. 5 NBN vertical (F) versus 0. 5×0. 25 NBN horizontal (G). In all these different treatments, proportions of the different parts were the same, e. g. treatments B and C, and F and G are made of 50% blue, 50% net, whereas black, blue and net are 25%, 50% and 25% respectively, in treatments D and E. This experiment was conducted only in Folonzo (Burkina Faso), for both G. p. gambiensis and G. tachinoides. In this experiment, we wanted to know if smaller designs were more cost efficient than the standard. Hence three different sizes of targets, 0. 75×0. 5 (treatments C, L and M), 0. 25×0. 5 (F and H) and 0. 25×0. 25 (I and K) were investigated and compared to the standard. The experiment was undertaken on G. p. gambiensis in Orodara, Solenzo and Folonzo (Burkina Faso), on G. tachinoides in Folonzo, and on G. p. palpalis in Azaguié (Côte d' Ivoire), respectively with BkBlBk and NBN cloths. Only the treatment M was in BkBlN. The experiment was designed to determine if the black cloth could be replaced by the less expensive netting and, second, if it was necessary to place netting on both sides of the target or only one. This experiment also was performed on the 3 sites in Burkina and in Azaguié in Côte d' Ivoire. The cost efficiency ratio of the devices was calculated and was expressed as the catch per unit area of one device divided by catch per unit area of the standard. Let us assume a target of 1 m2 catches 100 tsetse and a new design of 0. 1 m2 also catches 100. So the tsetse/m2 for each target is 100 (100/1 = 100) and 1000 (100/0. 1 = 1000). The improvement in cost efficiency would be viewed as being 1000/100 = 10 - ie we get 10x more tsetse per dollar/CFA/euro spent. Horizontal-oblong targets caught consistently more tsetse than the vertical ones for both G. p. gambiensis and for G. tachinoides (see tables 1 and 2), for the three pairs which were compared (treatment C vs B, E vs D, G vs F). Results showed catch indices according to the ratio horizontal/vertical of 1. 6,3. 5 and 4. 5 respectively, with the two latter being significant at P<0. 01 and P<0. 001 respectively for G. p. gambiensis (Table 1). For G. tachinoides, these ratios were 1. 60,3. 1 and 5 respectively (Table 2), these differences being highly significant for the two latter pairs (P<0. 001) but not for the first pair in both sexes. For all three species in all the study areas, catches were highly correlated with size (P<0. 001). As an example, in Solenzo while the catches of the standard were up to 18 flies/day for G. p. gambiensis, catches for targets H (0. 25×0. 5) and I (0. 25×0. 25) (both BkBlBk like the standard) were nearly 20 times less than the standard (p<0. 001) (see Table 3). This ratio was almost the same with the netting treatments F (0. 25×0. 5) and K (0. 25×0. 25) where captures decreased to almost 0 tsetse/day, both for males and females G. tachinoides in Folonzo (see Table 2), and for G. p. gambiensis in Folonzo (Table 1) and Solenzo (Table 3). The same pattern was also observed in Côte d' Ivoire for G. p. palpalis (table 4) where fly densities were high. One exception to this general trend was the 0. 75×0. 5 target which gave catch indices up to more than 0. 8 (compared to 1 for the standard) for G. p. gambiensis (treatment C in table 1, treatments L, C, and M in Table 5) and up to 1. 23 for the horizontal NBlN (treatment C) for G. tachinoides in Folonzo (table 2). For this particular size, there was no significant difference compared to the standard, whatever the colour combination. For G. p. gambiensis in Folonzo and in Solenzo (tables 1 and 3), there was no significant difference in the catch of the 1×1 NBlN (J) compared to the standard. For the same species in Orodara (table 5), no difference appeared when comparing the three types of 0. 75×0. 5 targets (BkBlBk, NBlN, BkBlN) in terms of global catches (respective mean catches of about 20) and individually, none of them was significantly different from the standard (catch indices ≈0. 9). For G. tachinoides, the NBlN (0. 75×0. 5 (C) and 1×1 (J) ) caught more than the standard with respective indices of 1. 2 and 1. 4, although these differences were not significant (table 2). For G. p. palpalis it appeared that the 1×1 NBlN (J) was significantly better than the standard (p<0. 001) with an index of 2. 4 (see Table 4). The general trend as regards landing behaviour is illustrated by experiment reported in table 5, where the following treatments were compared to the standard: (L) 0. 75×0. 5 BkBlBk, (C) 0. 75×0. 5 NBlN and (M) 0. 75×0. 5 BkBlN. All the targets had one blue coloured section and we compared the percentage of tsetse caught on that section. There was no significant difference in the landing responses of males and females and so the data were pooled. For all treatments, the blue section covered half the total area of the target but for only L was the catch (49%±3. 0) close to the expected (50%). For all other treatments, the catches on the blue section were significantly less (20%±2. 3,30%±2. 7, and 37±2. 9 for A, C and M respectively). Thus the majority of tsetse were contacting either the the netting or black sections. The targets offering the best cost efficiency ratios for the three tsetse species studied are listed in table 6. For G. p. gambiensis the highest cost efficiency ratios were obtained with the horizontal 0. 5×0. 25 BkBlN (E) with a cost efficiency of 2. 8, then with the 0. 75×0. 5 NBlN (C) and 0. 75×0. 5 BkBlN (P) both horizontal, with cost efficiency ratios respectively of 2. 37 and 2. 35. For G. tachinoides, the horizontal 0. 75×0. 5 NBlN (C) was the most cost-effective with a ratio of 3. 28, followed by the 0. 5×0. 25 BkBlN (E) and NBlN (G) with respective ratios of 2. 48 and 2. 4. For G. p. palpalis the best was the vertical 0. 5×0. 25 BkBlBk (H) with a cost efficiency ratio of 2. 4. Hence higher cost efficiency ratios (2–3) were achieved with targets smaller than the standard except for G. p. palpalis, and in all cases, including G. p. palpalis, it is worth noting the highest cost efficiency ratios were obtained with targets incorporating netting. Horizontal targets performed consistently better than vertical ones for both G. p. gambiensis and G. tachinoides in Burkina Faso. This confirms results previously reported for savannah flies of the Morsitans group in East and South Africa [17], [18]. However, this contrasts with observations on another species of the Palpalis group, G. p. palpalis, which has been reported to be more attracted to vertical rather than horizontal targets in forested area of Ivory Coast ([11], and I. Tirados et al. , In Press.). Hence there appears to be consistent differences, even between species of the same group (i. e. Palpalis group here), regarding visual attraction to given shapes. The biological explanation is not precisely known so far, and may include several factors such as habitat, visibility, and/or feeding behaviour. It would be interesting to know if other tsetse species living in forest habitat would react like G. p. palpalis or not. We observed that catches were highly correlated with target size for the species studied, but only up to a maximum size of 75 cm wide. Hence when the minimum size threshold of ∼50 cm2 was reached, increasing the size of the target did not increase catches, as illustrated by the similar results of the horizontal 0. 75×0. 5 m target compared to the 1×1 m standard. These results are in general accordance with observations made for Morsitans group species in East and southern Africa, except that increasing size beyond 50 and 75 cm does not improve the catch. For G. pallidipes, attraction improved several fold as the width of visual panels increased from 25 to 200 cm, and the percentage of tsetse landing on visual panels before flying round increased up to several times with wider panels [19]. Recent studies in southern Africa also reported the correlation between size and catches for Morsitans group species [18]. Again here for size, as for shape above, knowing the exact cause of these differences between species is not obvious. Presumably it relates to the habitat or hosts of the flies. For instance it is generally acknowledged that savannah flies of the Morsitans group feed mainly on wild and domestic mammals, which are generally bigger than the reptiles which constitute the main diet of tsetse such as G. fuscipes [29]. G. palpalis and G. tachinoides are regarded as being opportunistic in their feeding habits, feeding on mammals, including humans, as well as reptiles. They may have developed a particular ability to detect small hosts. Although the three designs of horizontal target (0. 75×0. 5 BkBlBk, BkBlN and NBlN) were not significantly different regarding the total tsetse catches, replacing the black cloth by the black net consistently increased, albeit slightly, the catches of the different species. This confirms earlier observations of flies trying to avoid landing immediately on the blue cloth and then hit the “invisible” net when circling around the target [17]. The strongest landing response was elicited by blue and black targets, as reported previously [11], [14]. Although the difference was not always significant in our work in Orodara, males G. p. gambiensis were more attracted to the blue cloth while females landed more on the black. Differences in landing behaviour between males and females had been reported previously for studies of G. p. palpalis in Côte d' Ivoire [11], although in this study, as well as the one of Green [14], the reverse was observed, i. e. the proportion of males landing on black was greater than the one of females. Regarding shape, the different trials showed that the horizontal target was always better when compared to the vertical for both G. tachinoides and G. p. gambiensis. When considering size, most flies were caught with the 0. 75×0. 5 cm targets, and also by the 0. 5×0. 25 cm. However especially for the latter, such small targets might be hidden by the dense vegetation found in humid savannahs and forested areas. More studies are underway to see if this is the case. Using the flanking netting gave very promising results for the three species studied. Hence one could make a practical choice about the cost and durability of using netting vs. cloth in the construction of targets. It is noteworthy that Laveissiere et al. [11] also found that using net resulted in more catches than using black cloth. They however did not recommend its use for targets due to the local price of netting in Ivory Coast at that time. However, given that the price of black net is currently estimated to be 1/3 that of black cloth (T. Frandsen, pers. com.), changing the design of the BkBlBk 1 m2 target to a 0. 75×0. 5 NBlN target (see fig. 1) would increase cost efficiency by six fold (3 fold for cloth surface plus 3 fold for price of net vs black) without losing efficiency. It would have the further practical advantage of being effective for both G. p. gambiensis and G. tachinoides which are the main vectors of pathogenic trypanosomes to humans and domestic animals in West Africa, and which are found often together in savannah areas. For G. p. palpalis in Ivory Coast, although new insights have been brought by this study regarding their visual behaviour related to size of targets and the use of netting, it would be interesting to expand such studies to other countries where this tsetse species occurs.

Tsetse flies transmit trypanosomes causing sleeping sickness and nagana. Controlling tsetse prevents transmission of these diseases. Insecticide impregnated targets are highly effective but are too costly. This study aims to improve the cost effectiveness of targets. Experiments were performed on three tsetse species in Burkina Faso or Côte d' Ivoire. Effects of target size, shape, and the use of netting instead of black cloth were measured. We observed that targets wider than they are high (horizontal target) killed 1. 6-5x more G. p. gambiensis and G. tachinoides than vertical ones. Catches were highly correlated with the size of the target up to a target size of 0. 75 m, beyond which there was no further increase in catches. Replacing the black cloth of the target by netting did not change catches, but was far cheaper. Hence reducing the size of the current 1 m×1 m black-blue-black target to an horizontal 0. 75×0. 5 m net blue net target will improve cost effectiveness six-fold for both G. p. gambiensis and G. tachinoides. Studying the visual responses of tsetse to different designs of target has allowed us to design more cost-effective devices for the effective control of sleeping sickness and animal trypanosomiasis in Africa.

Abstract Introduction Methods Results Discussion

medicine infectious diseases african trypanosomiasis neglected tropical diseases

2011

Towards an Optimal Design of Target for Tsetse Control: Comparisons of Novel Targets for the Control of Palpalis Group Tsetse in West Africa

Fragment-based drug discovery using NMR and x-ray crystallographic methods has proven utility but also non-trivial time, materials, and labor costs. Current computational fragment-based approaches circumvent these issues but suffer from limited representations of protein flexibility and solvation effects, leading to difficulties with rigorous ranking of fragment affinities. To overcome these limitations we describe an explicit solvent all-atom molecular dynamics methodology (SILCS: Site Identification by Ligand Competitive Saturation) that uses small aliphatic and aromatic molecules plus water molecules to map the affinity pattern of a protein for hydrophobic groups, aromatic groups, hydrogen bond donors, and hydrogen bond acceptors. By simultaneously incorporating ligands representative of all these functionalities, the method is an in silico free energy-based competition assay that generates three-dimensional probability maps of fragment binding (FragMaps) indicating favorable fragment∶protein interactions. Applied to the two-fold symmetric oncoprotein BCL-6, the SILCS method yields two-fold symmetric FragMaps that recapitulate the crystallographic binding modes of the SMRT and BCOR peptides. These FragMaps account both for important sequence and structure differences in the C-terminal halves of the two peptides and also the high mobility of the BCL-6 His116 sidechain in the peptide-binding groove. Such SILCS FragMaps can be used to qualitatively inform the design of small-molecule inhibitors or as scoring grids for high-throughput in silico docking that incorporate both an atomic-level description of solvation and protein flexibility. Fragment-based drug discovery relies on a simple premise: identify small-molecule fragments that bind to a target region of the protein and then evolve or link them to create a larger high-affinity molecule. To a first approximation, the binding free-energies of fragments bound in non-overlapping poses are additive [1]. Therefore, linking two such fragments with millimolar affinities (4 kcal*mol−1) will yield a single molecule with micromolar affinity (8 kcal/mol), which is of sufficient affinity to serve as a “hit” for lead optimization [2]. Since the chemical space spanned by small fragments is orders of magnitude smaller than that spanned by molecules of sufficient size to be hits, it becomes feasible to screen a fragment library representative of the full extent of chemical space [3]. Nature imposes an upper limit on the contribution per ligand heavy atom to the binding free-energy [4], commonly referred to as “ligand efficiency” (LE) [5]. This limit means that even the best fragments (LE 0. 4–0. 5 kcal*mol−1 per heavy atom [3]) still have weak affinities for their targets, making their screening by traditional assays difficult. Consequently, fragment-based drug discovery relies on sensitive biophysical methods to detect fragment binding. Among these methods are NMR spectroscopy (“SAR-by-NMR”) [6] and x-ray crystallography [7]. These two methods additionally benefit from the fact that they yield structural information about fragment binding poses, which is useful for confirming that two fragments indeed bind to two adjacent sites and can be productively linked. Despite their utility, there are significant time, labor, and materials costs associated with experimental fragment-based drug discovery approaches. Computational approaches to fragment-based drug discovery hold out the promise of mitigating the costs of experimental fragment-based drug discovery. Currently, in computational approaches the protein is assumed to be rigid and fragments sample the surface of the rigid protein using an energy function that models the solvent environment as a continuum [8]–[12]. As a result, these methods are limited in their ability to accurately account for protein conformational heterogeneity and solvation effects, contributions that are essential to compute free energies of binding [13]. In reality, proteins can accommodate ligands by undergoing conformational changes [14], [15], and water plays an important role in protein∶ligand binding affinity [16]–[18]. Significant advances have been made toward incorporating protein flexibility, for example by screening against multiple different rigid protein conformations [19]–[21], and toward more accurate modeling of solvation effects in energy functions [22]. Nonetheless, approximations used in computational approaches to date can still limit the accuracy of fragment placement and scoring, and, ultimately, the determination of the most suitable fragment for a selected region of the protein. All-atom explicit-solvent molecular dynamics (MD) simulations of proteins give an atomic-level-of-detail description of the motions of both protein and water atoms at relevant temperature and pressure [23]. MD samples a Boltzmann distribution of thermally accessible protein conformations, and with the ability of MD to reach the nanosecond timescale, the sampled conformations can include changes in sidechain dihedral angles as well as loop motions. Furthermore, MD simulation-based methods are able to determine the absolute binding free energy of a ligand to a protein to, in the best cases, within RT of the experimental value [15], [24]–[32]. However, such MD free-energy calculations are computationally expensive, limiting MD simulations from being used directly for high-throughput in silico screening. Toward overcoming present limitations in fragment-based computational drug design we describe a new method that combines ideas from experimental fragment-based drug discovery with all-atom explicit-solvent MD. The method (SILCS: Site Identification by Ligand Competitive Saturation) involves computationally immersing a protein in an aqueous solution simultaneously containing different types of small molecules, with each at a concentration of ∼1 M. The protein+small molecule+water system is then subjected to multiple MD simulations allowing for competitive binding of the small molecules to the protein. Snapshots from the MD trajectories are combined to generate 3D probability maps (FragMaps) that reveal what types of functionalities bind most strongly to different parts of the protein surface. Because they are generated from MD simulations, SILCS FragMaps incorporate both protein mobility, with a Boltzmann distribution of conformations, and atomic-level solvation effects, thereby yielding FragMaps that represent rigorous free energy distributions. Notably, the method requires minimal time, labor, and materials compared to experimental approaches. As a test case, SILCS FragMaps were generated for the BTB domain of the BCL-6 oncoprotein [33], [34]. The SILCS FragMaps, from MD simulations initiated using the BCL-6 conformation in the BCL-6∶SMRT protein∶peptide cocrystal, recapitulate the pattern of aliphatic, aromatic, hydrogen bond donor, and hydrogen bond acceptor interactions seen at the BCL-6∶SMRT cocrystal interface. Additionally, these same FragMaps also recapitulate the interaction pattern seen in the BCL-6∶BCOR protein∶peptide cocrystal, which has important differences arising from sequence and structure variation in the C-terminal halves of the SMRT and BCOR peptides. Furthermore, the simulations sample the BCL-6 His116 sidechain conformation seen in the BCL-6∶BCOR cocrystal, a conformation that is required for hydrogen bonding with BCOR Ser508 and significantly different from that in the SMRT-bound BCL-6 MD starting conformation, emphasizing the ability of the presented approach to account for protein flexibility. The majority of moieties on drug-like molecules that target proteins fall into four classes: aliphatic, aromatic, hydrogen bond donor, and hydrogen bond acceptor. This reflects the relatively limited chemical diversity of amino acid sidechains. Salt bridges between two amino acids are a special case of hydrogen bonding, since the interaction is never directly between two charged heavy atoms but between a negatively-charged oxygen and the proton on a positively-charged nitrogen. Fragment libraries generated from existing drugs and drug-like molecules reflect this limited diversity, being largely composed of hydrogen bond donors consisting of amides, hydrogen bond acceptors of carbonyls and ethers, hydrophobic groups of small-length aliphatic chains, and aromatic/cyclic groups of benzene [35]. The first goal in the choice of small molecules for use in this initial implementation of SILCS was to minimize the set of fragments so as to be able to maximize their individual concentrations, which in turn maximizes binding and helps convergence on the MD timescale. To this end, a minimalist small-molecule set was selected that contains hydrophobic aliphatic moieties, aromatic moieties, hydrogen bond donors, and hydrogen bond acceptors. Propane was chosen to represent hydrophobic aliphatic groups because the termini are small enough to fit into cavities only large enough to accommodate a methyl group, while the molecule itself is large enough to disrupt the hydrogen bonding structure of water so as to induce strong hydrophobic binding [36]. Additionally, unlike longer-chain alkanes, propane is essentially a rigid body excepting the rotation of the two terminal methyl groups, and thus convergence of internal degrees of freedom is not an issue. Benzene was selected to represent aromatic groups as it occurs in over 40% of drug-like compounds and is four times more common than the next most-common aromatic moiety [35]. Finally, water was used as a small molecule that contains both hydrogen bond donating and accepting capabilities. Water is at a concentration of 55 M in solution and also has no internal conformational degrees of freedom, again promoting convergence on the MD timescale. Other small-molecule possibilities for hydrogen bond donors and acceptors include acetone, formaldehyde, and small amides, but these would necessarily be at much lower concentrations than water, hindering convergence. Additionally, they contain several different functionalities, such as the methyl groups in acetone and the combined hydrogen bond donor and acceptor moieties in an amide, which can make binding analysis more difficult. The second goal in a choice of small molecules for use in SILCS was to minimize their sizes to maximize convergence, both by facilitating reversible binding on the MD timescale and allowing for fast diffusion through the bulk solvent. Even with a high-ligand efficiency, i. e. 0. 4 kcal*mol−1 per heavy atom, fragments consisting of 3–6 heavy atoms will have binding affinities of only 1. 2 to 2. 4 kcal*mol−1 (100 millimolar to 10 millimolar). While such weak binding affinity can be a liability in an experimental approach as it may push the limits of detection, it is an asset in the SILCS approach, allowing for ligand exchange from a binding site on the MD timescale, facilitating the implementation of a competitive in silico binding assay. Another benefit of molecules having only 3–6 heavy atoms is that their high diffusion rates lead to quick mixing and rapid translation to different regions of the protein surface. Thus, small molecules of minimal molecular size are beneficial both because of rapid binding exchange with the protein and rapid diffusion around the protein. It should be emphasized that the SILCS approach is amenable to a wide range of fragment-like small molecules. The fragment molecules selected for the present study were chosen for computational expediency, as proof-of-principle, and because they represent a minimal set that includes aliphatic, aromatic, hydrogen bond donor, and hydrogen bond acceptor moieties. Larger fragments and/or fragments with a greater diversity of functional groups may prove useful in developing a more fine-grained classification of preferred functionalities beyond simply aliphatic, aromatic, hydrogen bond donor, and hydrogen bond acceptor. These may encompass different types of hydrogen bonding groups such as ethers, amides, amines, esters, and carbonyls, as well as heterocyclic aromatics and molecules with halides, to name some possibilities. To ensure binding of small low-affinity molecules, a high concentration (∼1 M) of each small molecule is used in the simulations. However, a simulation of a solution of 1 M propane and 1 M benzene in water is prone to severe hydrophobic aggregation, as seen in the intermolecular carbon…carbon (C…C) radial distribution function g (r), which traces the relative probability of observing this pair of atoms at a given separation distance. The C…C g (r) has a large peak at 5 Å (Figure 1A, “no repulsion”), associated with the distance between carbons in two fragments that are in direct contact. This trace slowly decays with increasing distance, reflecting the fact that in an aggregate in water, it is much more likely to have two hydrophobic fragments adjacent to each other than at a larger separation. Such aggregation drastically reduces the effective concentration of the fragments, which in turn hampers sampling of the protein surface and prevents SILCS FragMap convergence. Because SILCS is a computational approach, it is possible to modify the interactions between hydrophobic/aromatic fragments to prevent aggregation. This can be done by introducing a repulsive interaction energy between fragments that comes into effect only when two fragments come closer than a given interaction distance. This repulsive interaction energy is only applied to selected fragment∶fragment interactions, while all fragment∶water, fragment∶protein, water∶water, water∶protein, and protein∶protein interactions remain unperturbed. For convenience, the repulsive interaction is implemented using the Lennard-Jones force field term [37] by adding an additional massless particle to the geometric center of each benzene molecule and the central carbon of each propane molecule. These particles serve as interaction sites for the inter-fragment repulsive interaction energy. Lennard-Jones parameters (ε = −0. 01 kcal/mol; Rmin = 24. 0 Å) combined with a switching function [38] operating between 5 Å and 8 Å yield an energy vs. distance profile that is purely repulsive (Figure 1B). With this additional repulsive interaction energy in effect, even at very high concentrations the small molecules will not aggregate. Thus, in the simulation of 1 M propane and 1 M benzene in water, the g (r) contact peak at 5 Å disappears, indicating the lack of direct intermolecular C…C contacts, and the flat g (r) trace at larger distances indicates a homogeneous distribution of molecules in solution (Figure 1A, “with repulsion”). In principle, such a repulsive term can make hydrophobic fragments that associate with the protein surface compete unphysically with other directly adjacent hydrophobic fragments. For example, the form of the repulsive potential (Figure 1B) will prevent the formation of a stacked benzene dimer in a binding pocket. It will, however, allow for two benzene molecules to simultaneously bind unimpeded in two adjacent pockets on the protein surface. The BTB domain of the BCL-6 protein was chosen as a test case for the SILCS method because of several favorable properties. The first is that it has two-fold symmetry, with two identical symmetry-related binding sites [33], allowing for measuring convergence of fragment sampling by analyzing the two-fold symmetry in the SILCS FragMaps. A second reason is that the binding of native ligands to the two binding sites shows no cooperativity [33]; thus, the binding sites are independent of each other and the occupancy of one site will not affect the occupancy of the other. A third reason is that two known ligands for BCL-6, SMRT and BCOR, are peptides 17 amino-acids in length that bind in extended conformations to the same groove over a large contact-area [33], [34], allowing for comparison of FragMaps over a large portion of the protein. Fourth, there is thermodynamic data available from competition assays using single-residue alanine or glycine-substituted analogs of these two peptides for every position in each peptide. Fifth, SMRT and BCOR have different binding modes in the BCL-6 peptide-binding cleft and lack sequence similarity. The different binding modes include BCL-6 sidechains in the binding cleft assuming different conformations in the presence of SMRT vs. BCOR. And finally, BCL-6 has clinical importance because of its association with diffuse large B-cell lymphoma, and competitive inhibitors that bind to the BCL-6 peptide-binding cleft may have therapeutic applications. Convergence of the SILCS FragMaps was facilitated by the selection of propane, benzene, and water as the “fragments, ” by the use of ∼1 M propane and benzene concentrations, and by combining results from 10 independent 5-ns SILCS MD simulations (see Methods). The two-fold symmetry of the BCL-6 protein with its two symmetric binding sites and non-cooperative binding allows for using two-fold symmetry in the FragMaps as a measure of convergence. Analysis of the separate 5-ns simulations shows them to yield somewhat different FragMaps that do not have exact two-fold symmetry (not shown); however, FragMaps generated as the ensemble average of all ten 5-ns simulations do exhibit the expected symmetry. To visualize the extent of convergence, slices of the aliphatic carbon atom FragMap from propane along with the protein molecular surface were taken perpendicular to the two-fold symmetry axis of the protein. These slices clearly demonstrate the expected two-fold symmetry in the FragMap, and hence convergence (Figure 2). Similarly converged results are seen for the aromatic carbon atom FragMap generated by mapping benzene carbon atoms and the hydrogen bond donor and acceptor FragMaps generated by mapping water molecules (Figures S1, S2, and S3). To more rigorously evaluate the extent of convergence, difference maps were obtained by subtracting FragMaps based on half the MD simulation data from those based on the other half. This was done for each type of FragMap by creating one map from five 5-ns simulations, a second from the remaining five 5-ns simulations, and then subtracting the first map from the second. For fully converged results these difference maps would have bin counts of zero for all volume elements (i. e. fragment atom counts in 1 Å×1 Å×1 Å cubic volume elements as described in Methods). Presented in Figure 3 are the frequency distributions of bin counts from the FragMaps (solid red) and the difference maps (dashed green), as well as bin count cutoff values used for the visualization of isosurfaces (see below) for the four fragment types. The difference map distributions are all centered around zero as expected for random errors, while the distributions from the FragMaps are all non-negative and have much higher bin counts. The difference distributions, with the exception of the aliphatic distribution (Figure 3A), go to zero below the cutoff value used for visualization, demonstrating convergence between the two data sets. In the case of the aliphatic difference map, the bin count at the isovalue cutoff is only 6% of that for the actual FragMap. These results indicate that while the FragMaps are not fully converged, the extent of convergence is adequate to identify regions of high probability for the different fragment types, which is ultimately the goal of the SILCS approach. Further, the difference map analysis shows that the different sets of SILCS simulations are generating the same affinity pattern for fragment molecules. Because each of the ten SILCS simulations was started with a different random ordering of fragment molecules on a cubic grid (see Methods), the similarities between the FragMap data from the grouping into two sets of five simulations likely reflect convergence as opposed to redundant unconverged results. SILCS FragMaps were compared with the crystal structures of the BCL-6∶SMRT and BCL-6∶BCOR complexes to validate the method' s ability to identify known binding interactions. FragMaps overlaid on the BCL-6∶SMRT and BCL-6∶BCOR structures are shown in Figures 4 and 5: Figure 4 focuses on interactions with the peptide backbones, while Figure 5 focuses on the C-terminal regions of the peptides, which contain the majority of the thermodynamically important interactions between the peptides and BCL-6 [34]. FragMap isosurfaces for hydrogen bond donors are in blue, hydrogen bond acceptors in red, aliphatic carbons in green, and aromatic carbons in purple, with the sites of discussion emphasized using arrows of the same color. BCL-6 binding interactions conserved between the non-homologous SMRT and BCOR peptides are exclusively hydrogen bonding interactions with the peptide backbones [34], and the hydrogen bond donor and acceptor FragMaps show these conserved interactions. Starting from the N-termini of the two peptides, the backbones of SMRT Ala1416 and Val1418, and BCOR Ser499 and Ile501 act as hydrogen bond acceptors, and of SMRT Val1418 and Glu1420, and BCOR Ile501 and Ser503 as donors, all of which are recapitulated by high-probability regions in the corresponding FragMaps (Figure 4A). Toward the middle of the peptides, high-probability regions overlap with SMRT Glu1420 and BCOR Ser503 as hydrogen bond acceptors to BCL-6, and SMRT Ser1424 and BCOR Ser507 as donors (Figure 4B). Finally, at the C-termini, hydrogen-bond acceptor FragMap overlap is observed with SMRT His1426 and Pro1429 as well as BCOR Trp509 and Pro512, while hydrogen-bond donor FragMap overlap is seen for SMRT Ile1428 and BCOR Val511 (Figure 4C). The only peptide backbone hydrogen bonding interactions for SMRT not detected by SILCS are at the ends of the peptide, namely Ala1416 as a hydrogen bond donor and Ile1428 as a hydrogen bond acceptor, which may be explained by the high crystallographic temperature factors of these residues [33]. In the case of the BCOR peptide, only the Ser508 backbone is not detected as a strong hydrogen bond donor. Thus, in eighteen out of twenty-one cases, high probability regions in SILCS FragMaps recapitulate the location of both SMRT and BCOR peptide backbone hydrogen bonds to BCL-6. More interesting than the conserved backbone hydrogen bonds are the non-conserved interactions involving sidechains from the C-terminal ends of the two peptides. These C-terminal amino acids have large contact areas and buried surfaces, correlating with these residues contributing most strongly to the peptide binding affinities, as measured by competitive fluorescence polarization titrations involving SMRT or BCOR peptides that have single amino acid substitutions to either alanine for non-alanine residues or glycine for alanine residues [34]. To be considered useful, the SILCS method should be capable of recapitulating these important interactions. The SILCS FragMaps capture every one of the thermodynamically important C-terminal sidechain interactions of the SMRT peptide with BCL-6. In the SMRT peptide, Arg1423, Ser1424, Ile1425, Asp1427, Ile1428, and Pro1429 in the C-terminal half make large contributions to the binding affinity [34]. Analysis of the crystal structures shows that the sidechains of Arg1423, Ser1424, and Asp1427 all form hydrogen bonds to BCL-6, while both the Ile1425 and Ile1428 aliphatic sidechains are buried in hydrophobic pockets. High-probability regions in the hydrogen-bond donor FragMap overlap with the polar hydrogens in the Arg1423 and Ser1424 sidechains, and high-probability regions in the hydrogen-bond acceptor FragMap overlap with the oxygens in the Ser1424 and Asp1427 sidechains (Figure 5A). High-probability regions in the aliphatic carbon FragMap encompass both the Ile1425 and Ile1428 sidechains (Figure 5A). Interestingly, only the Ile1428 sidechain and not the Ile1425 sidechain is also overlapped by a high-density region in the aromatic carbon FragMap. The lack of observable aromatic carbon FragMap density coincident with the Ile1425 sidechain occurs on both sides of the BCL-6 protein, and decreasing the isovalue threshold by half continues to yield no observable density on one side and only two small points of observable density on the other side that are overwhelmingly enveloped by the aliphatic carbon FragMap contour. This suggests that the Ile1428 pocket can accommodate both aliphatic and aromatic carbons, while the Ile1425 pocket will preferentially bind aliphatic carbons. Experimental evidence to this effect exists in the form of a crystal structure of BCL-6 with a small-molecule inhibitor, in which an aromatic moiety binds in the Ile1428 pocket (G. Privé, personal communication). Such differentiation emphasizes the ability of the SILCS method to account for the subtle energetic contributions that dictate the binding of different classes of hydrophobic moieties. Pro1429 is interesting in that it is the only amino acid in the C-terminal region of the SMRT peptide that makes a large experimental thermodynamic contribution to binding yet whose sidechain is not involved in an interaction with the BCL-6 protein. Rather, its backbone carbonyl acts as a hydrogen bond acceptor, and this interaction is indeed seen in the corresponding FragMap (Figures 4C and 5A). This result indicates that Pro1429Ala mutation likely has a strong affect on the SMRT binding affinity due to an increase in conformational entropy and the fact that proline occupies the extended region of φ/ψ space while alanine preferentially occupies the helical region [39]. The SILCS FragMaps also capture the thermodynamically important interactions of the BCOR peptide C-terminal residues 508–512 (Figure 5B). These include sidechain interactions for Ser508, Trp509 and Val511. Surprisingly, no overlap is seen for the Val510 sidechain with a high-density region in either the aliphatic or aromatic FragMaps. This may have an explanation similar to that for SMRT Pro1429, in that the Val510Ala mutation may cause a decrease in binding affinity due to replacement of an amino acid that prefers an extended conformation with the helix-promoting alanine. Finally, as with the homologous SMRT Pro1429, the BCOR Pro512 backbone overlaps with a high-density region in the hydrogen bond acceptor FragMap while no sidechain overlap is seen (Figures 4C and 5B). Because SILCS uses all-atom explicit-solvent MD simulations, protein flexibility is naturally included. As observed crystallographically, there are important differences in the conformations of BCL-6 sidechains in the peptide-binding groove between the BCL6 apo, BCL-6∶SMRT and BCL-6∶BCOR crystal structures. For example, BCL-6 Arg24 sidechain dihedral angles have significantly different values in crystal structures of the unliganded protein, the BCL-6∶SMRT complex, and the BCL-6∶BCOR complex, while the BCL-6 His116 sidechain undergoes a dramatic rearrangement between the SMRT and BCOR complexes. SILCS simulations seeded with a single BCL-6 structure capture this heterogeneity in both Arg24 (Figure S4) and His116, and can therefore inform the design of inhibitors targeting such flexible binding sites. The SILCS MD behavior of the BCL-6 His116 sidechain is especially relevant because of the large crystallographically-determined conformational change required in this sidechain for BCL-6 to accommodate both the SMRT and the BCOR peptides. SILCS MD samples both the His116 sidechain conformation observed in the BCL-6∶SMRT crystal structure used to initiate all the SILCS simulations, and the very different conformation in the BCL-6∶BCOR crystal structure (Figure 6). In the SILCS MD, the His116 sidechain reversibly shifts between the initial, BCL-6∶SMRT conformation (Figure 6A, purple) and a second conformation. In this second conformation, His116 forms a hydrogen-bonding complex with a water molecule that acts as a hydrogen bond donor to the sidechain and as an acceptor to the His116 backbone amide NH group (Figure 6A, colored by atom type), a complex not possible in the initial conformation due to the location of the sidechain. Furthermore, this MD second conformation is the same as in the BCL-6∶BCOR crystal structure and enables hydrogen bonding between BCL-6 His116 and the BCOR Ser508 sidechain hydroxyl in the BCL-6∶BCOR crystal structure (Figure 6B). The Ser508 hydroxyl donates a hydrogen bond to the His116 sidechain and accepts a hydrogen bond from the His116 backbone amide NH group (Figure 6B) in the same manner as the water molecule in the simulation (Figure 6A). Importantly, the hydrogen bond donor and acceptor FragMaps show that these are high-probability (favorable free energy) interactions. These results demonstrate the ability of SILCS to include protein flexibility and the ability of the method to identify locations of favorable interaction sites on the protein surface that arise from protein flexibility. The conformational changes that SILCS can take into account are, naturally, related to the timescales of the MD simulations and of the conformational changes themselves. The present results suggest that readily-accessible timescales can account for the conformational heterogeneity in biologically important surface-exposed sidechains, although in situations with, for example, strong sidechain hydrogen bonding or large structural changes like loop opening, this may not be the case. Described is a new computational method that combines ideas from experimental fragment-based drug discovery with all-atom explicit-solvent molecular dynamics. The SILCS (Site Identification by Ligand Competitive Saturation) method, by using all-atom explicit solvent molecular dynamics, incorporates atomic-level solvation effects and protein mobility. The resulting 3D free energy-based probability distributions (FragMaps) suggest the optimal placement of aliphatic hydrophobic, aromatic, hydrogen-bond donor, and hydrogen-bond acceptor functionalities in a binding pocket. As an example, SILCS FragMaps computed for the BCL-6 oncoprotein do an excellent job of reproducing the binding interactions of the non-homologous SMRT and BCOR peptides with the BCL-6 protein and include biologically relevant conformational changes in the binding pocket. SILCS FragMaps, when visualized as isosurfaces in conjunction with a protein (e. g. Figures 4–6), may potentially be used to guide the development of inhibitors at a particular site on the protein surface. The FragMaps contain information about protein flexibility and atomically-detailed solvation effects as they impact fragment binding. Additionally, the relative importance of interactions is represented by the values of the histogram counts in the 3D FragMap histograms, thus inhibitors can be optimally designed by targeting overlap with high-probability regions in the FragMaps. This can be done in an interactive, qualitative fashion, for example by informing the extension of small-molecule binders with known binding poses into larger, higher-affinity molecules that encompass nearby high-probability regions. Alternatively, this can be done in an automated, quantitative manner by taking the natural logarithm of the probabilities and multiplying by –RT; the resultant free-energy maps can be used as docking grids for high-throughput in silico docking of drug-like compound libraries, with an additional map of the protein atoms incorporated into a penalty function to account for steric clash between docked compounds and the protein. With this latter approach, some care must be taken regarding the direct interpretation of FragMaps in terms of free energies due to alterations to the chemical potential of bulk water, which is used to generate hydrogen bond donor and acceptor maps, arising from the high concentration of fragments. Additionally, some care may be required to delineate mutually-exclusive high-probability regions arising from protein conformational heterogeneity. Nonetheless, the use of SILCS free-energy FragMaps as docking grids has the potential to be a significant improvement over current high-throughput in silico methods, which are limited in their descriptions of protein flexibility and solvation [13]. Finally, an important part of the SILCS method is its computational feasibility. Each 5-ns SILCS simulation of BCL-6 took less than three days on a single 2×4-core node of a commodity computing cluster, and because each of the ten simulations was independent, they were all run simultaneously to yield converged FragMaps in under three days. The ability to achieve converged FragMaps probability maps in such a short time is a very important result, since MD simulations are often limited by the computational cost for simulations beyond the nanosecond regime, which in turn limits their utility in computer-aided drug discovery [13]. The experimental BCL-6 protein conformation from the BCL-6∶SMRT complex [33] [PDB ID 1R2B] was used to seed all SILCS MD simulations. The Reduce software [40] was used to place missing hydrogen positions and to choose optimal Asn and Gln sidechain amide and His sidechain ring orientations. Propane and benzene molecules were placed on a square grid, with the identity of the molecule at each grid point randomly determined. Ten such grids were generated with the grid spacing selected to yield a concentration of ∼1 M propane and ∼1 M benzene when combined with a box of water molecules at the experimental density of water. Ten protein+small molecule+water systems were generated by overlaying the coordinates of the BCL-6 protein and water molecules from the BCL-6∶SMRT co-crystal structure with each of the ten different solutions, removing all water, propane, and benzene molecules that overlapped the protein, and replacing two random water molecules with chloride ions to give a net neutral system charge. The final systems were rectangular boxes of size 72×58×43 Å to accommodate the protein with maximum dimensions of 64×48×35 Å. Harmonic positional restraints with a force constant of 1 kcal*mol−1*Å−2 were placed on all protein atoms and the system was minimized for 500 steps with the steepest descent algorithm [41] under periodic boundary conditions [37]. Molecular dynamics simulations were performed on each minimized system using the “leap frog” version of the Verlet integrator [37] with a 2-fs timestep to propagate the system. The SHAKE algorithm [42] was applied to constrain bonds to hydrogen atoms to their equilibrium lengths and maintain rigid water geometries, long-range electrostatic interactions were handled with the particle-mesh Ewald method [43] with a real-space cutoff of 8 Å, a switching function [38] was applied to Lennard-Jones interactions in the range of 5 to 8 Å, and a long-range isotropic correction [37] was applied to the pressure for Lennard-Jones interactions beyond the 8 Å cutoff length. With the positional restraints still in place, the system was heated to 298 K over 20 ps by periodic reassignment of velocities [44], followed by 20 ps of equilibration at 298 K, also using velocity reassignment. After the heating and equilibration periods, the positional restraints were replaced by restraints on only protein backbone Cα positions with a very weak force constant of 0. 01 kcal*mol−1*Å−2 so as to prevent rotation of the protein in the rectangular simulation box. Each system was subsequently simulated for 5 ns at 298 K and 1 atm, with the Nosé-Hoover thermostat [45], [46] and the Langevin piston barostat [47], for a total of 50 ns of simulation time. All simulations were done with the CHARMM molecular simulation software [48], the CHARMM protein force field [49] with CMAP backbone correction [50], and the TIP3P water model [51] modified for the CHARMM force field [52]. FragMaps were prepared for each SILCS simulation by binning atoms from SILCS MD snapshots taken at 2-ps intervals into 1 Å×1 Å×1 Å cubic volume elements of a grid spanning the entire system. For the aliphatic and aromatic carbon FragMaps, carbon atoms for propane and benzene molecules, respectively, were binned if they were within 5 Å of the protein. For the hydrogen bond donor and acceptor FragMaps, water hydrogen and oxygen atoms, respectively, were binned if they were within 2. 5 Å of the protein. For each type of FragMap, the respective FragMaps from each of the ten simulations were added together to create a single FragMap. A single isocontour value resulting in optimal visualization was empirically chosen for each FragMap type, and this value was used to generate all isocontour molecular graphics for that FragMap type. The ratio of the isocontour value to the average cubic volume element occupancy in an equilibrated system consisting of only propane, benzene, and water molecules was 9. 8 for propane carbons, 9. 8 for benzene carbons, 1. 3 for water hydrogens, and 1. 1 for water oxygens. Visualization of FragMaps and preparation of molecular graphics were done with VMD [53].

Fragment-based drug discovery is based on a simple yet powerful principle: instead of trying to screen through the vast number of possible drug-like compounds during the drug discovery process, screen representative drug-like fragments, which are far fewer in number. Once a suitable fragment is discovered, it can then be built up or linked with other fragments to give a drug-like molecule. Because such fragments are small, even “good” fragments bind weakly to their targets, therefore requiring significant time, labor, and materials costs for experimental detection and characterization of binding. In the present work, we describe a computational approach to the problem of detecting and characterizing fragment binding. Importantly, the method provides atomic-resolution results and also explicitly takes into account the effect that molecular water has on binding and the inherent flexibility of protein targets. The methodology is demonstrated by application to the BCL-6 protein, which is implicated in a variety of cancers, is conceptually easy to understand, and can yield results in a matter of days using present-day commodity computers.

Abstract Introduction Results/Discussion Methods

computational biology/molecular dynamics biochemistry/biomacromolecule-ligand interactions biochemistry/theory and simulation biophysics/theory and simulation

2009

Computational Fragment-Based Binding Site Identification by Ligand Competitive Saturation

p38 has long been known as a central mediator of protein kinase A (PKA) signaling in brown adipocytes, which positively regulate the transcription of uncoupling protein 1 (UCP-1). However, the physiological role of p38 in adipose tissues, especially the white adipose tissue (WAT), is largely unknown. Here, we show that mice lacking p38α in adipose tissues display a lean phenotype, improved metabolism, and resistance to diet-induced obesity. Surprisingly, ablation of p38α causes minimal effects on brown adipose tissue (BAT) in adult mice, as evident from undetectable changes in UCP-1 expression, mitochondrial function, body temperature (BT), and energy expenditure. In contrast, genetic ablation of p38α in adipose tissues not only markedly facilitates the browning in WAT upon cold stress but also prevents diet-induced obesity. Consistently, pharmaceutical inhibition of p38α remarkably enhances the browning of WAT and has metabolic benefits. Furthermore, our data suggest that p38α deficiency promotes white-to-beige adipocyte reprogramming in a cell-autonomous manner. Mechanistically, inhibition of p38α stimulates the UCP-1 transcription through PKA and its downstream cAMP-response element binding protein (CREB), which form a positive feedback loop that functions to reinforce the white-to-beige phenotypic switch during cold exposure. Together, our study reveals that inhibition of p38α is able to promote WAT browning and confer metabolic benefits. Our study also indicates that p38α in WAT represents an exciting pharmacological target to combat obesity and metabolic diseases. White adipose tissue (WAT) and brown adipose tissue (BAT) are two major types of adipose tissues, which play different physiological roles in whole-body energy homeostasis [1,2]. The main function of WAT is to store excess energy as triglycerides (TGs) for utilization during nutrient shortage and to produce bioactive adipokines—such as leptin, adiponectin, and resistin, which take part in glucose and lipid metabolism [3]—while BAT dissipates the chemical energy stored in TGs as heat to preserve core temperature through uncoupling of fatty acid oxidation from ATP production by uncoupling protein 1 (UCP-1) during hypothermia [4]. BAT was thought to function primarily in rodents and in newborn babies until functional BAT was discovered in adult humans [5,6]. The BAT identified in human adults might consist of not only classic brown adipocytes but also brown-like adipocytes (beige adipocytes). Similar to classic brown adipocytes, beige adipocytes display multilocular lipid droplet morphology, have high mitochondrial content, and express UCP-1, although they differ from classic brown adipocytes in their origin and molecular identity [7,8]. It has been indicated that the beige adipocytes interspersed among white adipocytes in rodents are able to alleviate cold stress to restore thermal homeostasis [8]. Due to the same ability to convert fat into heat through uncoupled respiration as brown adipocyte, beige adipocyte has been also considered an attractive target to promote weight loss. Indeed, promoting the development and formation of beige adipocytes in WAT, also called the browning of WAT, increases energy expenditure, prevents diet-induced obesity, and improves glucose metabolism in rodents [9,10], while suppressing WAT browning leads to obesity and insulin resistance [11,12]. It is worth noting that beige adipocytes have been shown to contribute to systemic energy handling even at room temperature (RT) [13]. Given that beige and brown adipocytes have many distinguishing characteristics [14], it is probable that the regulation of the thermogenic program differs in beige and brown adipocytes, which has yet to be studied. p38 mitogen-activated protein kinases (MAPKs) are key mediators in cellular responses to extracellular stimuli. p38 MAPKs play critical roles in a wide variety of cellular processes such as proliferation, differentiation, regeneration, and metabolism [15–19]. The p38 family of proline-directed serine/threonine kinases has 4 members (p38α, β, γ, and δ), each encoded by individual genes. p38α is highly abundant in most cell types, while p38γ and p38δ have more restricted expression patterns [20,21]. It has been proposed that p38 family members function in a cell context–specific and cell type–specific manner to integrate signals that affect cellular processes [22,23]. p38 has long been known as a central mediator of cAMP/protein kinase A (PKA) signaling, which positively regulates the transcription of UCP-1 in brown adipocytes by phosphorylating activating transcription factor 2 (ATF2) directly, a member of the cAMP-response element binding protein (CREB) /ATF family of transcription factors [17,24,25]. The in vitro effect of p38 on the thermogenic program in brown adipocytes has been well established [17,24–26]; however, the physiologic role of p38 during cold exposure has never been validated by employing a loss-of-function genetic approach in mice. Moreover, whether cAMP/PKA/p38/ATF2 cascade plays a similar role in beige adipocytes is largely unknown. Here, we show that mice lacking p38α in adipose tissues exhibited a lean phenotype and improved metabolism. To our surprise, adipocyte-specific deletion of p38α using the aP2-recombinase (Cre) line caused minimal effects on the morphology of interscapular brown adipose tissue (iBAT), the UCP-1 expression in iBAT, mitochondrial function, and body temperature (BT), as well as oxygen consumption and carbon dioxide production in adult mice. Interestingly, we found that genetic ablation of p38α in adipose tissues not only facilitated WAT browning upon cold stress but also prevented diet-induced obesity. The effect of adipocyte-specific p38α deficiency on WAT browning was subsequently verified by using the Adipoq-Cre line. Consistently, pharmaceutical inhibition of p38α promoted the browning of WAT and had beneficial effects. Further study revealed that p38α deficiency promoted white-to-beige adipocyte reprogramming in a cell-autonomous and cell type–specific manner. Mechanistically, suppression of p38α in WAT could stimulate the UCP-1 transcription through the PKA/CREB pathway. Our study indicates that p38α in WAT represents an exciting pharmacological target to combat obesity and metabolic diseases. To investigate the role of adipocyte p38α in vivo, we generated adipocyte-specific p38α knockout (Fp38αKO) mice using the Cre-lox system (p38αf/f; aP2-Cre+/–). As controls, floxed p38α (Floxed) mice that did not express Cre recombinase were used. As expected, p38α protein expression was greatly reduced in the iBAT, inguinal white adipose tissue (iWAT), and epididymal white adipose tissue (eWAT) of Fp38αKO mice as compared to Floxed mice (Fig 1A–1C, Fig A-C in S1 Fig, S1 Data). Accordingly, the protein levels of p-p38 were markedly decreased in iBAT, iWAT, and eWAT of Fp38αKO mice, suggesting that the p38 signaling in adipose tissues was greatly impaired (Fig 1A–1C, Fig A-C in S1 Fig, S1 Data). Also as expected, the protein levels of p38α were not changed in the liver and skeletal muscle of Fp38αKO mice as compared to Floxed mice (Fig 1D and 1E, Fig D and E in S1 Fig, S1 Data). We also determined the p38α protein levels in the macrophages to see whether aP2 Cre-mediated deletion of p38α could be detected in this cell type. Consistent with previous studies [27–29], we did not observe any decrease in p38α expression in intraperitoneal macrophages derived from the Fp38αKO mice (Fig 1F, Fig F in S1 Fig, S1 Data). Accordingly, the amount of either macrophages or neutrophils was not different in the peripheral blood between Floxed and Fp38αKO mice (Fig G in S1 Fig, S1 Data). Fp38αKO mice were fertile and displayed normal energy intake and excretion when maintained at RT (Fig 1G, S1 Data). Growth curve analysis for body weight (BW) revealed that adult Fp38αKO mice had reduced BW compared to age-matched Floxed mice at RT (Fig 1H and 1I, S1 Data). The results of body composition analysis suggest that the lean phenotype of Fp38αKO mice might be due to the reduction in fat mass (FM), since lean mass (LM) was not affected (Fig 1I and 1J, Fig H in S1 Fig, S1 Data). The hematoxylin-eosin staining (HE staining) of iWAT and eWAT revealed that the size of adipocytes was smaller in Fp38αKO mice compared to Floxed mice at RT (Fig I-L in S1 Fig, S1 Data). Consistent with the lean phenotype, both glucose tolerance and insulin sensitivity were increased in Fp38αKO mice (Fig 1K and 1L, S1 Data). Similar results were obtained in a glucose tolerance test (GTT) when the glucose dose was adjusted on the basis of LM (Fig M in S1 Fig, S1 Data) [30]. Moreover, the levels of glucose and TGs in Fp38αKO mice were lower than those in Floxed mice at RT (Fig N in S1 Fig, S1 Data). Since p38α has been shown to act as a central regulator of cAMP/PKA signaling and controls the transcription of UCP-1 in brown adipocytes [17,25], we speculated that the Fp38αKO mice would have reduced BT and/or altered energy expenditure. To our surprise, we did not observe any significant changes in BT, oxygen consumption, or carbon dioxide production in Fp38αKO mice compared to Floxed mice maintained at RT (Fig 2A–2C, S1 Data). To test whether loss of p38α in adipose tissue would affect cold-induced adaptive thermogenesis, we exposed Fp38αKO mice to a cold environment for 2 d. The change of BW after 2 d of cold exposure was not different between Floxed and Fp38αKO mice (Fig A in S2 Fig, S1 Data). However, the difference in BT between Fp38αKO and Floxed mice still could not be detected after cold challenge for 2 d (Fig B in S2 Fig, S1 Data), suggesting that p38α deficiency in adipose tissues could not impair the adaptations to cold exposure. Consistently, the weight of iBAT, the histological morphology of iBAT, and the size of adipocytes in iBAT from Fp38αKO mice appeared indistinguishable from Floxed mice either at RT or after 2 d of cold exposure (Fig 2D–2F, Fig C-E in S2 Fig, S1 Data). The difference in iBAT morphology and the size of adipocytes in iBAT still could not be detected between Floxed and Fp38αKO mice after 7 d of cold exposure (Fig F and G in S2 Fig, S1 Data). The staining results of UCP-1 in iBAT were similar in Floxed and Fp38αKO mice maintained at RT (Fig 2G). Electron microscopy images of iBAT of Fp38αKO mice at RT revealed that the mitochondrial content and morphology were not altered (Fig 2H). The ratio of mitochondrial DNA (mitDNA) to nuclear DNA (nuDNA) was not different in iBAT between Floxed and Fp38αKO mice either maintained at RT or after 2 d of cold exposure (Fig 2I, Fig H in S2 Fig, S1 Data). The oxygen consumption rate (OCR) of the isolated iBAT mitochondria was comparable between Floxed and Fp38αKO mice after 2 d of cold exposure (Fig 2J, S1 Data). Consistently, the mRNA levels of those genes related to mitochondria function were either not or only slightly altered in the iBAT of Fp38αKO mice at RT or after 2 d of cold exposure, compared to Floxed mice (Fig I and J in S2 Fig, S1 Data). These results suggest that the mitochondrial function was not affected in the iBAT of Fp38αKO mice. In agreement with the above findings, the mRNA levels of UCP-1 and other thermogenic genes were not changed in the iBAT of Fp38αKO mice compared to Floxed mice either at RT or after 2 d of cold exposure (Fig 2K, S1 Data). Additionally, the mRNA expression of those genes involved in fatty acid metabolism was either not or only slightly altered in the iBAT of Fp38αKO mice compared to Floxed mice after 2 d of cold exposure (Fig K in S2 Fig, S1 Data). Accordingly, the protein levels of UCP-1 were also not changed in the iBAT of Fp38αKO mice compared to Floxed mice either at RT or after 2 d of cold exposure (Fig 2L and 2M). Interestingly, the protein levels of p-ATF2 were slightly decreased in the iBAT of Fp38αKO mice compared to Floxed mice, but the differences did not quite reach statistical significance (Fig 2M, Fig L in S2 Fig, S1 Data). Since quantitative real-time PCR results revealed that p38α and p38β are the most abundant isoforms, and the expression of p38δ is very low in mouse iBAT (Fig M in S2 Fig, S1 Data), we determined the protein levels of p38β and p38γ in the iBAT of Fp38αKO mice. We found that the protein abundance of both p38β and p38γ was similar in iBAT between Floxed and Fp38αKO mice (Fig N and O in S2 Fig). We also measured the protein levels of tyrosine hydroxylase (TH), a marker of sympathetic innervations in iBAT of Fp38αKO mice, and found that the TH protein levels were not changed in iBAT from Fp38αKO mice compared to Floxed mice either at RT or in a cold environment for 2 d, indicating that sympathetic outflow was comparable between Floxed and Fp38αKO mice (Fig P in S2 Fig). To examine whether Fp38αKO mice were unable to maintain BT upon acute cold exposure, mice were exposed to cold for 4 h. However, the difference in BT between Fp38αKO and Floxed mice still could not be detected after 4 h of cold challenge (Fig 2N, S1 Data). Interestingly, after an acute cold challenge for 4 h, a significant decrease in mRNA levels of UCP-1 and peroxisome proliferative activated receptor gamma coactivator 1α (PGC1α) was observed in iBAT from Fp38αKO mice compared to Floxed mice (Fig 2O, S1 Data), indicating that there is a defect in iBAT of Fp38αKO mice. Although the mRNA expression of UCP-1 was decreased in the iBAT of Fp38αKO mice, the UCP-1 protein levels were compensatorily increased in the iBAT of Fp38αKO mice after the acute cold exposure, compared to Floxed mice (Fig 2P and 2Q, S1 Data). We also analyzed the mRNA expression of thermogenic genes and genes involved in fatty acid metabolism in the iBAT of these mice. We found that the mRNA levels of positive regulatory domain containing 16 (PRDM16), deiodinase 2 (DIO2), and elongation of very long chain fatty acids (FEN1/Elo2, SUR4/Elo3, yeast) -like 3 (ELVOL3) —as well as adipose triglyceride lipase (ATGL), monoglyceride lipase (MGL), and hormone-sensitive lipase (HSL) —were all elevated in the iBAT of Fp38αKO mice after the acute cold exposure compared to Floxed mice, suggesting that the transcription of other thormogenic genes and lipolysis-related genes was compensatorily increased during acute cold exposure (Fig 2R and 2S, Fig Q in S2 Fig, S1 Data). In addition, the nonesterified fatty acid (NEFA) levels were also comparable between Floxed and Fp38αKO mice either maintained at RT or after acute cold exposure (Fig R in S2 Fig, S1 Data). There were no differences in the creatine kinase activity in serum, gastrocnemius (GAS) muscle, and heart between Floxed and Fp38αKO mice (Fig S in S2 Fig, S1 Data). These results suggest that the supply of fatty acids in Fp38αKO mice was normal. Since increased browning of WAT has been observed in many knockout mouse models that show improved metabolism, the finding of improved metabolism in Fp38αKO mice prompted us to investigate the browning of WAT in these mice without apparent alterations in BAT function. Interestingly, profoundly increased browning was observed in iWAT from Fp38αKO mice exposed to a cold environment for 2 d, as indicated by significantly increased emergence of multilocular adipocytes, reduced adipocyte size, increased vascular density, increased expression of UCP-1 and other thermogenic or beige adipocyte genes, and increased ratio of mitDNA to nuDNA (Fig 3A–3G, S1 Data). In contrast to iBAT, iWAT showed significantly decreased p-ATF2 levels in Fp38αKO mice compared to Floxed mice after 2 d of cold exposure (Fig 3H and 3I, S1 Data). The levels of p-CREB (Ser133) were increased in the iWAT of Fp38αKO mice after 2 d of cold exposure, which might contribute to the up-regulation of UCP-1 in these mice (Fig A in S3 Fig). Since it is widely assumed that the browning of WAT can increase energy expenditure, we speculated that the energy expenditure would be affected in Fp38αKO mice after the induction of WAT browning. As expected, increased energy expenditure was observed within 48 h after β3-adrenoceptor agonist (CL316,243) injection in Fp38αKO mice compared to Floxed mice (Fig 3J and 3K, Fig B in S3 Fig, S1 Data). Similar results were obtained in mice exposed to a cold environment for 7 d before analysis (cold-adapted mice) (Fig C in S3 Fig, S1 Data). These results further indicate that deletion of p38α in adipose tissues facilitates WAT browning upon cold stress. Increased emergence of multilocular adipocytes, reduced adipocyte size, and increased expression of UCP-1 and other thermogenic genes were also observed in the iWAT from Fp38αKO mice exposed to cold for 7 d compared to Floxed mice (Fig 3L–3O, S1 Data), further suggesting that lacking p38α in adipose tissues could lead to an increase in WAT browning upon cold stress. In addition, distinct histological morphology and smaller adipocyte size; increased expression of thermogenic genes, including UCP-1; and increased ratio of mitDNA to nuDNA were observed in the iWAT from 5-wk-old Fp38αKO mice maintained at RT, which was indicative of enhanced browning (Fig 3P, Fig D-F in S3 Fig, S1 Data). Accordingly, smaller adipocyte size was also observed in eWAT from 5-wk-old Fp38αKO mice compared to age-matched Floxed mice (Fig G-H in S3 Fig, S1 Data). To test whether the increased browning of iWAT observed in adult Fp38αKO mice upon cold exposure and 5-wk-old Fp38αKO mice maintained at RT was due to an increase in sympathetic input, we measured the protein levels of TH in iWAT of these mice. We found that the protein levels of TH in iWAT were not altered in adult Fp38αKO mice compared to Floxed mice upon cold stress for 2 d (Fig 3D). Similarly, we did not detect any changes in TH protein levels in iWAT from 5-wk-old Fp38αKO mice maintained at RT compared to age-matched Floxed mice (Fig I in S3 Fig). These results indicate that the increased browning of iWAT observed in cold-exposed adult Fp38αKO mice or 5-wk-old Fp38αKO mice maintained at RT was independent of sympathetic action. It has been reported that different adipocyte-specific Cre lines displayed different degrees of efficiency and specificity. Another Cre line driven by the mouse adiponectin promoter regions within the brown adipocyte cell line (BAC) transgene (Adipoq-Cre) was bred to p38αf/f mice to generate Ap38αKO mice. As expected, p38α protein expression was greatly reduced in iBAT and iWAT of Ap38αKO mice (Fig J and K in S3 Fig). Consistent with the observation in Fp38αKO mice, the UCP-1 levels were elevated in iWAT of cold-exposed Ap38αKO mice (Fig K in S3 Fig), further supporting the notion that ablation of p38α in adipose tissues could promote WAT browning in the adaptive response to cold environments. To see whether the adipogenesis was affected in the iWAT of Fp38αKO mice after cold exposure, in vivo BrdU-labeling experiments were performed. We found that the proportion of BrdU-positive adipocytes in iWAT was not different between Floxed and Fp38αKO mice after cold challenge for 7 d (Fig L and M in S3 Fig, S1 Data), suggesting that the adipogenesis was not affected in the iWAT of Fp38αKO mice. Additionally, we determined the mRNA levels of p38 isoforms in mouse iWAT and found that p38α and p38β are the most abundant isoforms in mouse iWAT, and the expression of p38δ is very low in mouse iWAT (Fig N in S3 Fig, S1 Data). We then examined the protein levels of p38β and p38γ in the iWAT of Fp38αKO mice and found that the protein expression of both p38β and p38γ was not altered in Fp38αKO mice either maintained at RT or after 2 d of cold exposure compared to Floxed mice (Fig O and P in S3 Fig). According to a previous study [31], the total protein of adipose tissues would increase after cold exposure for 7 wk, especially for iBAT. To take this into consideration, we quantitatively determined the total protein content in iBAT and iWAT, and UCP-1 content per mg protein, then calculated the total UCP-1 content per depot in both genotypes. We found that total protein content in iBAT and iWAT was comparable between Floxed and Fp38αKO mice either maintained at RT or exposed to cold for 2 d (Fig 3Q and 3R, S1 Data). The total UCP-1 content per depot for iBAT was similar in both Floxed and Fp38αKO mice (Fig 3S, S1 Data). In contrast, increased total UCP-1 per depot for iWAT was observed in Fp38αKO mice compared to Floxed mice after 2 d of cold exposure (Fig 3S, S1 Data). To determine the effect of adipocyte-specific p38α deficiency on diet-induced obesity, we treated adult Fp38αKO and Floxed mice with a high-fat diet (HFD). After 3 months of HFD challenge, adult Fp38αKO mice gained less BW and had smaller iWAT and eWAT weight compared to Floxed mice (Fig 4A–4C, S1 Data). No significant difference was observed in iBAT weight and GAS muscle weight between these 2 groups of mice after HFD treatment (Fig A and B in S4 Fig, S1 Data). When normalized to the BW, the relative weight of either iWAT or eWAT was significantly lower in Fp38αKO mice than that in Floxed mice (Fig 4D, S1 Data). Although the liver weight of HFD-fed adult Fp38αKO mice was similar to that of Floxed mice (Fig C in S4 Fig, S1 Data), histological analysis and Oil Red O staining results revealed that the fatty liver was improved in HFD-fed adult Fp38αKO mice (Fig D and E in S4 Fig). In agreement with these findings, the GTT and insulin tolerance test (ITT) experiments revealed that HFD-fed adult Fp38αKO mice displayed improved glucose tolerance and insulin sensitivity (Fig 4E and 4F, S1 Data). As expected, we observed increased browning as indicated by formation of multilocular adipocytes, reduced adipocyte size, and increased expression of UCP-1 and other thermogenic genes in iWAT from HFD-fed Fp38αKO mice after 2 d of cold exposure (Fig 4G–4J, S1 Data). The adipocyte size was also reduced in eWAT from HFD-fed Fp38αKO mice exposed to cold for 2 d (Fig F and G in S4 Fig, S1 Data). Together, these results suggest that ablation of p38α in adipose tissues was able to prevent diet-induced obesity and improve metabolism, which might be attributed to the increased browning potential of WAT. Although we could not detect any differences in either oxygen consumption or carbon dioxide production between HFD-fed Floxed and Fp38αKO mice maintained at RT (Fig H and I in S4 Fig, S1 Data), we did see an increase in macrophage infiltration in the iWAT of Fp38αKO mice after 2 d of cold exposure, as evident from increased number of CD68+ cells (Fig J and K in S4 Fig, S1 Data). We also found that the mRNA expression of M2-related genes was increased, while the mRNA levels of proinflammatory cytokines were reduced in the iWAT of these HFD-fed Fp38αKO mice after cold exposure (Fig L and M in S4 Fig, S1 Data). These findings are in agreement with our current knowledge of the browning process. In an in vivo BrdU-labeling experiment, we found that the proportion of BrdU+ adipocytes in iWAT from HFD-fed Fp38αKO mice was reduced compared to Floxed mice after cold challenge for 7 d (Fig N and O in S4 Fig, S1 Data). Interestingly, the percentage of BrdU−UCP-1+ adipocytes relative to the total numbers of UCP-1+ adipocytes was higher in the iWAT of these HFD-fed Fp38αKO mice (80%) than that in control animals (40%), suggesting increased conversion or transdifferentiation of existing white adipocytes (UCP-1−) into beige adipocytes (UCP-1+) in these HFD-fed Fp38αKO mice during cold exposure (Fig O in S4 Fig, S1 Data). The finding that Fp38αKO mice were resistant to diet-induced obesity encouraged us to test whether pharmaceutically targeting p38α using SB203580 could have a similar beneficial effect. We found that 48 h of SB203580 treatment reduced the adipocyte size in iWAT and eWAT from C57BL/6J mice, accompanied by an increase of adipocyte size in iBAT from the same animal (Fig 5A and 5B, S1 Data). We also found that 4 wk of SB203580 treatment led to a lean phenotype, as evident from a decrease in BW and relative weight of iWAT, eWAT, and iBAT, but had no effect on liver weight (Fig 5C and 5D, Fig A and B in S5 Fig, S1 Data). Additionally, glucose levels were decreased after 4 wk of SB203580 treatment (Fig C in S5 Fig, S1 Data). The decrease in iWAT and eWAT weight, as wells as in glucose levels, became more evident in SB203580-treated mice upon cold exposure for 2 d (Fig 5E, Fig D in S5 Fig, S1 Data), which was accompanied by increased expression of UCP-1 and other thermogenic genes in iWAT (Fig 5F and 5G, S1 Data), suggesting the increased capacity of browning in these mice might contribute to the decreased adiposity and glucose levels. To be noted, the protein levels of p-CREB (Ser133) were also increased in SB203580-treated mice upon cold exposure, which might contribute to the elevation of UCP-1 expression (Fig 5G). The positron emission tomography (PET) tracer 18F-FDG was used to monitor the CL316,243-induced brown adipocyte recruitment into WAT in C57BL/6J mice after 4 wk of SB203580 treatment. As expected, PET/computed tomography (CT) showed an increase in 18F-FDG uptake in iWAT of SB203580-treated mice compared to control mice (Fig 5H and 5I). Ex vivo measurement of 18F-FDG uptake in different adipose tissues revealed that the 18F-FDG uptake in iWAT was increased in mice receiving SB203580 treatment (Fig 5J, S1 Data). A small increase in 18F-FDG uptake in eWAT of SB203580-treated mice was also observed, although the difference did not quite reach statistical significance (Fig 5J, S1 Data). We did not observe any differences in 18F-FDG uptake in iBAT between SB203580-treated mice and control mice. The 18F-FDG uptake in skeletal muscle was decreased in SB203580-treated mice, although there was no evidence for myofiber-type conversion (Fig E and F in S5 Fig, S1 Data). These results further suggest that SB203580 treatment could promote WAT browning without affecting BAT function in mice. SB203580 was also employed to treat obese db/db mice. Obese db/db mice gained less weight and had lower BW after receiving 3 wk of SB203580 treatment compared to control animals (Fig G and H in S5 Fig, S1 Data). Further investigation revealed that 3 wk of SB203580 treatment not only reduced the weight of iWAT but also decreased the adipocyte size in iWAT from these db/db mice (Fig 5K–5M, S1 Data). As expected, we observed increased infiltration of CD68+ macrophages in the iWAT from SB203580-treated db/db mice, accompanied by increased mRNA expression of M2-related genes and decreased mRNA expression of proinflammatory cytokines (Fig I-L in S5 Fig, S1 Data). Together, these results indicate that pharmaceutically targeting p38α might have a beneficial effect on metabolism. Given that genetic ablation of p38α was capable of increasing browning of WAT without affecting sympathetic activation, we hypothesized that p38α in WAT might function in a cell-autonomous manner. To test our hypothesis, we injected adenovirus-expressing p38αAF (Ad-p38αAF), which is a dominant-negative form of p38α, into the iWAT of C57BL/6J mice (Fig A and B in S6 Fig, S1 Data). As expected, a distinct histological morphology, significantly reduced adipocyte size, and increased staining of UCP-1 were observed in Ad-p38αAF-infected mice exposed to cold for 2 d (Fig 6A–6C, S1 Data). Additionally, Ad-p38αAF-infected mice had decreased glucose and TG levels compared to control mice (Fig C in S6 Fig, S1 Data). To further explore the role of p38α in WAT, we performed cell-based experiments. Interestingly, we found that SB203580 treatment increased the mRNA levels of UCP-1 and DIO2 in matured 3T3L1 adipocytes while suppressing the mRNA expression of UCP-1 and other thermogenic genes in matured brown adipocytes differentiated from either a BAC or stromal vascular fraction (SVF) isolated from iBAT (iBAT-SVF) of C57BL/6J mice (S6D–S6F Fig, S1 Data). Consistently, the mRNA expression of UCP-1 and other thermogenic genes was up-regulated by SB203580 treatment in a time-dependent manner in matured adipocytes derived from SVF isolated from iWAT (iWAT-SVF) of C57BL/6J mice (Fig 6D, S1 Data). Given that p38α is the major p38 isoform in either matured iBAT-SVF-derived adipocytes or matured iWAT-SVF-derived adipocytes (S6G Fig, S1 Data), similar results were obtained by using a p38α inhibitor (p38αMAPK-IN-1) (Fig 6E, Fig H in S6 Fig, S1 Data). Additionally, the mRNA expression of UCP-1 and other thermogenic genes was elevated in matured 3T3L1 adipocytes infected with lentivirus expressing p38αAF (Lenti-p38αAF) (Fig I and J in S6 Fig, S1 Data). We did not detect any effects of Lenti-p38αAF on lipid accumulation in these matured 3T3L1 adipocytes (Fig K in S6 Fig). Similarly, infection of Lenti-p38αAF increased the mRNA levels of UCP-1 and other thermogenic genes in iWAT-SVF-derived matured adipocytes (Fig 6F, Fig L in S6 Fig, S1 Data). To substantiate the evidence that inhibition of p38 signaling has differential effects on thermogenic program in brown adipocytes and white adipocytes, matured adipocytes derived from either iWAT-SVF or iBAT-SVF of Fp38αKO mice were used. We found that loss of p38α did not influence the adipogenic capacity of iWAT-SVF-derived cells (S6M and S6N Fig). In agreement with the results obtained by using either inhibitors or Lenti-p38αAF, the mRNA expression of thermogenic genes was increased in matured adipocytes derived from iWAT-SVF of Fp38αKO mice (Fig 6G, S1 Data). In contrast to what we observed in iWAT-SVF-derived matured adipocytes, the mRNA levels of thermogenic genes were decreased in matured adipocytes derived from iBAT-SVF of Fp38αKO mice (S6O Fig, S1 Data). An increase in mitochondria staining was also observed in iWAT-SVF-derived matured adipocytes infected with Lenti-p38αAF, indicating that the thermogenic program was activated upon p38α inhibition (Fig 6H). Accordingly, by using Agilent Seahorse XF24 Analyzer, we found that basal, uncoupled, and maximal OCRs were all significantly increased in matured adipocytes derived from iWAT-SVF of Fp38αKO mice as compared to Floxed mice, further suggesting that loss of p38α could enhance mitochondrial function in white adipocytes (Fig 6I and 6J, S1 Data). Consistently, the immunofluorescence staining of UCP-1 in iWAT-SVF-derived matured adipocytes was increased by Lenti-p38αAF infection (Fig 6K). Consistent with above findings using Lenti-p38αAF, we found that the UCP-1 expression was significantly increased in matured adipocytes derived from iWAT-SVF of Fp38αKO mice compared to control cells (Fig 6L). In addition, an increase in UCP-1 and DIO2 mRNA expression was also observed in matured adipocytes derived from WAT-SVF of Ap38αKO mice compared to control cells (Fig P in S6 Fig, S1 Data). Together, these results suggest that inhibition of p38α might promote white-to-beige adipocyte reprogramming in a cell-autonomous manner. In agreement with the findings that the protein levels of p-CREB (Ser133) were elevated in iWAT from either Fp38αKO mice (Fig 7A, Fig A in S3 Fig) or SB203580-treated mice (Fig 5G) after 2 d of cold exposure, we also observed that the protein levels of p-CREB (Ser133) were increased in either iWAT-SVF-derived matured adipocytes lacking p38α (Fig A in S7 Fig) or iWAT-SVF-derived matured adipocytes infected with Lenti-p38αAF (Fig B in S7 Fig). Consistent with these results, chromatin immunoprecipitation (ChIP) assay revealed that loss of p38α resulted in an enrichment of p-CREB (Ser133) on 2 cAMP response elements (CRE2 and CRE4) in the UCP-1 enhancer in iWAT-SVF-derived matured adipocytes (Fig C and D in S7 Fig). These results prompted us to explore the upstream signaling mediator that might contribute to the increased p-CREB (Ser133) levels. Since it has been reported that PKA is able to phosphorylate CREB at Ser 133, we tested whether PKA expression was altered when p38α was absent. Interestingly, increased staining of PKA catalytic subunit (PKA C) was observed in iWAT from cold-exposed adult Fp38αKO mice, accompanied by increased staining of p-CREB (Ser133) in the same animal (Fig 7A). Similar results were obtained in iWAT from 5-wk-old Fp38αKO mice maintained at RT (Fig E in S7 Fig). Consistently, the protein levels of PKA C and p-CREB (Ser133) were elevated in matured adipocytes derived from iWAT-SVF of Fp38αKO mice (Fig 7B). In agreement with these findings, either inhibition of p38α by Lenti-p38αAF or genetic deletion of p38α increased the immunofluorescence staining of PKA C in iWAT-SVF-derived matured adipocytes (Fig 7C and 7D). Additionally, the levels of phosphorylated PKA substrates in matured 3T3L1 adipocytes infected with Lenti-p38αAF were elevated, suggesting the PKA activity was increased by p38α inhibition (Fig F in S7 Fig). To test whether activation of p38 signaling could suppress the expression of PKA C, a lentivirus expressing a constitutive active mutant of an MAPK kinase (Lenti-MKK6E) was employed. As expected, in contrast to the effect of Lenti-p38αAF on the protein levels of PKA C, Lenti-MKK6E infection led to a decrease in PKA C protein levels in iWAT-SVF-derived matured adipocytes (Fig 7E). These data suggest that p38α in WAT is a negative regulator of PKA/CREB pathway. To test whether the PKA/CREB pathway mediated the effect of p38α inhibition on the thermogenic program in iWAT-SVF-derived matured adipocytes, H-89, an inhibitor of PKA, was employed. We found that suppression of PKA activity by H-89 was able to attenuate the effect of p38α inhibition either by SB203580 or Lenti-p38αAF on the UCP-1 mRNA expression in iWAT-SVF-derived matured adipocytes (Fig 7F and 7G, S1 Data). Consistently, we found that H-89 could abolish the effect of Lenti-p38αAF on the protein levels of UCP-1 in iWAT-SVF-derived matured adipocytes by using immunofluorescence staining (Fig 7H). Together, our results suggest that inhibition of p38α might stimulate the thermogenic program through PKA/CREB pathway in WAT. Interestingly, we found that the protein levels of p-CREB (Ser133) were reduced instead of increased in SB203580-treated BAC, matured adipocytes derived from iBAT-SVF of Fp38αKO mice, and iBAT from Fp38αKO mice maintained at RT compared to their controls, respectively (Fig G-I in S7 Fig). In contrast to p-CREB, we found that the protein levels of p-ATF2 were not only decreased in the matured adipocytes derived from the iWAT-SVF of Fp38αKO mice but also reduced in SB203580-treated matured adipocytes derived from iWAT-SVF of C57BL/6J mice (Fig J in S7 Fig). Thus, our data suggest that the molecular mechanism underlying the transcriptional regulation of UCP-1 in WAT is distinct from that in BAT. More interestingly, we found a putative CRE in the promoter region of the PKA C gene (Fig 8A). ChIP assay using the CREB antibody revealed an enrichment of CREB on this putative CRE in the PKA C promoter in either matured 3T3L1 adipocytes or iWAT of C57BL/6J mice (Fig 8B and 8C). Accordingly, luciferase assay demonstrated that overexpression of CREB could enhance the activity of PKA C promoter containing this putative CRE in HEK293T cells (Fig 8D, S1 Data). These results suggested that CREB could regulate the expression of PKA C at transcriptional level. In this mode, PKA and CREB form a positive feedback loop that serves to activate a thermogenic program that leads to white-to-beige phenotypic switching during adaption to cold (Fig 8E). Based on our current data, we propose that adipose p38α is a critical regulator of energy homeostasis. p38α functions as a brake of the PKA signaling pathway in WAT, thereby conferring robust and precise controls on the adaptation of adipose tissues to cold exposure and the whole-body energy metabolism (Fig 8E). Inhibition of p38α in WAT promotes browning, which could serve as a new therapeutic approach to combat obesity and improve metabolic homeostasis (Fig 8E). Maintenance of a proper BT is essential for survival in homeotherms. The sophisticated mechanisms of thermoregulation may participate in the control of energy homeostasis and the development of metabolic disorders [32,33]. Since both brown and beige adipocytes are present in adult humans and have remarkable capacity to dissipate stored energy, these two types of adipocytes hold great promise to treat obesity and metabolic diseases [34,35]. In this study, we demonstrate that p38α signaling controls the development and formation of beige adipocytes in WAT. We found that genetic ablation of p38α in adipose tissues facilitated WAT browning during cold exposure. Moreover, loss of p38α in adipose tissues led to a lean phenotype, improved metabolism, and resistance to diet-induced obesity. Given that pharmaceutical inhibition of p38α promoted the browning of WAT and had beneficial effects, we propose that p38α in WAT could serve as an exciting pharmacological target to combat obesity and metabolic diseases. p38α is activated in response to a variety of extracellular stimuli and mediates signal transduction from the cell surface to the nucleus. p38 has been long proposed to positively regulate the thermogenic program in brown adipocytes as a downstream mediator of cAMP/PKA signaling [25]. Our finding that the ablation of p38α in adipose tissues of mice caused minimal effects on thermogenic function of iBAT, as indicated by undetectable changes in UCP-1 expression, BT, morphology, and mitochondrial function, as well as energy expenditure in adult mice (Fig 2A–2M, Fig A-K in S2 Fig, S1 Data). These unexpected findings yielded from our study suggest that the loss of p38α in brown adipocytes is not as deleterious as we previously thought. On the other hand, to our knowledge, the role of p38 in the browning of WAT, especially the regulation of the thermogenic program in WAT, has never been extensively studied. Recently, PKA–apoptosis signal-regulating kinase 1 (ASK1) -p38 axis has been shown to contribute to the induction of brown adipocyte–specific gene expression in response to cAMP signaling [36]. Either CL316,243 treatment or enhancing the expression of ASK1 was able to induce UCP-1 expression, which was accompanied by an increase in p-p38 levels. However, the effect of p38 inhibition on CL316,243-induced ASK1 activation and UCP-1 expression was not investigated in that study. Therefore, the roles of adipocyte p38 in the regulation of thermogenic program remain unclear. Here, our data suggest that p38α acts as a negative regulator of browning in WAT (Fig 8E). Suppression of p38α, instead of activation of p38α, is able to stimulate the thermogenic program through enhancing PKA signaling in WAT. Thus, p38α differentially regulates PKA signaling and the thermogenic program upon cold exposure in BAT and WAT. These findings also indicate that p38α plays an important role in coordinating energy homeostasis by controlling the thermogenic program in an adipose depot–specific manner. Recent studies have greatly expanded our knowledge of beige adipocytes. In contrast to classic brown adipocytes, beige adipocytes express relatively low levels of theromogenic genes under nonstimulated conditions, which are dramatically induced upon cold exposure [8]. Although it is possible that common regulatory mechanisms may operate, since the thermogenic program in BAT and the browning of WAT are both stimulated by cold stress, it remains to be established whether cell type–specific mechanisms exist in brown and beige adipocytes [37]. In brown adipocytes, the activation of p38 by cAMP/PKA signaling stimulates the transcription of UCP-1 through direct phosphorylation of ATF2, which binds to a well-characterized enhancer located 2. 5 kb upstream of the UCP-1 gene [25]. Whether the cAMP/PKA/p38/ATF2 cascade also plays a similar role in WAT browning remains unclear. In our study, we did observe the effect of loss of p38α on the phosphorylation of ATF2 in WAT during browning; however, the reduced levels of p-ATF2 could not explain the elevation of UCP-1 expression, suggesting there is a distinct regulatory mechanism in WAT (Fig 3H and 3I, S1 Data). Interestingly, we found genetic or pharmaceutical inhibition of p38α was able to stimulate rather than suppress the transcription of UCP-1 through PKA/CREB cascade in WAT. These findings suggest that cell type-specific signaling cascades exist in BAT and WAT, which provide a means to fine-tune the expression of UCP-1 in all adipose tissues across the whole body. It has been widely accepted that beige fat is metabolically important, especially during cold exposure and nutrient overload [38–40]. Notably, obesity resistance in mice appears to be mostly related to browning of WAT rather than to adaptive thermogenesis of classic BAT, suggesting that beige fat is a key contributor to metabolic health [41]. Previous studies have clearly demonstrated that stimulating the browning process improves metabolism and protects mice from diet-induced obesity, whereas ablation of beige adipocytes results in metabolic dysfunction [42,43]. It has also been proposed that beige fat has nonthermogenic functions and regulates energy metabolism through various mechanisms in response to nutrient stress. In light of the presence of beige adipocytes in adult humans, these findings have been generating considerable interest, as understanding the molecular mechanisms underlying the browning of WAT could lead to novel therapeutic strategies for treating obesity and metabolic disorders [8,44]. Adipocyte p38 has been investigated before by using commercially available human adipocytes, preadipocytes, SVF, or directly using adipocytes collected from human subjects. These studies suggest that the adipocyte p38 is involved in the regulation of the response to inflammatory stress, cardiac natriuretic peptide-induced thermogenic program, and cellular aging [35,45–48]. More importantly, it has been reported that phosphorylation p38 was increased either in type 2 diabetic adipocytes or omental fat from obese women [49,50]. In our study, we observed that p38α deficiency in adipose tissues increased the browning in WAT, which was accompanied by resistance to obesity and improvement of metabolism. Mechanistically, p38α deficiency could promote white-to-beige adipocyte reprogramming in a cell-autonomous manner. Our findings, together with those reported by others [49,50], suggest that inhibition of the overactivated p38 in WAT may be beneficial for obese or diabetic subjects. Although we could not achieve WAT-specific delivery, our data suggest that pharmaceutical inhibition of p38α by SB203580 treatment is able to reduce the fat weight and glucose levels (Fig 5, S5 Fig, S1 Data). However, whether the pharmaceutical inhibition of p38α would promote white-to-beige adipocyte reprogramming in human adipocytes is not known and requires further study. In agreement with previous reports [31,51], our results indicate that although there was a dramatic increase in the protein levels of UCP-1 in the iWAT of mice upon cold exposure, at the system level the contribution from classic brown-fat UCP-1-mediated thermogenesis would still predominate (Fig 3S, S1 Data). Therefore, the changes in UCP-1 levels in the iWAT of Fp38αKO mice upon cold exposure might not explain the whole phenotype we observed in this study. UCP-1-independent and/or nonthermogenic mechanisms need to be investigated in future studies. Taken together, we establish an important role for p38α in the browning of WAT and energy homeostasis. Based on our findings, we propose that p38α in WAT could serve as a novel druggable target to combat obesity and metabolic diseases. All animal protocols were approved by the Animal Care Committee of Institute for Nutritional Sciences (INS), Shanghai Institutes for Biological Sciences (SIBS), and Chinese Academy of Sciences (CAS) (Approval number 2015-AN-12). All in vivo experiments described in this study were in accordance with institutional guidelines for the care and use of animals. Mice with a targeted deletion of p38α in adipose tissues were generated by crossing the p38αflox/flox (p38αf/f) mice with transgenic mice expressing Cre-recombinase under the control of the fatty acid binding protein 4 promoter (aP2-Cre) (Fp38αKO mice) or the adiponectin promoter (Adipoq-Cre) (Ap38αKO mice). Littermates expressing no Cre (Floxed mice) were used as a control group throughout the experiments. Mice were fed 60 kcal% HFD for 3 mo since 6 wk old or injected with SB203580 (20 mg kg−1, Medchemexpress Company) every 7 d for 1 mo. Mice were fed 60 kcal% HFD for 5 wk. Then, these HFD-fed mice were maintained in a cold environment (4 °C) and injected intraperitoneally with BrdU (200 mg kg−1, Sigma) twice a day for 7 d. The db/db mice were injected with SB203580 (20 mg kg−1) every 7 d for 3 wk from 6 wk old. Mice used in this study were aged between 2 to 4 mo if not specially pointed out. Male mice were used in the experiments. Cold treatment was conducted by sending mice to a cold room (4 °C) for 2 d or 7 d supplied with food and water. For acute cold challenge, mice were exposed to cold (4 °C) only with water. Mice were randomly assigned to each group; however, blinding was not possible. Mice with similar age or from same litters had the priority of use. During the experiments, mice were monitored daily. Any mice with significantly abnormal signs of rapid weight loss, inability to eat or drink, clinical symptomatology, toxicity, or unresponsiveness would be recorded, and the data from these mice were excluded for statistical analysis. We estimated the sample size by using an online program from http: //www. powerandsamplesize. com/Calculators/ for animal study. If the known value is 1, expected value is 1. 2 (20% difference between groups), standard deviation is 0. 1, and alpha is 0. 05, the power of the test will reach 0. 93 when sample size is 3, or the power will reach 0. 99 when sample size is 5. If the expected value is 1. 1–1. 5 (smaller difference), 5–10 or even more samples will be used to have the power bigger than 0. 9. The food intake was evaluated by weighing out the grams of food every 12 h (7 AM–7 PM day, 7 PM–7 AM night). Minispec TD-NMR Analysers were used to evaluate living body composition. Rectal temperature was measured with a model BAT-12 thermometer (Physitemp Instruments). To measure energy expenditure, mice with or without β3 agonist (CL316,243 Sigma) (1 mg kg−1 BW) injection were placed in metabolic cages (Columbus Instruments) to assess their O2 consumption and CO2 production. Gross energy content of food and feces in 24 h was determined using oxygen bomb calorimeter (IKA Oxygen Bomb Calorimeter C 6000). For GTT, mice were fasted for 14 h and injected with D-glucose (2 g kg−1 BW or 2. 5 g kg−1 LM) (Sigma) intraperitoneally. For ITT, mice were fasted for 4 h and injected with recombinant human insulin (1 U kg−1, Roche) intraperitoneally. We measured mouse blood glucose levels with whole blood from the tail vein using a glucose meter (Abbott). For TG measurement, plasma was collected through fresh whole blood centrifuged for 10 min after 30 min of standing. TG and NEFA levels were measured using a commercial ELISA kit (Labassay). The activity of creatine kinase in plasma, skeletal muscle, and hearts was measured using an ELISA kit (Abnova). To activate browning process in WATs, mice were treated with β3 agonist (CL316,243 Sigma) (1 mg kg−1 BW) daily by intraperitoneal injections for 8 d at RT. Mice PET-CT imaging was performed by Siemens Inveon PET-CT Multimodality System. In brief, mice were fasted overnight, lightly anesthetized using 3% isofluorane, and injected with approximately 150 μCi of 18F-FDG into the tail vein. After that, the animal was permitted to roam freely in the cage for 1 h to uptake 18F-FDG. Subsequently, the animal was placed onto the imaging bed under 2% isofluorane anesthesia for the duration of imaging. All the PET/CT experiments were operated under RT. Tissues of mice after PET/CT imaging were ex vivo measured with γ counter (SN-695 γ RIA Counter) and corrected with tissue weight, respectively. Mouse tissues were fixed in 4% paraformaldehyde and embedded in paraffin. Sections were stained with hematoxylin and eosin or oil red according to standard protocols. Immunohistochemical staining of paraffin sections was carried out with 1: 50 anti-UCP-1 (Abcam), 1: 50 p-CREB (Ser133) (CST), and 1: 50 PKA C (CST) and detected by Inverted microscope (Olympus). The cell sizes and areas of adipose tissues were measured by Image J. Immunofluorescence staining of paraffin sections for BrdU incorporation was carried out with primary antibodies 1: 50 anti-UCP-1 (Abcam), 1: 100 BrdU (Santa Cruz), or 1: 200 CD68 (Bio-Rad, Formerly AbD Serotec) and secondary antibodies Alexa Fluor 594 conjugated goat antibody to rabbit IgG, Alexa Fluor 488 conjugated goat antibody to mouse IgG, or Alexa Fluor 594 conjugated goat antibody to rat IgG (Invitrogen, 1: 1,000). The Elecron microscopic observations were conducted through scanning electron microscope (PHILIPS CM120). All the representative images were repeated in at least 3 independent experiments. Macrophages and neutrophils were isolated from blood of Floxed and Fp38αKO mice. RBCs were lysed using ACK lysis buffer. A single-cell suspension was used for staining cell surface markers F4/80, Gr-1, and Mac-1 (eBioscience) following standard protocols, and data acquisition was performed using a FACSAria II cytometer (BD). Flow cytometric data were analyzed with FLOWJO software. Total RNA was extracted from cells or tissues using TRIzol reagent (Invitrogen) in accordance with the manufacturer' s instructions. One microgram of RNA was transcribed to complementary DNA with the RT Reagent Kit (Takara). Real-time PCR was carried out on the 7900 System (ABI) using SYBER Green Supermix (Takara). Primers used in this study were provided in S1 Table. Data were normalized to 18S and analyzed using the ΔΔCT method. To quantify the expression of p38 isoforms, absolute quantification of p38α, p38β, p38γ, and p38δ was performed. Four pairs of primers were designed to amplify fragments from mice cDNA, which were used as templates for standard curves. The other 4 pairs of primers were designed to perform regular real-time PCR to get specific copy numbers according to standard curves respectively. Proteins of cells or tissues were extracted by RIPA buffer. All protein samples were subjected to 5 ng/μL and immunoblot assay with the indicated antibodies. Hsp90 (CST) or α-tubulin (Sigma) were used as internal controls. Detailed information on the antibodies used in this study is provided in S2 Table. The representative blotting bands were repeated at least 3 times. The gray intensity of blotting bands was evaluated through Image J. 3T3L1 preadipocyte cell line was cultured in DMEM supplemented with 10% NCS (Gbico) and 1% penicillin/streptomycin at 37 °C with 5% CO2. SVF-derived preadipocytes were isolated as described previously from iWAT. Firstly, inguinal pads from 5-wk-old mice were minced and digested with 2% collagenase type I in DMEM for 30 min at 37 °C, followed by quenching with complete medium. Cell suspensions were centrifuged washed and filtered through a 70 μm strainer (BD Biosciences) and were plated onto 10 cm dishes in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin (Invitrogen). BAC cell line was immortalized from SVF-derived preadipocytes from iBAT of newborn C57BL/6J mice by SV40 retriovirus and cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin at 37 °C with 5% CO2. Adipocyte differentiation of confluent cells—including 3T3L1, BAC, and SVF-derived preadipocytes—was induced in growth medium with 5 μg/ml insulin, 0. 5 mM IBMX (Sigma), 1 μM DEX (Sigma), 1 nM T3, and 5 μM Rosiglitazone (Sigma) for 48 h and replaced with growth medium supplemented with insulin, T3, and Rosiglitazone for 4 d. After that, matured adipocytes were infected with lentivirus GFP, p38αAF, or MKK6E for 48 h. Ten micromoles of SB203580 (Merck/millipore) —the inhibitor of p38α and p38β—20 μM p38α MAPK–IN–1 (MCE), or 20μM H-89 (Selleck) —the inhibitor of PKA—were preincubated before other treatments or sample collection. Ten nanomoles of Forskolin (Sigma) were added to matured adipocytes for 3 h before RNA/immunofluorescence collection or 30 min before protein collection unless pointed out. HEK293T cell line was cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin at 37 °C with 5% CO2. The mitochondria were stained by Mito-Tracker Green (Beyotime) for 30 min in DMEM. Cells were fixed with 4% paraformaldehyde and incubated with 1: 50 anti-UCP-1 (Abcam) or 1: 100 anti-PAK C (CST) or 1: 100 anti-Perilipin (CST) antibody and further detected by secondary antibodies Alexa Fluor 594 conjugated goat antibody to rabbit IgG (Invitrogen 1: 1,000). Cells were then washed in PBS and stained with 40, DAPI. Images were acquired by fluorescence microscopy (Zeiss System) or High Content Screening Microplate Imaging Reader (Thermo Fisher Scientific). The representative images were repeated in at least 3 independent experiments. Genome DNA from adipose tissues of Floxed and Fp38αKO mice was extracted using a DNA Mini Preparation Kit with Spin Column (Beyotime, Shanghai). The presence of amplifiable mitDNA and nuDNA in the extract was assayed through real-time PCR. Differentiated primary adipocytes were trypsinized on day 8 after differentiation and plated into the XF24 V7 cell culture microplate. After 48 h, the OCRs were determined by a Seahorse Bioscience XF24 Extracellular Flux Analyzer (Seahorse Bioscience), with Oligomycin 2 μM, FCCP 1. 5 μM and Antimycin A/Rotenone 1 μM injected during fixed time intervals. The OCRs were normalized by proteins in each well. Mitochondrial respiration of BAT after cold exposure was determined using an XF24 Extracellular Flux Analyzer (Seahorse Bioscience) using 5 μg mitochondrial protein in a buffer containing 50 mM KCl, 4 mM KH2PO4,5 mM HEPES, and 1 mM EGTA, 4% BSA, 10 mM Pyruvate, 5 mM Malate, 1 mM GDP. Mitochondria were plated and centrifuged 2,000 g for 20 min to promote adherence to the XF24 V7 cell culture microplate. One millimole of ADP, 4 mM Oligomycin, 6mM FCCP, and 2 mM each of Antimycin A/Rotenone were added during fixed time intervals. The recombinant adenovirues of GFP (control) and p38αAF were generated as previously described in another study [52]. In order to generate the lentivirus for GFP, p38αAF, and MKK6E, the synthesized sequences were inserted into Fugw-vector plasmids, which were PCR from plasmids described previously. The above Fugw plasmids and packaging plasmids PMD2. G and PSPAX were cotransfected into HEK293T cells. The virus particles were collected from supernatant. Titers were determined by using dilution methods and counting the number of GFP-positive colonies using fluorescence microscope. PKA C (CRE) promoters were amplified with primers: PKA C (CRE) KpnI-f: CGG GGT ACC CCG GAC CTA GTC AGA CTT TGG AG; PKA C (CRE) XhoI-r: CCG CTC GAG CGG ATC AGT TTG TCT TGG GGA CT. For the luciferase reporter assay, HEK293T cells were plated in 48-well plates and transfected with PKA C (CRE) reporter constructs and MSCV or CREB vectors (Addgene). The pRL-TK vector (expressing Renilla luciferase) was used to normalize the luciferase activity. Cells were lysed 48 h after transfection, and luciferase activity was measured using a dual-luciferase reporter assay system (Promega). ChIP assays of SVF-derived matured adipocytes were performed using an EZ Magna ChIP G kit (Minipore) according to the manufacturer’s protocol. Immunoprecipitation was performed using an anti-pCREB (Ser133) antibody or with rabbit IgG (Santa Cruz) as a negative control. Primers used for amplifying CRE2 and CRE4 were CRE2-chip-f, GAT AAG AAG TTA CGA CGG GA; CRE2-chip-r: TCT GAG GAA AGG GTT GAC CT; CRE4-chip-f, GAA GAG TGA CAA AAG GCA CC; and CRE4-chip-r: TAT ATA GCC CCT TGC CGG AG. Immunoprecipitation to identify putative CRE in PKA C promoter was performed using an anti-CREB1 antibody (Abcam) or with rabbit IgG (Santa Cruz) as a negative control. Primers used for amplifying putative CRE in PKA C promoter were PKA C (CRE) -chip-f: AGG GAC AGT GCC TCA AAC CT; PKA C (CRE) -chip-r: TGA CAA GCC TGT ACC AGA GA. The PCR cycle parameters were 95 °C for 5 min, then 30 cycles of 95 °C for 25 s, 60 °C for 30 s, and 72 °C for 30 s, followed by a final extension at 72 °C for 3 min for both the ChIP product and input (represent 0. 2%). PCR products were resolved by electrophoresis in a 2% Agarose-gel (Invitrogen). Data were expressed as means ± SEM. The statistical differences in mean values were assessed by Student t test. All experiments were performed at least twice, and representative data are shown. Supporting information includes S1 Table (RT-qPCR primers used in this study), S2 Table (primary antibodies used in this study), 7 Supplemental Figures (S1–S7 Figs), and S1 Data (Excel spreadsheet containing, in separate sheets, the underlying numerical data for all figures).

The functional brown adipose tissue (BAT) identified in human adults consists of not only classic brown adipocytes but also brown-like adipocytes (beige adipocytes), both of which are important for energy homeostasis. Due to the same ability to convert fat into heat as brown adipocytes, beige adipocytes have been considered as a novel pharmacological target to combat obesity. Growing evidence suggests that promoting the development and formation of beige adipocytes in white adipose tissue (WAT), also called the browning of WAT, is able to prevent diet-induced obesity and improve metabolism in rodents. Thus, understanding the molecular basis for the regulation of browning in WAT may help us to develop new strategies to counteract obesity and metabolic diseases. In this study, adipocyte-specific p38α knockout (Fp38αKO) mice are generated that display a lean phenotype, improved metabolism, and resistance to diet-induced obesity. Interestingly, we found that adipocyte p38α deficiency facilitates the browning in WAT. Then, we show that pharmaceutical inhibition of p38α enhances the browning in WAT and has metabolic benefits. We propose that inhibiting p38α in WAT, possibly combined with cold exposure, could constitute an exciting pharmacological target to combat obesity and metabolic diseases.

Abstract Introduction Results Discussion Materials and methods

group-specific staining body weight medicine and health sciences hematoxylin staining brown adipose tissue animal models model organisms adipocytes physiological parameters connective tissue cells obesity experimental organism systems mitochondria pharmacology bioenergetics cellular structures and organelles drug metabolism research and analysis methods specimen preparation and treatment staining animal cells connective tissue biological tissue mouse models pharmacokinetics biochemistry cell staining cell biology anatomy adipose tissue physiology biology and life sciences cellular types energy-producing organelles

2018

Metabolic benefits of inhibition of p38α in white adipose tissue in obesity

The surface of polyomavirus virions is composed of pentameric knobs of the major capsid protein, VP1. In previously studied polyomavirus species, such as SV40, two interior capsid proteins, VP2 and VP3, emerge from the virion to play important roles during the infectious entry process. Translation of the VP3 protein initiates at a highly conserved Met-Ala-Leu motif within the VP2 open reading frame. Phylogenetic analyses indicate that Merkel cell polyomavirus (MCV or MCPyV) is a member of a divergent clade of polyomaviruses that lack the conserved VP3 N-terminal motif. Consistent with this observation, we show that VP3 is not detectable in MCV-infected cells, VP3 is not found in native MCV virions, and mutation of possible alternative VP3-initiating methionine codons did not significantly affect MCV infectivity in culture. In contrast, VP2 knockout resulted in a >100-fold decrease in native MCV infectivity, despite normal virion assembly, viral DNA packaging, and cell attachment. Although pseudovirus-based experiments confirmed that VP2 plays an essential role for infection of some cell lines, other cell lines were readily transduced by pseudovirions lacking VP2. In cell lines where VP2 was needed for efficient infectious entry, the presence of a conserved myristoyl modification on the N-terminus of VP2 was important for its function. The results show that a single minor capsid protein, VP2, facilitates a post-attachment stage of MCV infectious entry into some, but not all, cell types. The icosahedral polyomavirus capsid is constructed from 72 pentameric capsomers of the major capsid protein, VP1 [1]. VP1 mediates attachment of the virus to cell surface glycan receptors [2], [3], initiating infectious entry and delivery of the encapsidated circular ∼5 kb dsDNA viral genome to host cells. The VP1 protein of previously studied polyomaviruses, such as simian virus 40 (SV40) and murine polyomavirus (MPyV), associates with two minor capsid proteins, called VP2 and VP3, that are thought to emerge from the capsid interior to play important roles during the infectious entry process [4]–[7]. VP2 and VP3 are translated from a single, un-spliced open reading frame (ORF) [8]. The initiating AUG codon of VP3 is in-frame with and downstream of the initiating VP2 AUG. Thus, VP3 is an N-truncated isoform of VP2. In most polyomaviruses, the unique N-terminus of VP2 carries a consensus sequence for N-terminal myristoylation (MGXXXS/T). The VP2 proteins of SV40 and MPyV have been shown to be myristoylated, and mutations that prevent myristoylation reduce the infectivity of MPyV [9]–[11]. In 2008, a new human polyomavirus was discovered in a rare and aggressive form of skin cancer, known as Merkel cell carcinoma (MCC) [12]. The virus was named Merkel cell polyomavirus (MCV or MCPyV). A large number of studies indicate that accidental integration of MCV DNA into the host genome is a causal factor in most cases of MCC [13], [14]. Although MCC is uncommon, chronic MCV infection of human skin appears to be very common, and most infected people will experience no known symptoms as a result of their infection with MCV [15], [16]. Since the discovery of MCV, renewed interest in polyomaviruses and advances in deep sequencing technologies have led to the discovery of many additional human and animal polyomaviruses [17]–[35]. The newly expanded view of this viral family permits greater resolution of significant sequence differences between members. For instance, alignment of VP2 sequences reveals that a subset of polyomaviruses, including MCV, lack a conserved peptide motif that comprises the N-terminus of known VP3 proteins. The VP1 and VP2 proteins of this subset of polyomaviruses have in common other distinctive sequence features, which will be discussed in detail below. These sequence differences led us to question whether the expression and function of the MCV minor capsid proteins might differ from more extensively studied members of the polyomavirus family. Several publications have addressed the role of VP2 and VP3 in the life cycle of different polyomaviruses. However, it is difficult to draw universal conclusions about the precise function of the minor capsid proteins from these reports. Results consistent with a defect in virion assembly and viral genome packaging were reported when one or both minor capsid proteins of JC polyomavirus (JCV or JCPyV) were deleted [36]. In contrast, SV40 VP1-only virions packaged and protected viral DNA as efficiently as VP1/2/3 virions [4], [5]. SV40 attachment to cell surfaces was determined to be normal in the absence of both minor capsid proteins in one report [4], but binding of VP2-deleted SV40 virions appeared reduced in another report [5]. In nearly every case, infection or plaque formation by the virus is reduced or abrogated when either of the minor capsid proteins is deleted [4]–[6], [11], [36], [37], but the mechanism by which the minor capsid proteins facilitate infection remains controversial. The work of Daniels et al. [5] suggests that the role of both SV40 minor capsid proteins is to promote escape of the virus from the endoplasmic reticulum (ER). Inoue et al. [6] agreed that VP3 was needed for ER escape, but concluded that the principal role of VP2 was to direct trafficking of the virus to the ER. Conversely, Nakanishi et al. [4] provided evidence that the role of the SV40 minor capsid proteins lay in nuclear import of the viral genome during the infectious entry process and found that the virus could escape from the ER into the cytoplasm without the benefit of minor capsid proteins. Reports concerning the role and importance of VP2 myristoylation also lack uniformity [9], [11], [36]–[38]. We have examined, for the first time, the expression and function of the MCV VP2 protein as well as the previously proposed VP3 protein of MCV using both pseudoviruses and native MCV virions propagated in cell culture. VP3 protein was not detected in MCV-infected cells or in purified MCV virions. Our phylogenetic analyses suggest that this feature of MCV is typical of an emerging clade of polyomaviruses. The late 19S SV40 mRNA was previously shown to code for multiple proteins through a mechanism of “leaky” ribosomal scanning [8], [39]. The protein products of the SV40 19S mRNA can be predicted based on what is known of translation initiation at Kozak consensus sequences. The optimal Kozak consensus sequence is minimally defined as GCC RCC AUG G (where the most important positions are in bold) [40]. The relatively weak Kozak sequence surrounding the SV40 VP2 translation initiation site (AGG UCC AUG G) is thought to result in it being bypassed by scanning ribosomes with a frequency of about 70%, while the stronger Kozak context of VP3 (CCA GGA AUG G) results in more frequent translational initiation of this gene product [39]. In contrast to SV40, the MCV VP2 Kozak sequence is strong (UUC AGG AUG G) and the Kozak context surrounding the proposed MCV VP3 initiation codon (Met46) [12] is very weak (AGU UUA AUG A). The Kozak sequence surrounding the only other methionine codon in VP2 (Met129) is weak (GCA CUU AUG G), and ribosome access to Met129 is also obstructed by four out-of-frame upstream AUG codons (in addition to Met1 and Met46). Since conventionally scanning ribosomes would initially encounter the strong Kozak sequence surrounding the VP2 initiation site before they could reach either of the weaker initiation codons that might conceivably initiate a VP3-like protein, VP2 seems likely to be the primary product of the MCV late mRNA. An additional consideration in our sequence-based analyses is that documented VP3 proteins carry a consensus N-terminal motif defined as MALXXΦ, where Φ represents an aromatic residue. MCV VP2 does not encode any sequence with homology to this conserved VP3 N-terminal motif (Figure S1). We have previously published methods for production of native MCV virions from 293TT cells transfected with MCV genomic DNA [41], [42]. Purified MCV virions can be propagated in a cell line named 293-4T, which stably expresses the MCV small t (sT) and Large T (LT) antigens. The 293TT system can also be used for intracellular production of recombinant MCV-based reporter vectors (pseudovirions) [15]. Using these tools, we sought to determine the minor capsid protein make-up of MCV virions. When purified MCV virions produced by transfection of 293TT cells were initially examined by SDS-PAGE and western blot with a rabbit polyclonal anti-VP2 antiserum, VP2 could readily be detected but no proteins migrating faster than VP2 were apparent (data not shown). To increase the possibility of detecting a VP3 protein that was simply not abundant, larger amounts of 293-4T-propagated native MCV virions were concentrated by immunoprecipitation using an anti-VP1 monoclonal antibody. The concentrated native virions were compared to recombinant pseudovirion standards produced with or without VP2 and VP3 (Figure 1A). Western blotting of concentrated native virions showed a strong VP2 band and no visible VP3 bands (Figure 1B). Thus, MCV virions produced using the native viral regulatory sequences do not contain detectable amounts of VP3 capsid protein. Analysis of recombinant pseudovirions shows that recombinantly expressed VP3 protein is stably incorporated into VP1 capsids (Figure 1A). The absence of VP3 proteins in native MCV virions thus presumably reflects a lack of VP3 protein expression in MCV-infected 293-4T cells. When whole lysates of 293-4T cells transfected with MCV genomic DNA were analyzed by western blot with VP2 antiserum, VP3 proteins were again not visible, confirming that MCV does not express detectable amounts of VP3 protein (Figure S2). In an effort to genetically rule out the possibility that undetectably low levels of VP3 might play a role in MCV infection, both of the internal VP2-frame methionine codons (Met46 and Met129) were mutated in the context of the MCV genome. Based on homology to the VP2 proteins of close relatives of MCV, Met46 was changed to a valine (mutant dVP3a) and Met129 was changed to a leucine (mutant dVP3b). A third variant was created (dVP3d) in which both methionine residues were mutated. Purified virion preparations for each of the mutants were produced and their infectivity was compared to that of a preparation of wild-type (WT) MCV virions. The yield of each virus was similar, there was a similar ratio of VP2 to VP1, and the number of MCV genomes packaged was nearly equivalent in each of the viruses (Figure 2A and data not shown). When the WT or mutant viral genomes were transfected into 293-4T cells, there was no evidence that the mutations affected VP1 or VP2 expression or altered cell viability (Figure S2 and S3). The infectivity of the viruses was assessed in 293-4T cells, which were inoculated with equivalent amounts of each virus, standardized by genomic copies (10 viral genomes/cell). Viral genome replication was determined by quantitative PCR (qPCR). Individual mutation of the VP3a and VP3b methionine residues in native MCV virions resulted in less than a two-fold decrease in infectivity of the viruses (Figure 2B). The apparent infectivity of the double mutant was reduced by about four-fold, suggesting the effect of the mutations was additive. Unfortunately, it is not possible to determine if the modest reduction in infectivity is a consequence of mutating the VP2 protein (causing VP2 to function less effectively) or if the WT virus expresses an undetectable amount of VP3 that confers a minor improvement in infectivity. Nevertheless, these results show that both conceivable VP3 proteins are largely or entirely dispensable for native MCV infection. MCV reporter pseudoviruses were used to confirm the native virion results. Production of pseudovirions is less costly and less time-consuming than native virion production, assays for pseudovirion transduction have much higher throughput, and pseudovirions can transduce a wide variety of cell lines that do not support native MCV replication. The pseudovirus system also allows ectopic expression of candidate VP3 proteins. We have previously reported that supplementing the pseudovirus production system with VP3 expression plasmids does not improve the infectivity of MCV reporter pseudovirions [15]. However, the VP2 expression construct used in the previous report was intentionally designed to have a poor Kozak sequence context surrounding VP2, allowing limited amounts of VP3 expression by leaky scanning (Figure 1A). In an initial set of pseudovirus-based experiments, we found that improving the Kozak context surrounding the recombinant VP2-initiating AUG codon reduced the amount of “leaky” VP3 expression to undetectable levels (data not shown). Despite the reduced VP3 expression, pseudoviruses made using the improved-Kozak VP2 expression construct transduced 293TT cells with efficiency similar to pseudoviruses made using the original leaky VP2 expression construct (data not shown). To eradicate VP3 expression, the proposed VP3-initiating Met46 was comprehensively mutated to various other amino acids, anticipating that some VP2 mutations might be better tolerated than others. A confounding variable in these experiments was that some mutant VP2 proteins appeared to be expressed at a lower level than WT VP2, while VP1 expression was similar (Figure 3A). The transduction efficiency of WT and mutant pseudovirions, normalized to VP1 concentration, was determined in 293TT cells (Figure 3B). The majority of Met46 mutants (Ile, Val, Ala, and Asn) exhibited a ∼20–30% reduction in infectivity, while transduction by the M46D mutant was reduced by ∼65%. This shows that some Met46 mutations have a modest impact on VP2 biology and confirms that eradication of VP3 expression has little or no effect on pseudovirion infectivity. The data show that MCV is distinct from previously studied polyomaviruses in that it does not employ a VP3 minor capsid protein. In light of this discovery, the importance of the VP2 protein was next explored. MCV pseudovirions were produced by expression of VP1 in the absence of VP2 (“None”), with a low level of VP2 (“Low”), or with a high level of VP2 (“High”). The difference in VP2 incorporation in purified pseudovirus preparations can be observed in a protein stain of denatured pseudovirions in SDS-PAGE (Figure 4A). Quantitative PCR analysis of the encapsidated reporter plasmid associated with each pseudovirus preparation suggested that VP2 had no role in the encapsidation of the reporter plasmid (data not shown). When pseudovirion stocks were normalized by reporter gene copy number and inoculated onto 293TT cells, a VP2 dose-dependent effect on infectivity was observed (Figure 4B). Of note, transduction of cells by the VP1-only particles was also non-zero and VP1 dose-dependent. To verify that VP2 plays an important role in efficient MCV entry in other cell types, several additional cell lines that had previously been shown to be highly transducible by MCV [42] were challenged with pseudovirions containing varying levels of VP2. Intriguingly, the effect of VP2 on MCV pseudovirus transduction efficiency differed dramatically from one cell type to the next (Figure 4C–F). Transduction of NCI/ADR-RES cells (an ovarian cancer line) was strongly enhanced by high levels of VP2, but low levels of VP2 improved transduction very little relative to VP1-only pseudoviruses. Transduction of A549 cells (a lung cancer line) was also improved by the presence of VP2, but the degree to which VP2 enhanced infectivity was not as great as in 293TT cells. In contrast to the other cell lines, MCV infectivity of UACC-62 cells and SK-MEL-2 cells (both melanoma lines) was nearly unchanged when VP2 was present at any level, suggesting that VP2 is dispensable for MCV infectious entry in some cell types. To determine whether minor capsid protein-independent entry in some cell lines is a feature unique to MCV, we tested the VP2/VP3 dependence of another polyomavirus pseudovirus. BK polyomavirus (BKV) genotype IV pseudovirion transduction in multiple cell lines has also previously been established [42]. BKV pseudovirions were produced with VP1 alone, VP1+VP2, VP1+VP3, or VP1+VP2+VP3. Infectivity of the BKV pseudovirions was then examined in seven different cell lines, four of which were previously challenged with MCV pseudovirions (Figure 5A–D). The other three cell lines were chosen because they are highly BKV-transducible and represent diverse tissue types (Figure 5E–G). The infectivity of the BKV VP1-only pseudovirions was dramatically lower than the VP1+VP2+VP3 pseudovirions on all tested cell lines (Figure 5A–G). This analysis included the SK-MEL-2 cell line that MCV pseudovirions transduced in a VP2-independent fashion (Figure 4E and 5D). Thus, infectious entry of the BKV pseudovirus appears to differ from the MCV pseudovirus with regard to its dependence on the presence of minor capsid proteins. That a minor capsid protein would not contribute to MCV infection in some cells is highly unexpected, and it would be desirable to verify the results with native MCV virions. Unfortunately, few cell types replicate MCV DNA to an appreciable level [43], [44], and 293-4T cells are the only cell type demonstrated to replicate MCV genomic DNA delivered by MCV infection to detectable levels above background [42]. As 293TT cells are the parent of 293-4T cells, the presence of VP2 in virions would be expected to contribute to efficient native MCV infection of 293-4T cells. To test this prediction, MCV genomic DNA was mutated with a frame-shift mutation to prevent VP2 production, and additional changes were made to the downstream sequence to ensure VP3 would not instead be produced by translational reinitiation. The mutant (dVP2) and WT native virions were produced as before and purified. A western blot of the virion preparations shows that only the WT virus contains VP2 protein (Figure 6A). The yield of dVP2 virions was similar to WT, and the rate of viral genome encapsidation was nearly identical between the two viruses (data not shown). Binding of the dVP2 and WT viruses to 293-4T cells proved to be equivalent in qPCR-based measurements of cell-associated MCV genome copies after one hour of incubation at 4°C or 37°C (Figure 6B). Infectivity of the dVP2 mutant was determined similarly to the dVP3 mutants. Duplicate wells of 293-4T cells were collected one day and four days post-infection with WT or dVP2 virions. In addition, a negative control was performed in which anti-MCV neutralizing antibodies were added at the time of infection or the following day. The addition of neutralizing antibodies at the time of infection demonstrates the validity of the infection assay, while neutralization one day after an initial round of infection controls for the possibility of multiple rounds of infectious spread. Both the WT and dVP2 viruses displayed increases in MCV genome copy number between day one and day four (Figure 6C), suggesting both were capable of infecting 293-4T cells. However, the apparent infectivity of the dVP2 virus was much less than WT, with ∼100-fold fewer replicated MCV genomes on day four. These results confirm that the MCV VP2 protein has an important function during infectious entry of native MCV virions into 293-4T cells. Since the VP2 protein occupancy of MCV appears to be a strong determinant of infectious entry in multiple cell types, we carefully measured the apparent ratio of VP1∶VP2 in MCV pseudovirions and native virions. MCV pseudovirions with high VP2 content and WT native MCV virions were analyzed alongside BSA standards by SDS-PAGE and SYPRO Ruby stain (Figure 7). This protein stain is believed to show little protein-to-protein variability in staining intensity, as it binds primarily to the polypeptide backbone, with minor contributions from basic amino acid residues [45]. Since the MCV VP1 protein contains a higher percentage of basic residues than MCV VP2, the stain might be expected to slightly over-estimate the relative abundance of VP1. Quantitative analysis of the stained gels showed a VP1∶VP2 molar ratio of 5∶2 for native MCV virions (i. e. , two molecules of VP2 per pentameric VP1 capsomer). MCV pseudovirions exhibited slightly lower VP2 occupancy, with a VP1∶VP2 ratio of 5∶1. 4. There are several possible roles that VP2 may be playing to promote the infectious entry of MCV into some cell lines. One possibility is that VP2 directs MCV through a cellular pathway that is suitable to accomplish infection, while particles lacking VP2 only rarely find this pathway. If this were true, it might be possible to visualize differences in subcellular localization of VP1-only and VP1+VP2 MCV particles by fluorescence confocal microscopy. Previously published studies found that EdU-labeling of human papillomavirus type 16 (HPV16) pseudovirion DNA is a sensitive method for the detection of virus-associated DNA throughout the entry process [46]. We therefore chose to label the DNA encapsidated within pseudovirions with EdU for high-specificity detection by Click-iT chemistry and imaging with co-localization of organelle markers. Since 293TT cells do not adhere well and do not have morphology suitable for microscopic analysis, the analyses were performed using the NCI/ADR-RES cell line, which supports VP2-dependent infectious entry (Figure 4C). MCV, with or without VP2, was visible in a punctate pattern within the cell by roughly six hours after pseudovirion inoculation (data not shown). This pattern changed little over a period of several days. The EdU signal did appear to increase significantly over time, likely resulting from increasing accumulation of particles in a single location (data not shown). Co-staining of EdU with LAMP-1 revealed strong co-localization of encapsidated DNA in the late endosome/lysosomal compartment (Figure 8A). The MCV VP1 protein also appeared to co-localize with LAMP-1+ compartments (data not shown). At no time was MCV convincingly co-localized with the ER markers calreticulin or ERp72, or the Golgi marker giantin, and EdU signal was rarely observed in the nucleus unless the nucleus was undergoing division (Figure 8B and data not shown). The presence or absence of VP2 did not discernibly alter these patterns of sub-cellular localization. To address the concern that the cells might non-specifically sequester a majority of MCV particles in lysosomes for non-infectious degradation, BKV was also produced with EdU-labeling of DNA and examined alongside MCV. BKV was not found co-localized with the LAMP-1 marker, and the pattern of EdU detected in BKV-transduced NCI/ADR-RES cells clearly differed from MCV (Figure 8A). The majority of BKV particles amassed in a single location overtime, the identity of which was not revealed by any of our subcellular markers (Figure 8B and data not shown). The result shows that MCV' s trafficking to LAMP-1+ compartments is a distinctive feature of MCV' s biology. MCV particles +/− VP2 were also examined by differential detergent subcellular fractionation [47], and, again, no effect of VP2 was detected using these methods (data not shown). The significance of these results is uncertain considering the particle-to-infectivity ratio of the polyomavirus pseudoviruses is such that it may be difficult to observe the small fraction of particles at a particular step of the productive infectious pathway at any given moment during the slow and potentially asynchronous process of entry into cells. The intriguing differences between MCV and previously studied polyomaviruses inspired further investigation of the VP2 protein. Whether the VP2 N-terminal consensus sequence for myristoylation resulted in the covalent attachment of a myristoyl group was next examined. Traditional methods for analyzing myristoylation use tritiated substrates, and detection is time-consuming and cumbersome. We chose to instead utilize a recently developed approach involving Click-iT chemistry. Cells transfected with VP1 and VP2 expression plasmids were metabolically labeled (or mock-labeled) with myristic acid derivatized with an azide group. Pseudovirions were extracted from the cells, and then reacted with a Tetramethylrhodamine (TAMRA) fluorochrome linked to an alkyne. The highly specific copper-catalyzed reaction between the azide and alkyne permits in-gel fluorescent detection of myristoylated proteins following SDS-PAGE. The same gel that is examined for TAMRA fluorescence can then be stained for total protein. The results clearly indicate that the MCV VP2 protein is myristoylated, as only VP2 of the metabolically labeled pseudovirus displayed significant TAMRA fluorescence (Figure 9). In previously studied polyomaviruses, the glycine at position two of the VP2 protein sequence is required for covalent attachment of a myristoyl group [10]. Mutation of MPyV VP2 Gly2 has previously been shown to reduce the infectivity of virions [9], [11]. We mutated the Gly2 of MCV to valine (G2V), serine (G2S), or phenylalanine (G2F). Pseudovirions were produced and purified with the mutated VP2, then examined by SDS-PAGE (Figure 10A). Infectivity of the mutant pseudoviruses was reduced by ∼10 fold on 293TT cells (Figure 10B). VP2-null MCV pseudovirions displayed a ∼20–30 fold decrease in 293TT cell transduction relative to high-VP2 occupancy pseudovirions. Thus, myristoylation is an important feature of the MCV VP2 protein. A large number of human and animal polyomavirus species have been discovered in recent years, and the family Polyomaviridae now includes about 65 known species. Inspection of VP2 protein sequences reveals that more than a quarter of known polyomavirus species lack the consensus VP3 N-terminal MALXXΦ motif (Table S1 and Figure S1 alignment position 170–185). For the purposes of phylogenetic analyses, we defined polyomavirus species with a clear homolog of the VP3 N-terminal motif as “VP3+” and species lacking the motif as “VP3-less. ” When mapped onto a phylogenetic tree drawn based on an alignment of the complete nucleotide sequences of polyomavirus genomes, VP3-less species cluster together into a discrete monophyletic clade (Figure 11). The VP3-less clade is sub-divided into two separate monophyletic lobes. Similar patterns were observed for VP3-less species in phylogenetic trees drawn based on alignments of VP1, VP2, and Large T antigen protein sequences (data not shown). A simple model for these results is that members of the VP3-less clade all descended from a single viral ancestor that was not shared with the VP3+ viruses. In other words, the loss of VP3 appears to have involved a single bottlenecked historical event, as opposed to having arisen repeatedly through convergent evolution. The VP2 ORFs of VP3-less species are strikingly shorter than those of VP3+ species. VP3-less VP2 ORFs range from 229–243 codons, while VP3+ VP2 ORFs range from 304–416 codons (Figure S4). VP3-less species also tend to encode relatively long VP1 ORFs (435±44 codons in VP3-less species versus 372±22 codons for VP3+ species, with error representing one standard deviation). The shorter VP2 ORF length of the VP3-less viruses is largely attributable to a sharply defined deletion of a ∼50 amino acid segment within the C-terminal 1/3rd of all VP3+ VP2 proteins (see Figure S1, alignment positions 300–390). Although the missing segment of VP2 has not previously been assigned any function, it includes the methionine codon that initiates translation of VP4, a cytolytic protein expressed by SV40 [48]. The deleted segment contains a conserved extended sequence motif that can be summarized as YxxLxxYYxxL (x) PxxPxxxR. There are no discernible homologs of the deleted VP2 segment in VP1 proteins (see Discussion). Intriguingly, the L2 minor capsid proteins of many papillomaviruses encode a motif (YYxxLxPxxP) similar to the core of the VP2 motif. In addition to the ∼50 amino acid deletion, the VP3-less polyomaviruses, as well as a handful of VP3+ polyomavirus species, are missing an additional ∼30 amino acid segment at the C-terminus of VP2. This patch is highly basic, and a nuclear localization signal (NLS) has been identified in SV40 VP2/3 in this region [49]. In SV40, the VP2/3 NLS overlaps a proposed DNA-binding motif [50]. MPyV VP2/3 proteins appear to have a truncated C-terminus, consistent with the observation that nuclear localization of MPyV VP2/3 requires co-expression of VP1 [51]. Various polyomavirus VP1 and VP2 protein sequences were analyzed using an NLS-prediction algorithm, cNLS Mapper [52]. The algorithm correctly predicted the NLS phenotypes of SV40 and MPyV VP1 and VP2/3 proteins and predicted an MPyV-like phenotype for MCV, with a strong NLS motif predicted near the N-terminus of MCV VP1 and no predicted NLS within MCV VP2. Like MCV, all VP3-less polyomaviruses encode predicted NLS sites near the N-terminus of VP1 and no predicted NLS within VP2 (Table S1 and Figure S5). In contrast, nearly all VP3+ polyomaviruses carry predicted strong NLS sequences near the C-terminus of VP2 (with MPyV being a noteworthy exception). About half of VP3+ species do not encode a predicted NLS within VP1. To confirm the predicted NLS phenotypes of MCV VP1 and VP2, we examined the subcellular localization of each protein using confocal microscopy in 293TT cells (Figure 12). MCV VP2 primarily appeared in large, peri-nuclear cytoplasmic punctae when expressed alone. In contrast, VP1 expressed alone showed a diffuse nuclear pattern with apparent enrichment around the interior rim of the nucleus. When VP1 and VP2 were co-expressed in 293TT cells, there was a clear shift in the majority of VP2 localization to a diffuse nuclear pattern. MPyV, SV40, BKV, and JCV have been studied extensively in the roughly half-century since their discovery. MCV is very distantly related to these better-studied viruses and thus presents an important opportunity to learn which biological features are shared among all polyomaviruses and which features have diverged during the course of the family' s evolution. At the time MCV was discovered, it was the only example of a polyomavirus with a VP2 protein lacking the consensus MALXXΦ motif that forms the N-terminus of known VP3 proteins. It was thus unclear whether MCV encoded an unusual VP3 protein or entirely lacked a VP3 protein. In this work, we show conclusively that MCV lacks a VP3 minor capsid protein. This feature of MCV appears to be typical of an emerging clade of polyomaviruses encompassing more than a quarter of currently known species. Although high-sensitivity western blotting did not detect VP3 proteins in MCV virions or in MCV-infected cells it is difficult to absolutely rule out the possibility that VP3-like proteins might be expressed under some conditions in vivo or that VP3-like proteins might play a non-virion role in the viral life cycle. One example of a possible non-virion role is illustrated by the VP4 protein of SV40. VP4 is not incorporated into SV40 virions, but instead exhibits a cell-lytic activity that is important for virion release [48]. SV40 VP2, VP3, and VP4 proteins have all been shown to permeabilize membranes [53]. In contrast, we have not observed cytopathic or lytic effects in cell cultures replicating and producing native MCV virions, nor have others [41]–[44]. In addition, we observed no change in cell proliferation when VP2 or possible VP3 proteins were deleted from the replicating virus. Consistent with these observations, transient overexpression of individual VP2 or VP3 proteins in the presence or absence of VP1 does not appear to cause cytolytic effects in 293TT cells (data not shown). It thus seems unlikely that any products of the MCV minor capsid protein ORF could play an SV40 VP4-like role in the MCV life cycle. It is possible that MCV, like papillomaviruses, does not require an active lytic process for release. It is currently uncertain which cell types are productively infected by MCV, but the virus appears to reside in the skin, where the natural host process of keratinocyte desquamation might facilitate release of virions into the environment [21], [54], [55]. Most current literature states that polyomavirus virions contain an average of one minor capsid protein (either VP2 or VP3) per capsomer [56]. However, a careful review of older primary literature suggests that this established view is probably incorrect. An early report by Estes, Huang, and Pagano used radiolabeling and spectrophotometry to show that the molar ratio of VP1∶VP2∶VP3 in SV40 virions is 6∶1∶1. 5 [57]. This ratio measurement implies a total of about two minor capsid protein molecules per pentameric VP1 capsomer. A later study by Lin and colleagues asserted, as data not shown, that SV40 virions have a capsid protein ratio of 16. 88∶1∶2. 65, indicating about one molecule of minor capsid protein per capsomer [58]. Lin and colleagues' conflicting claim was based on Coomassie-stained SDS-PAGE gels, raising the caveat that the ratio calculation may have been distorted by differential staining of VP1, VP2, or VP3 by the Coomassie dye. Our current analyses of native MCV virions using a less variable staining reagent appear to confirm Estes and colleagues' original biophysical results showing that native polyomavirus virions can accommodate an average of two molecules of minor capsid protein per VP1 capsomer. Our findings show that MCV VP2 is required for efficient transduction of some cell types but not others. A model for this finding is that, in cells where VP2 is beneficial, there is a barrier to entry that VP2 helps overcome. During infection of cells such as UACC-62 or SK-MEL-2, which VP1-only pseudoviruses transduce efficiently, this barrier seems not to exist or is not encountered. Interestingly, all apparently VP3-less viruses encode larger VP1 proteins than those found in most putatively VP3+ species (Figure S4). We speculate that the larger VP1 proteins of the VP3-less viruses may have assumed some of the functions of the minor capsid proteins, thus making VP2 conditionally dispensable and rendering VP3 entirely dispensable. If so, then the expansion of VP1 would presumably have evolved prior to the internal VP2 deletion events that appear to have destroyed VP3 and truncated C-terminal portions of VP2. MCV VP2 does not exhibit strong nuclear localization unless it is co-expressed with VP1 (Figure 12). Analyses using an NLS prediction algorithm suggest that, like MCV, all VP3-less polyomavirus species encode an NLS near the N-terminus of VP1 and do not encode an NLS within VP2. In contrast to MCV, nearly all VP3+ species have a predicted NLS near the C-terminus of VP2, and about half of these species do not contain a predicted NLS within VP1. Polyomavirus genomes are organized such that the coding regions for the N-terminus of VP1 and C-terminus of VP2 overlap (Table S1). It is thus possible to envision frameshift mutations or tandem duplication events that would result in the transfer of NLS motifs (and possible nearby DNA-binding motifs) from the C-terminus of VP2 to the N-terminus of VP1 (or vice versa). This might represent a concrete example of transfer of a VP2 function to VP1 during the evolution of VP3-less species. Intriguingly, several polyomavirus species show evidence of sequence duplication in or near the VP2/VP1 overlap region (noted in Table S1). Polyomaviruses are believed to be constrained by the small size of their ∼5 kb genomes and it seems unlikely that unnecessary genes would be maintained over evolutionary timeframes. Thus, the fact that the VP2 gene is maintained in all known polyomavirus species suggests that VP2 plays an important role in MCV biology in vivo. However, it remains possible that the apparently VP2-independent infectious route is also relevant to pathobiology. A recent publication from Muñoz and colleagues documents an outbreak of hamster polyomavirus (HaPyV) -induced lymphoma [59]. Interestingly, most HaPyV genomes isolated from lymphoma cells were found to have deletion mutations encompassing the N-terminus of VP2 and, in one instance the N-terminus of VP3 as well. Although it is unclear whether these natural VP2/3 deletion mutations play a causal role in lymphoma development, our data open the possibility that loss of VP2 might lead to altered viral tropism, conceivably with pathogenic consequences. It is not yet known if the cell types MCV productively infects in vivo are the same cells that MCV pathogenically transforms, nor whether the virus infects both classes of cells via the same entry mechanism. We have previously examined the transduction efficiency of MCV in four MCC lines and found each MCC line to be highly resistant to transduction with, at best, ∼2% of cells becoming transduced with maximum doses of MCV pseudovirus [42]. Although it may seem counter-intuitive that MCC lines carrying integrated MCV genomes are resistant to MCV infection, it is important to note that progression to malignancy and adaptation to culture are typically associated with a large number of genetic, epigenetic, and phenotypic changes [60]. This problem is illustrated by a current debate over whether MCC tumors arise from differentiated Merkel cells, epidermal stem cells, or B cells [61], [62]. To partially address the problem that cultured cell lines may not faithfully represent their tissue of presumed origin, we examined an assortment of different cell lines that MCV readily transduces. Given the abundance of MCV in human skin, it is intriguing that the two cell lines that MCV VP1-only particles transduced efficiently were derived from skin tumors (melanoma). However, further analysis would be required to determine if this reflects a common feature of primary melanocytes in vivo or is simply a coincidence. Initially, we hypothesized that VP2 may direct the trafficking of MCV down a productive infectious entry pathway, but our data suggest that the MCV VP2 protein does not alter the bulk trafficking of particles. VP2 also does not affect packaging of viral DNA nor does VP2 affect the binding of particles to cell surfaces. Thus, it appears that VP2 functions at a step late in entry. Given our observation that myristoylation of VP2 is important for its function during entry, it is tempting to postulate that this hydrophobic modification is important for disruption of cellular membranes during entry. However, our microscopy data are equally consistent with other possible functions of the MCV VP2 protein during the late phase of infectious entry. A controversial question in the taxonomy of the viral family Polyomaviridae is whether polyomavirus species usually arise through co-evolution with a particular host animal lineage, or whether individual polyomavirus species have sometimes evolved after being productively transmitted between distantly related animal families [17], [63]. Finding highly similar polyomavirus sequences in distantly related animal families would constitute strong evidence for transmission between animal families. Our phylogenetic analysis shows that the VP3-less clade can be divided into two separate lobes (Figure 11). Within each lobe, the phylogenetic relationships of individual polyomavirus species roughly resemble the phylogenetic relationships of the host animal families in which the viruses were discovered. Most notably, there are no examples of bat-derived viruses co-occupying short branches with primate-derived viruses. A simple model for the observed phylogenetic relationships would be that two distinct VP3-less polyomavirus species both infected the most recent common ancestor of placental mammals (or its near relatives), and the two VP3-less lineages evolved within different host animal families during the ensuing ∼100 million years. Such a languid evolutionary pace is consistent with molecular clock estimates previously proposed for papillomaviruses, which also have a small circular dsDNA genome that is replicated by host cell polymerases [64]. Taken together, the results clearly indicate that the MCV VP2 gene typifies a large clade of polyomaviruses that differ significantly from previously examined polyomaviruses, in that they do not encode a VP3 protein. Awareness of this fundamental difference between MCV and other polyomaviruses will be important for future studies aiming to understand the pathobiology of MCV and may guide efforts to prevent this extremely common and occasionally pathogenic infection. 293TT cells [65] were maintained in DMEM (Mediatech, Inc.) with 10% fetal bovine serum (FBS, Sigma), Glutamax-I (Invitrogen) and MEM non-essential amino acids (Invitrogen) supplemented with hygromycin B (250 µg/ml; Roche). 293-4T cells [41] were maintained in the same medium as 293TT, except supplemented with zeocin (100 µg/ml; Invitrogen) and blasticidin S (5 µg/ml; Invitrogen) rather than hygromycin. NCI/ADR-RES cells, A549 cells, SK-MEL-2 cells, UACC-62 cells, RXF 393 cells, OVCAR-4 cells, and T-47D cells were obtained from the Developmental Therapeutics Program (NCI/NIH). These lines were cultured in RPMI 1640 medium (Mediatech, Inc.) supplemented with 5% FBS and 1% Glutamax-I. Anti-VP2 polyclonal serum was raised by immunizing a rabbit (Lampire Biological Products) with purified recombinant MCV VP2 immunogens expressed in bacteria. The rabbit was initially primed with a maltose binding protein (MBP) -VP2 fusion protein expressed from plasmid pMVP2M, which was made by PCR-mediated transfer of the VP2 ORF of ph2m [15] into the HindIII and BamHI sites of pMXB10 (NEB). The fusion protein was expressed in T7 Express lysY/Iq E. coli (NEB) and purified over amylose resin according to the manufacturer' s instructions. The VP2 PCR product incorporated a recognition site for tobacco etch virus (TEV) protease between the MBP and VP2. The purified fusion protein was cleaved using an improved TEV protease variant, S219P, expressed from plasmid pRK792 [66], which was a generous gift from David Waugh (NCI). The rabbit was primed with the MBP-VP2 immunogen in complete Freund' s adjuvant, boosted once with MBP-VP2 in incomplete Freund' s adjuvant, then given a final boost with a 6× histidine (His) -tagged VP2 protein expressed in bacteria using plasmid pHisMVP2. This plasmid was generated by transferring an EagI/NheI fragment of ph2m into pProEXhta (Invitrogen) cut with EagI and XbaI. The His-tagged booster VP2 immunogen was not treated with TEV protease. The final hyper-immune serum did not show any detectable neutralizing activity against MCV reporter pseudovirions (data not shown). To validate the VP2 antiserum, we generated MCV pseudovirion standards by co-transfecting 293TT cells with plasmids containing codon-modified versions of the MCV VP1 and VP2 genes. We also produced pseudovirions using a separate VP3-only expression plasmid. Comparison of the VP2 and VP3 pseudovirus standards in SYPRO Ruby stained gels (data not shown) and western blots showed that the VP2 antiserum recognizes both VP2 and VP3 with similar efficiency. MCV virions were produced using previously described methods [42], with slight variations. Briefly, 2. 5 million 293TT cells were plated in a 25 cm2 flask the day prior to transfection. The cells were co-transfected with 5 µg re-ligated MCV isolate R17b (GenBank accession number HM011556. 1) genomic DNA as well as 3. 5 µg of MCV small t antigen (pMtB) and 4 µg of MCV large T antigen (pADL*) expression plasmids. Transfected cells were expanded for 5–6 days, and virions were harvested and purified over Optiprep ultracentrifuge gradients as previously described. For propagative amplification of native MCV virions, 293-4T cells were infected with these native MCV virions produced by transfection. Propagated virions were harvested from a portion of the infected culture every few days beginning on day 10 after initial virion inoculation. Infected cells were lysed, clarified, and purified over Optiprep ultracentrifuge gradients as previously reported [41]. Quantification of encapsidated viral DNA in WT and mutant native virions was determined by first digesting 5 µl of viral capsids from each purified preparation with 20 mM Tris, pH 8. 3,20 mM DTT, 20 mM EDTA, 0. 5% SDS, and 0. 2% proteinase K for 20 min at 50°C. DNA was then purified out of the digest using the QIAquick PCR purification kit (QIAGEN), and qPCR was performed with primers 5′-GCTTGTTAAAGGAGGAGTGG-3′ and 5′-GATCTGGAGATGATCCCTTTG-3′ to quantitate viral genomic DNA using the DyNAmo SYBR Green qPCR kit (NEB). Comparison to pMCV-R17a DNA standards permitted calculation of the concentration of viral DNA in each virus stock. Mutation of MCV isolate R17b genomic DNA to create the dVP2 and dVP3 (a/b/d) mutants, was accomplished using overlap PCR. PCR products were transferred into the pMCV-R17b vector using AvrII and BsrDI restriction site cloning. Approximately 400 µl of Optiprep-purified virions from a mixture of day 13 and day 17 post-infection harvests were immunoprecipitated with the anti-VP1 monoclonal antibody MV23 [67], which had been pre-complexed with protein G Dynabeads (Invitrogen). The sample was then eluted in NuPAGE LDS sample buffer (Invitrogen) with 40 mM DTT and electrophoresed through a NuPAGE Novex 4–12% bis-tris gel alongside the VP2 and VP3 standards described below. The gel was transferred to nitrocellulose and western blotted using rabbit VP2 antiserum diluted 1∶100. The ratio of VP1 to packaged viral DNA was nearly identical in WT and dVP2 viral preparations, such that an equal concentration of VP1 (0. 05 ng virus/well, ∼2. 4×108 copies) was added to 2×105 293-4T cells that were brought into suspension by trypsin treatment. To measure cell binding by the WT and mutant virus, each was incubated with cells for one hour at either 4°C or 37°C in an untreated, round-bottom 96 well plate. Cells were then washed three times before freezing in modified Hirt buffer I [68]. To examine dVP2 infectivity, WT or dVP2 virions were added in quadruplicate to cells that were just plated in a 24-well plate. As a control, neutralizing rabbit polyclonal antibody raised against MCV capsids was added to one of the quadruplicate wells, while pre-immunization serum was added to the other three. Roughly 24 hours later, all wells were incubated in trypsin to resuspend the cells, the trypsin was neutralized with growth medium, and cells were pelleted. Medium was changed in the sample containing neutralized virus and in two non-neutralized samples, and cells were re-plated in a larger well for continued growth. At this time, neutralizing serum was replaced in the population that had received it previously. Additionally, neutralizing serum was added to another infected population to prevent further spread of the virus. One infected cell population remained exposed only to pre-immunization serum. The fourth sample (Day 1) was resuspended in Hirt buffer I, then frozen. A total of four days after virus inoculation, the re-plated cells were collected by trypsinization and resuspended in Hirt buffer I. Low molecular weight DNA from all samples was isolated by modified Hirt extraction [68], and the number of genomic copies of MCV in the samples was determined by quantitative PCR as described previously [41]. Measurements of dVP3 (a/b/d) infection were performed similarly to dVP2, but without the added neutralization controls. A volume of virus equaling 2×106 copies of genomic DNA was added to 2×105 293-4T cells that were just plated in duplicate. The next day, cells were resuspended, pelleted and then either resuspended in Hirt buffer I and frozen, or resuspended in growth medium and re-plated for continued growth. On the fourth day after virus inoculation, the final set of cells was collected and all samples were subjected to modified Hirt extraction and MCV genome copies were measured by qPCR. MCV reporter pseudoviruses were produced using methods reported previously [15], [41]. In brief, 293TT cells [65] were transfected with plasmids expressing codon-modified versions of the major and minor capsid genes, as described below. Pseudoviruses carrying GFP reporter plasmids were co-transfected with pYafw [65] and/or pEGFP-N1 (Clontech). These plasmids express the GFP reporter under control of recombinant human elongation factor 1α promoter or cytomegalovirus (CMV) immediate early promoter, respectively. Separate pseudovirus stocks carried a mixture of reporter plasmids carrying a Gaussia luciferase reporter gene under control of CMV promoter (pCLG) or human elongation factor 1α (pLGluc). Except where indicated, transfected cells were harvested forty-eight hours after transfection. Cells were lysed in PBS supplemented with 0. 5% Triton X-100 as well as Benzonase (Sigma) and Plasmid Safe (Epicentre) nucleases. Pseudovirions were purified over Optiprep gradients according to previously reported methods. The pseudoviruses containing WT or mutated VP2 proteins that were analyzed in Figure 3 were not Optiprep purified, but the cell lysates containing matured pseudovirions were clarified by centrifugation at 5,000×g. A different detergent, Brij 58 (0. 35%; Sigma), which is less cytotoxic than Triton X-100, was used to lyse the cells at the time of harvest in order to preserve the health of target cells during transduction experiments. The mutation of M46 in the VP2 protein was accomplished by PCR with ph2m [16] as a template. The XhoI restriction site used in cloning fell near enough to the mutation site to be incorporated into the primer. Degeneracy was engineered into the primer and multiple clones were sequenced in order to obtain the various changes at this site. Pseudovirions were produced by transfection of 293TT cells with individual M46 mutants or the WT ph2m plasmid with pwM [16] and the pEGFP-N1 reporter plasmid. Mutation of VP2 Gly2 was also accomplished with PCR using ph2m as the template and incorporation of a cloning restriction site (SnaBI) in the primer used for mutagenesis. Again, a degenerate primer was used to obtain multiple changes at the mutation site. In addition, the sequence upstream of VP2 was modified to improve the Kozak sequence. A sister construct, phK2m, in which the Kozak sequence was improved upstream of the WT VP2 gene, was also made. The primer used for construction of the myristoylation mutants is as follows: 5′-TTTTTTACGTAATATTGCCGCCACCATGKYNGGGATCATTACCCTGCTCGC-3′. Pseudovirions were produced by transfection of 293TT cells with individual Gly2 mutants or the WT ph2m plasmid with pwM and the pEGFP-N1 reporter plasmid. MCV pseudoviruses with varying levels of VP2 incorporated were produced by transfection of pwM (VP1 only), pwM2m [67] (low VP2), or pwM2m plus ph2m (high VP2). In addition, the pLGluc and pCLG reporter plasmids were co-transfected with each of the MCV capsid protein plasmids. The pwM and ph2m plasmids, as well as both Gaussia luciferase plasmids, have within their plasmid backbones the gene for EGFP under the control of the SV40 promoter. The ph2m and pwM plasmids may be packaged by the pseudovirus particles in addition to the intended reporter plasmids, although pwM is slightly too large (6. 6 kb) for efficient packaging. Since the number of luciferase and EGFP reporter gene copies relative to VP1 potentially differs between these pseudovirus preparations, the concentration of both reporter genes in each pseudovirus preparation was quantified by qPCR. To do this, the concentration of VP1 was first determined by densitometry of SYPRO Ruby-stained SDS polyacrylamide gels. DNA was then extracted from 100 ng of VP1 and analyzed by qPCR as above using primers specific for the reporter genes [41]. Comparison to plasmid standards permitted calculation of the concentration of reporter plasmid in each virus stock. BKV pseudovirions were produced with the pLGluc and pCLG reporter plasmids and analyzed similarly to the MCV pseudovirions with varying VP2 levels. In addition to the reporter plasmids, BKV VP1-only pseudovirion production used the pwB plasmid, BKV VP1+VP2 used pwB2b and ph2b, BKV VP1+VP3 used pwB3b and ph3b, and BKV VP1+VP2+VP3 used pwB2b, pwB3b, ph2b, and ph3b. Harvest of BKV from transfected cells was performed similarly to MCV, except that cell pellets are treated with neuraminidase. The yield of each pseudovirus was similar, and DNA was extracted from 2 µl of each preparation for qPCR analysis of reporter gene content. The purified pseudovirions used as protein standards to verify reactivity of the VP2 antiserum with both possible minor capsid proteins in western blot were produced by co-transfecting 293TT cells with the plasmid pwM with pC2m (VP2 under control of a CMV promoter) and/or pC3m (VP3 under control of a CMV promoter). The concentration of VP2 and VP3 in the pseudovirus stocks was first determined by densitometry of SYPRO Ruby-stained SDS polyacrylamide gels with comparison to BSA standards (data not shown). Maps of plasmids used in this work and detailed virus production methods are available from our lab website http: //home. ccr. cancer. gov/Lco/ Cells transduced with MCV or BKV pseudoviruses were plated the day prior at the following concentration in a 96-well plate: 293-TT = 1. 5×104, A549 = 7. 0×103, SK-MEL-2 = 1. 0×104, UACC-62 = 5. 0×103, NCI/ADR-RES = 1. 0×104, RXF 393 = 7. 0×103, OVCAR-4 = 7. 0×103, T-47D = 1. 0×104. Cells were subconfluent at the time of inoculation, and transduced cells were incubated for three days prior to analysis. A two-fold dilution series of each virus stock was analyzed and, within each experiment, virus stocks were normalized to each other by VP1 concentration or reporter gene content. The top dose of VP1/well ranged from 20–50 ng, and VP1 standardization was used in all but the analysis of varying levels of MCV VP2 in pseudoviruses and the BKV +/− VP2 and VP3 pseudovirus infection. As indicated above, BKV pseudovirions and the MCV pseudovirions made with no, low, or high levels of VP2 were normalized by reporter gene content. The top dose of GFP reporter gene used was 3. 0×108 copies/well for MCV and 6. 0×108 copies/well for BKV. To measure pseudovirus-mediated transduction of the GFP gene, adherent cells were incubated with trypsin to detach them from the plate, transferred to an untreated 96-well plate and suspended in wash medium (WM; DPBS with 1% FBS, antibiotic-antimycotic, and 10 mM HEPES, pH 8). Cells were then analyzed by flow cytometry for GFP expression in a FACS Canto II with HTS (BD Biosciences). Cell Proliferation Reagent WST-1 (Roche Applied Science) was added to the medium of 293-4T transfected with MCV isolate R17b genomic DNA, dVP2, dVP3 (a/b/d), or pMtBS (MCV small t antigen). We have previously observed the toxic effects of transient MCV small t antigen over-expression [41], [69], so pMtBS was used as a positive control. Cells were transfected in a 24 well plate and split the following day into a 96 well plate in triplicate and a new 24 well. Five days later, WST-1 was added to the 96 well plate and absorbance was measured at several time points from 30 minutes to 3 hours. The cells in the 24 well plate were lysed and analyzed by SDS-PAGE and western blot for MCV VP1 (1∶8,000 dilution of rabbit VP1 antiserum) or MCV VP2 (1∶100 dilution of rabbit VP2 antiserum). MCV pseudovirions containing only VP1 or VP1+VP2 were produced by transfection of 293TT cells with pwM or pwM2m and ph2m, respectively. The plasmids pIaw [70], ph2b and ph3b [41] were used to produce BKV pseudovirions. The pLGluc reporter plasmid was also included in each transfection mixture. Six hours after transfection, the medium of each culture was replaced with fresh medium containing 50 µM EdU-alkyne (Click-iT EdU Imaging Kit, Molecular Probes/Invitrogen). MCV pseudovirions were harvested ∼48 hours after transfection and purified normally using the methods described above. BKV pseudovirion harvest has been published previously and differs from MCV harvest by the addition of neuraminidase V during maturation [70]. The VP1 concentration of each pseudovirus preparation was determined by SYPRO Ruby stain of an SDS-polyacrylamide gel and densitometry with comparison to BSA standards. For fluorescent imaging of EdU-labeled pseudovirions during cell entry, 3. 5×104 NCI/ADR-RES cells were plated on glass coverslips in a 24-well plate. The next day, MCV VP1-only or MCV VP1+VP2 pseudovirions equaling 20 ng of VP1/well or BKV pseudovirions equaling 40 ng VP1/well were inoculated onto the cells. After ∼48 hours, the cells were fixed with 2% paraformaldehyde and labeled with Alexa Fluor 488-azide using the Click-iT EdU Imaging Kit (Molecular Probes) according to the manufacturer' s instructions. Co-staining was performed with an anti-LAMP-1 mAb (H4A3; Developmental Studies Hybridoma Bank) diluted 1∶300 or anti-ERp72 rabbit polyclonal antibody (Stressgen) diluted 1∶300. Alexa Fluor-488 conjugated MCV VLPs [67] were used to examine VP1 localization. Coverslips were inverted and mounted with Prolong Gold (Molecular Probes) containing DAPI. Images were acquired with a Zeiss LSM 780 confocal system interfaced with a Zeiss Axio Observer microscope. Images were collated with Adobe Photoshop Elements software, where the red levels of Figure 6A and green levels of Figure 6B were adjusted equally among images. 293TT cells were transfected with plasmids encoding MCV VP1 (pcM), MCV VP2 (p2mw) or MCV VP1 and VP2 (pMmw) were plated on poly-D-lysine/laminin coated coverslips. They were fixed with 2% paraformaldehyde and incubated in 0. 1% Brij 58 in PBS with rabbit VP2 antiserum (1∶200) and mouse monoclonal MV16 hybridoma supernatant (1∶10). Alexa fluor (AF) secondary antibodies were used for VP1 (AF-594) and VP2 (AF-488) detection. Coverslips with mounted with Prolong Gold containing DAPI. Images were acquired with a Zeiss LSM 710 NLO confocal microscope and collated with Adobe Photoshop Elements software. Detection of myristoylation was performed by producing pseudovirions in cell growth medium supplemented with Click-iT myristic acid-azide (Molecular Probes) to a final concentration of 25 µM. Metabolic labeling was allowed to proceed for ten hours, and then pseudovirions were harvested using standard methods. The concentration of VP1 in this pseudovirus preparation and an unlabeled MCV pseudovirus preparation was determined by densitometry of a SYPRO Ruby-stained SDS-PAGE gel. An amount of each pseudovirion preparation equaling 1. 9 µg of VP1, was mixed with 42 mM Tris, pH 8 and 0. 83% SDS in a final volume of 60 µl. Tetramethylrhodamine (TAMRA) -alkyne was then reacted with the samples according to the manufacturer' s instructions using the Click-iT Protein Reaction Buffer Kit. Proteins were then precipitated by methanol/chloroform extraction, and then resuspended in NuPAGE LDS sample buffer with 40 mM DTT. Proteins were separated by SDS-PAGE of the samples, and the gel was imaged under green epi-illumination in an ImageQuant LAS 4000 (GE Healthcare). The same gel was then SYPRO Ruby stained and imaged under epi-illumination with blue light. MacVector version 12. 6 software was used to perform a MUSCLE alignment [71] on the nucleotide sequences of complete polyomavirus genomes. For display, each polyomavirus species was assigned a nickname based on the common English name for the animal host in which the virus was discovered or based on commonly used polyomavirus name abbreviations. A naming key, including accession numbers, is provided in Table S1. Curated compilations of sequences used for the analyses can be found at the following link: <http: //home. ccr. cancer. gov/Lco/PyVE. asp>. NLS prediction was performed using the following website: <http: //nls-mapper. iab. keio. ac. jp/cgi-bin/NLS_Mapper_form. cgi> [52]. Bipartite NLS searching was restricted to terminal 60-amino-acid regions. For VP1, sequences with any NLS score of 4. 5 or greater were considered NLS positive. In nearly all instances, the predicted VP1 NLS involved an N-terminal portion of the protein. Two exceptions were JCV and California sea lion VP1 proteins, for which the algorithm predicted only an internal NLS. The predicted JCV VP1 NLS, 282-QLRKRRVKN-291, is poorly surface-exposed in the crystal structure of the protein [72], consistent with the poor nuclear localization of JCV VP1 [73]. For VP2, sequences with a monopartite NLS score of greater than 3 were considered NLS-positive. A neighbor-joining tree was constructed using random resolution of ties and Jukes-Cantor maximum likelihood method with proportional distribution of gaps. The tree was arbitrarily rooted on finch polyomavirus. Midpoint rooting or rooting on other Avipolyomaviruses or Wukipolyomaviruses also resulted in the appearance a monophyletic VP3-less clade, with similar relationships among members within the clade. For the tree shown in Figure 9, the root of the VP3-less clade and the two VP3-less sub-lobes each had bootstrap values of 100. Assignment of the coloring scheme in Figure 9 was based on general comparisons between various sources, including Wikipedia and [74]–[76]. Greater weight was assigned to newer molecular phylogeny-based estimates than fossil record-based estimates.

Merkel cell polyomavirus (MCV or MCPyV) is a recently discovered member of the viral family Polyomaviridae. The virus plays a causal role in Merkel cell carcinoma, a highly lethal form of skin cancer. MCV encodes a major capsid protein, VP1, which forms the non-enveloped surface of the virion. Other polyomavirus species encode two minor capsid proteins, VP2 and VP3, which associate with the inner surface of the capsid and facilitate infectious entry. In this report we show that MCV does not have a VP3 minor capsid protein. Sequence analyses suggest that more than a quarter of known polyomavirus species share MCV' s lack of a VP3 protein. In contrast to VP3, VP2-knockout MCV mutants displayed dramatically reduced infectivity. Consistent with native virion findings, MCV pseudovirions lacking VP2 or carrying mutations in the VP2 myristoylation motif displayed reduced infectivity on several cell lines. Puzzlingly, MCV pseudoviruses lacking VP2 successfully transduced other cell lines with high efficiency. Taken together, the data show that the lone MCV minor capsid protein, VP2, plays an important role during infectious entry into some cell types, but is dispensable for entry into other cell types.

Abstract Introduction Results Discussion Materials and Methods

viral entry viral transmission and infection virology biology microbiology viral structure viral evolution

2013

The Merkel Cell Polyomavirus Minor Capsid Protein

Genetic prion diseases are late onset fatal neurodegenerative disorders linked to pathogenic mutations in the prion protein-encoding gene, PRNP. The most prevalent of these is the substitution of Glutamate for Lysine at codon 200 (E200K), causing genetic Creutzfeldt-Jakob disease (gCJD) in several clusters, including Jews of Libyan origin. Investigating the pathogenesis of genetic CJD, as well as developing prophylactic treatments for young asymptomatic carriers of this and other PrP mutations, may well depend upon the availability of appropriate animal models in which long term treatments can be evaluated for efficacy and toxicity. Here we present the first effective mouse model for E200KCJD, which expresses chimeric mouse/human (TgMHu2M) E199KPrP on both a null and a wt PrP background, as is the case for heterozygous patients and carriers. Mice from both lines suffered from distinct neurological symptoms as early as 5–6 month of age and deteriorated to death several months thereafter. Histopathological examination of the brain and spinal cord revealed early gliosis and age-related intraneuronal deposition of disease-associated PrP similarly to human E200K gCJD. Concomitantly we detected aggregated, proteinase K resistant, truncated and oxidized PrP forms on immunoblots. Inoculation of brain extracts from TgMHu2ME199K mice readily induced, the first time for any mutant prion transgenic model, a distinct fatal prion disease in wt mice. We believe that these mice may serve as an ideal platform for the investigation of the pathogenesis of genetic prion disease and thus for the monitoring of anti-prion treatments. Inherited prion diseases, such as gCJD and Gerstmann–Sträussler–Scheinker (GSS), are autosomal dominant disorders linked to mutations in the gene encoding the prion protein (PrP), denominated PRNP [1], [2]. The largest focus of gCJD was identified among Libyan Jews carrying a missense mutation in codon 200 of PRNP (substituting lysine for glutamate, E200K) [3], [4]. This same mutation was also found in other communities around the world [5]. As of today, therapeutic intervention in human prion diseases has failed [6] [7]. Indeed, some protocols reduced the rate of patients' deterioration for short periods of time [8], but none could hope to reverse the severe neurological deficits apparent already at diagnosis. We therefore propose that efforts should be directed mostly to develop preventive treatments for subjects at risk, as is the case for asymptomatic carriers of genetic prion diseases. Candidate anti-prion reagents will need to be tested in transgenic models mimicking gCJD. Such transgenic mice should succumb spontaneously to neurological disease in a high attack rate and in a short time frame, allowing for long term treatments and measurable delay of onset well within the life span of the animals. The model mice should also present prion related biochemistry and pathology, and if possible transmit disease directly to wt animals, as is the case for humans suffering from gCJD [9] [10]. Indeed, several animal models of genetic prion disease were generated in the past, thereby demonstrating that late onset and spontaneous genetic human prion diseases can be reconstructed in mice [11], [12]. While very useful in the study of prion disease pathogenesis, not all these models presented all the properties described above. The first transgenic (Tg) mice imitating human genetic prion disease carried a P102L-PrP GSS mutation on a mouse background and succumb spontaneously to prion disease after about 4–6 months [13]. However these mice transmitted infectivity only to unique recipients [14], [15], and in addition presented poor PrP pathology. Tg lines mimicking the PrP insertional mutation [16], the A117V mutation [17], as well as both the CJD [18] and the FFI D178N [19] mutation presented prion-like clinical disease with low to marginal disease related PrP. The FFI D178N mice transmitted disease to mice overexpressing wtPrP as well as those expressing wtPrP with the 3F4 epitope, and the recipient mice developed prion-related neuropathology in the absence of disease related PrP [19]. Two Tg lines mimicking the E200K PrP mutation, one on a human PrP gene and another on a mouse PrP gene did not present disease or other prion related properties [20] [21]. In this work, we describe a transgenic mouse model for E200K gCJD expressing a chimeric mouse/human PrP [15], [22] both on a wt and a null PrP background, hereby denominated TgMHu2ME199K/wt and TgMHu2ME199K/ko respectively. The line on the wt background mimics most gCJD patients, who are heterozygous for the PrP mutation [2]. Mice from both lines presented progressive neurodegenerative disease starting from 5 to 6 month of age, deteriorated and died several months thereafter. Their brains comprise age related pathology characteristic of prion disease, such as gliosis and accumulated disease related PrP, which was shown by immunoblots to be resistant to digestion by high concentrations of proteinase K (PK). Most important, brain extracts from both lines transmitted prion disease to wt mice. We believe that these animals will play a significant role in the investigation of genetic prion disease pathogenesis and most important, in the development of novel anti-prion prophylactic treatments. TgMHu2ME199K on both a wt and a PrP ablated background were constructed (as described in the methods) by inserting an E to K substitution at position 199 of a chimeric mouse human (MHu2M) PrP construct. As of today a total of 300 mice were generated (240 on an ablated background and 60 on a wt background), and used for the different experiments described in this manuscript. These include characterization of clinical disease as well as investigation of kinetic of disease progression. We also studied pathological and biochemical prion disease properties of the Tg mice at different time points before and throughout disease progression and collected samples for expression and transmission studies. The most prominent symptom of disease, which appeared in all Tg MHu2M E199K mice already at 5–6 months of age, is an a-symmetric hind limbs weakness that develops with time to paraplegia. This sign was followed by leg clasping and lower body atrophy. Contrarily, some of the most characteristic clinical signs of prion symptoms, i. e. plastic tail and tremor were only apparent in some of the mice. Figure 1 depicts affected mice suffering from hind limbs plegia, lower body atrophy and leg clasping. While the mice in the figure are each from a different line (Tg/ko and Tg/wt), the different signs appear in all sick mice. The clinical symptoms of the TgMHu2ME199K mice by order of appearance are described in Table 1. Table 2 demonstrates the score we constructed out of these clinical symptoms for kinetic analysis of disease progression. The time point of death (score 5) was determined when a mouse was too paralyzed to reach food and water independently (according to local committee ethical requirements). To evaluate the kinetics of disease progression in these mice, a designated group of 62 TgMHu2ME199K/ko and 14 TgMHu2ME199K/wt, (half male, and half female) was followed carefully from birth throughout disease progression to death. Figure 2a shows the average age of disease onset (score 1) and disease end point (score 5) in the transgenic mice. Figure 2 b presents the severity of disease as related to age (each point represent the average score in groups of 2–8 littermates, which were averaged together to avoid individual differences), while figure 2 c demonstrates disease prevalence in these same groups as related to age. As stated above, our results indicate that all TgMHu2ME199K mice demonstrate first disease symptoms between the ages of 5 to 7 months old. No significant differences were observed in clinical and kinetic parameters between male and female mice. Small differences (non significant) in disease presentation and progression were observed between mice expressing the chimeric mutant PrP on null as compared to wt PrP background (figure 2a), however this may result from the smaller numbers of mice in the TgMHu2ME199K/wt group. Additional transgenic mice used in time course experiments showed similar disease parameters. We next investigated the levels of PrP expression in the brains of the TgMHu2ME199K mice. To this effect, mRNA samples purified from brains of wt and TgMHu2ME199K/ko 6 months old mice (4 for each group) were subjected to reverse transcriptase and subsequently to amplification by real time PCR of PrP and control genes (see methods). Figure 2d shows that while mRNA levels of PrP were 20 fold higher in the brains of the transgenic mice as compared to wt mice, the actual levels of the PrP protein, as tested by immunoblotting of brain homogenates with α PrP mAb 6H4, were only increased by 2 folds (figures 2e &f). Whether this discrepancy between the PrP mRNA and protein levels of mutant PrP in the Tg mice is of biological significance is unknown at this point. Four µm thick sections of formalin-fixed, paraffin-embedded brains and spinal cords of TgMHu2ME199K mice of different ages and gene array were evaluated for neuropathology and PrP immunoreactivity with different α PrP antibodies (see figure 3a for epitope description of all α PrP antibodies used in this project). Figure 3b depicts the results for 8 months old mice (at least 3 in each group) of the different lines (TgMHu2ME199K/ko, TgMHu2ME199K/wt, PrP0/0 and wt mice). Figure S1 presents results for TgMHu2ME199K/ko mice at different ages (3 in each group), which are also summarized in figure 4. None of the Tg mice brains exhibited inflammatory infiltrates, demyelination, axonal swellings, or abnormal neurites, in accordance with classical prion disease–related pathology [23]. The predominant form of disease related PrP immunoreactivity in the TgMHu2ME199K mice was an intraneuronal dot-like and granular immunostaining in widespread distribution but mainly in neurons of the spinal cord, basal ganglia, thalamus, frontal cortex, and in brainstem nuclei. This was detected by the C-terminally directed αPrP pAb RTC, and less so by αPrP mAb 12F10 (shown for human in Figure S1). Plaque-like or coarse PrP immunoreactivity was not seen in any of the sick TgMHu2ME199K mice. These patterns of immunopositivity, in particular the intraneuronal staining, is strikingly reminiscent of recently described human E200K gCJD [24]. In addition to the intraneuronal PrP detected by RTC, a fine granular immunostaining reminiscent of the so-called synaptic immunoreactivity was observed by immunostaining for α PrP mAb 6H4 (figure 3b and figure S1). Interestingly, while the intraneuronal PrP immunoreactivity was prominent in many regions including the spinal cord, the synaptic type was rather seen in subcortical gray matter structures (see figure S1 and figure 4 for time course and summary of pathological results). Both forms of disease-associated PrP immunodeposits are present in humans affected by E200K linked gCJD [24], further supporting the similarity between the human disease and this animal model. For technical reasons, the pathological examination of heterozygous E200K human patients [24] could not establish whether the intracellular staining was associated with E200K PrP, wtPrP or both PrP forms. However, the examination of our TG model, which allows for the comparison of Tg/ko with Tg/wt, provides a partial answer to this question. Since there was no apparent difference between the PrP immunoreactivity of both these lines with RTC and 6H4 (figure 3), we may conclude that mutant PrP thus accumulate intraneuronally in all sick Tg mice. To establish whether also wt PrP in the TgMHu2ME199K/wt mice can accumulate inside neurons or produce any form of disease related PrP, we need an antibody that will recognize only this PrP form. Regretfully, we could not allocate a reagent with exclusive specificity for wt PrP as opposed to MHu2M PrP that will also be suitable for pathological studies. Spongiform changes, a common feature of scrapie RML strain in mice [25], were observed only focally at the end-stage of disease, mostly in the frontal cortex and in the basal ganglia (figure S1). Mild degree of neuronal loss and reactive astrogliosis was observed already in 3 months old Tg mice in the basal ganglia, thalamus, and circumscribed areas of the frontal cortex, as well as in the spinal cord. These alterations became more prominent in later stages (at 8 months old and at the end point of disease) concomitantly with the clinical symptoms described above (see figure S1 and figure 4). As described above, intraneuronal PrP in the TgMHu2ME199K mice was visualized mostly with C-terminal αPrP antibodies, in particular RTC, a polyclonal antibody which detects the 201–205 PrP epitope [26], suggesting disease related PrP may accumulate in the Tg mice as an N-terminally truncated form. To further test this possibility by biochemical methods, as well as to evaluate other prion like biochemical properties of PrP in the Tg mice, we used diverse αPrP antibodies (see figure 3a for specific epitopes) to immunoblot brain homogenates from 8 months old mice from different genetic backgrounds (figure 5a). These include TgMHu2M E200K mice on both the ablated (lane 1) and wt (lane 2) PrP background, as well as from age matched wt TgMHu2M (wt chimeric human PRNP transgene mice (lane 3), which have not developed spontaneous neurological disease during their life span [15]. As additional controls, we also tested brain homogenates from normal (lane 4) and RML scrapie infected (lane 5) mice (also see insert in figure 5 for sample description). Panel 5a shows that while αPrP mAb IPC1 reacted preferentially with samples 2,4, and 5 which express PrP from a wt mouse allele, mAb 3F4 reacted only with the samples expressing TgMHu2MPrP, regardless of the presence of the E200K mutation or of the additional expression of wt PrP. Contrarily to the antibodies with species selective immunoreactivity (IPC1 and 3F4), mAb 6H4 recognized PrP forms from all brain samples at comparable levels (see also figure 1c). Last, the C-terminal RTC antibody detected equally the established PrP bands in all samples, but in addition recognized some truncated PrP forms (of about 10 and 20 kDA) in the samples comprising a TgMHu2ME199K allele. These bands (see arrows for truncated forms) were absent from samples of both wt mice and TgMHu2M controls, suggesting they are specific for PrP in the TgMHu2ME199K mice. To learn more about the PrP bands recognized by pAb RTC in the Tg mice brains, we subjected the samples presented in panel a to digestion by PNGase, an enzyme which removes N-linked sugars from proteins [27]. Figure 5b shows that while the 26 KDa band, representing deglycosylated full length PrP was detected by RTC in all samples, the TgMHu2ME199K samples presented additional and unique deglycosylated bands (see arrows), different also in their molecular weight from the 19 Kda band representing deglycosylated PK resistant PrP in scrapie brains (lane 5). To investigate whether PrP forms present in the brains of sick TgMHu2ME199K mice are aggregated and PK resistant, properties established as the hallmark of disease related PrP, we subjected Sarkosyl extracted brain homogenates from sick TgMHu2ME199K mice and controls (same lane numbering as above) to centrifugation at 100,000 g. In parallel, similar homogenate samples were digested with 30 µg/ml PK for 30 minutes at 37°C. The samples generated by these experiments were immunoblotted with α PrP antibodies 6H4 and RTC. Figure 5c shows that a significant fraction of the PrP protein present in the sick mice (lanes 1&2) pelleted under these conditions, resembling the fraction of aggregated PrP in scrapie infected mice (lane 5). This was not the case for PrP in the samples from the control chimeric or from the wt mice (lanes 3&4). As in the previous panel, immunoblotting of the same samples with pAb RTC revealed additional lower bands of about 10 to 20 Kda, in both the pellet and the supernatant. Following digestion of the homogenates with PK, and consistent with the pathological results (figure 2); it was again RTC that could detect the PrP bands resistant to protease digestion. To establish whether truncated PK resistant PrP in the brains of TgMHu2ME199K mice are a feature of the mutated PrP chimera at all ages or represent the onset of disease in older mice, we looked for their presence in the brains of young and asymptomatic TgMHu2ME199K mice. Figure 5d shows that PK resistant PrP was absent at 1 month of age (lane 1), barely present at 3 months of age (lane 2), but was clearly apparent at 8 months of age (lanes 3 and 4), when animals were severely sick. These results demonstrate that, consistent with the immunohistochemistry results described in figure 4, the appearance of PK resistant PrP forms correlate with age and disease progression, and are not an automatic feature of mutated PrP. We have recently shown that pAb RVC, a polyclonal antibody generated against reduced 203–214 human/mouse PrP peptides, could not detect Human PrPSc in brains of genetic or sporadic CJD patients [26]. This and other experiments demonstrated that Methionine residues (Met) in human PrPSc are present in an oxidized form. This was also the case for Met residues in recombinant human E200K PrP. To test the oxidation status of PrP in our sick TgMHu2ME199K mice, we subjected Sarkosyl extracted brain homogenates from wt and from TgMHu2ME199K/ko mice, as well as from mice infected with RML prions to 10–60% sucrose gradients. Subsequently, the gradient fractions were immunoblotted with both RTC and RVC α PrP antibodies. Figure 6 shows how PrP in the TgMHu2ME199K mice (both full length and truncated) was detected in all the gradients fractions when the blots were challenged with pAb RTC, indicating the mutant protein may be present in the brains of these mice at diverse aggregation states. In contrast, pAb RVC detected only full length PrP in the lighter fractions, suggesting that TgMHu2ME199K PrP may be oxidized and aggregated during its metabolic pathway in the Tg mice, as is also the case for PrPSc in the infected brains. These experiments indicate that most mutant PrP in the Tg mice is not oxidized immediately upon its generation, but becomes oxidized concomitantly with its aggregation during its metabolic pathway. Brain samples from heterozygous patients carrying the E200K PrP mutation were shown to transmit prion disease to primates [9] as well as to both wt and TgMHu2M PrP mice [10], [28]. To test if our mice also produce infectious prions in addition to fatal spontaneous disease, we inoculated the brain homogenates from an asymptomatic TgMHu2ME199K/wt mouse, from a sick TgMHu2ME199K/wt mouse, as well as from a sick TgMHu2ME199K/ko to groups of wt (C57Bl/6) mice. We speculated that the presence of a wt allele in the E199K PrP Tgs may induce the formation of some levels of wt PrPSc, thereby facilitating transmission of disease to wt mice following their infection with brains of the Tgs. To this effect, we inoculated the samples from the heterozygous mice only intraperitoneally (i. p.), which although resulting in a longer inoculation time is a less invasive pathway, while the brain homogenate from the sick TgMHu2ME199K/wt was inoculated both i. c. (intracerebrally) and i. p. , to maximize the possibility of transmission. As control for the experiment, the brain homogenate of a wt Tg MHu2M mouse was inoculated i. p. into a C57B/6 group. These mice were shown previously to remain healthy for more then 640 days [15]. In addition, a wt C57B/6mouse brain homogenate was inoculated i. c. to a group of mice in the same room as a general control for contamination. All inoculated animals were evaluated twice a week for clinical signs. Figure 7a shows a typical sick mouse infected with any of the TgMHu2ME199K brain samples, demonstrating lower body atrophy and hind limbs weakness, both properties reminiscence of the spontaneous disease of the donor Tg mice. The transmitted mice also showed “tip toe” walking, a rare feature described only in some prion related models [29]. Clinical signs present in other infectious prion strains, such as kyphosis and plastic tail, were also observed in these mice, as opposed to the donor Tgs. The transmitted disease affected the 6 mice of group 9. 9 (inoculated with brain homogenate from a sick TgMHu2ME199K/wt mouse), 3 out of 5 mice of group 9. 3 (inoculated with the brain extract from an asymptomatic TgMHu2ME199K/wt mouse), 2 out of the 5 mice inoculated i. c. and 1 out of 6 mice inoculated i. p with a brain extract of a sick TgMHu2ME199K/ko mouse. Disease signs appeared first in the mice infected i. p. with TgMHu2ME199K/wt at about 160–180 days and progressed to their death 2–3 months thereafter (see figure 7c for survival results). After infection with TgMHu2ME199K/ko, some mice became sick at 210 days (i. c) and 300 days (i. p.). While disease in these mice was apparently shorter than in the ones infected with TgMHu2ME199K/wt brains (2–3 weeks), no conclusions can be drown from this observation due to the small numbers. None of the control mice (Tg MHu2M and wt) develop any signs of disease for more than 400 days. Our results therefore indicate that, like in human E200K brains [10], [28], infectious prions are spontaneously formed in brains of TgMHu2ME199K mice. Most important, potential infectivity is generated in these mice brains before the appearance of clinical signs, as seen by the fact that brains of asymptomatic mice transmitted disease to some of the wt mice. The levels of infectious prions may further increase with disease progression, as seen by the fact that mice infected with the sample from a sick TgMHu2ME199K/wt mice (group 9. 9) succumbed to disease in a relative short time, as compared to asymptomatic sample 9. 3. Our results also suggest that the wt allele in heterozygous mice may facilitate the transmission of infectivity to naïve wt mice, since transmission from a sick TgMHu2ME199K/ko mice required a very long incubation time and occurred only occasionally, in particular after i. p. inoculation. Facilitation of disease transmission by a wt allele may result from the in-vivo generation of wt PrPSc in the TgMHu2ME199K/wt mice (even if at marginal levels), concurrently with the quantitative spontaneously generation of mutant disease related PrP. This may indicate that the “species barrier” between both forms of PrP may have been abrogated to some extent in the TgMHu2ME199K/wt mice. It also implies that while wt PrP has little or no effect on the actual presentation of spontaneous disease and its progression, low levels of wt PrPSc in these animals may be very central for the further passage of disease to naïve wt mice. Figure 7c shows an α PrP immunoblot (pAb RTC) of brain homogenates from individual mice, either infected with scrapie RML, naïve C57B/6, or infected with a brain homogenate from an asymptomatic TgMHu2ME199K/wt mouse, 9. 3 (see table 3 for the numbering of samples in figure 7c &d). As can be seen in the figure, sample 1 (derived from sick wt mouse 224 days post infection) presents a similar pattern of disease related PrP as in the RML infected sample, as opposed to, sample 2 (derived from a healthy wt mouse 413 days post infection) in which no PrPSc can be detected. Figure 7d presents an immunoblots in which individual PK digested samples from infected mice were overdeveloped to allow for the detection of low levels of PK resistant PrP. Consistent with the results in figure 7c, PrPSc was not be detected in the asymptomatic mice from group 9. 3. Contrarily, figure 7d shows that PK resistant PrP forms could be detected in the brains of a selection of brains from mice that succumbed to disease, however the levels and pattern of disease related PrP differed significantly between individual samples (see summary in Table 3). This was true even for extracts of mice infected with the same inoculum, and presenting the same symptoms, as was the case for samples 5–8, which were infected with brain homogenate of a sick TgMHu2ME199K/wt mouse, and samples 9 and 10, both infected with a TgMHu2ME199K/ko brain extract. As opposed to the donor Tg mice, no truncated PrP forms were observed with this antibody in any of the samples. The different levels of PrPSc in wt mice inoculated with the same prion homogenate are consistent with results from experiments describing the transmission of BSE into wt mice [30]. In that case the presence of PrPSc in the direct transmission from cow brains could be detected only in about 50% of the mice, while fatal disease presented in all of the animals. PrPSc became apparent in all mice following adaptation of the new strain by additional mice to mice passages. Sections of formalin-fixed, paraffin-embedded brains of mice infected with TgMHu2ME199K/wt, TgMHu2ME199K/ko and RML prions were examined for prion parameters. Figure 8a presents sections of the frontal cortex. Brains infected with TgMHu2ME199K samples present minor to moderate spongiform changes, distinctly different from the high levels of spongiform changes apparent in the RML strain [31]. The infected mice also showed severe astrogliosis and neuronal loss, as well as prominent diffuse synaptic type disease-related 6H4 PrP immunoreactivity, similar to the ones seen for the RML sections. As opposed to the spontaneous disease of TgMHu2ME199K mice (figure 3), only low levels of RTC related immunostaining were observed in the infected mice' s brains, as shown in figure 8a for the sample infected with a TgMHu2ME199K brains. RTC immunostaining was not observed in the RML samples. To test whether RTC related immunostaining can distinguish better between the RML and TgMHu2ME199K generated prions at a different experimental setup, we immunostained sections of TgMHu2ME199K and RML infected mice with RTC following a less harsh epitope revealing treatment (no formic acid after heating with citrate). Figure 8b shows that under these conditions, pAb RTC detected intracellular PrP aggregates in the mice infected with TgMHu2ME199K, but not in those infected with RML homogenates. Both brain samples presented a diffused immunoreactivity reminiscence of PrPC. No immunoreactivity of any kind was observed in brains of PrP ablated mice, indicating that the positive stain in the infected sample is indeed PrP. In conclusion, clinical, biochemical and pathological results presented in this section demonstrate that brains from TgMHu2ME199K mice may generate de-novo prions with specific properties. These prions may readily transmit to wt mice, and are particularly infectious when in the brains of sick TgMHu2ME199K on a wt background. Whether other organs of these mice, and in particular blood and immune cells, may also transmit infectivity remains to be established. Results from such experiments may be very important to assess blood safety in the community. Constructing a clinically relevant mouse model has proven to be a hard task for most neurodegenerative diseases [32]. The existing models, in particular for Alzheimer' s of Parkinson diseases present mostly pathological markers and in some cases behavioral changes, but not obvious clinical symptoms, or age dependent deterioration that correlates with those observed in human patients [33]. As opposed to models for the more common neurodegenerative conditions, several genetic prion diseases linked to pathological mutations in the PRNP gene have been reconstructed clinically in transgenic mice lines. Each of these models demonstrate several of the basic features of genetic fatal prion disease, as is the case for those mimicking GSS linked to the PrP P101L mutation [3] [21], the D177N CJD or FFI mutations [18], [19] or the insertional PrP modification [34]. These mouse models were seminal in proving that spontaneous prion disease may indeed result from the presence of a pathological PrP mutation. In this work, we describe the properties of a Tg line mimicking the most common genetic prion disease [3], i. e CJD linked to the human E200K PrP mutation. Our Tg line presents all prion relevant properties, spontaneous fatal disease, PrP pathology and transmission of prion disease to wt mice. This is particularly intriguing in view of the fact that two other models of this same mutation failed to generate disease in transgenic mice. While they may be other explanations for the different results in our case, we assume that the introduction of the E200K mutation into a chimeric mouse human PrP, as opposed to a mouse PrP [21] or a human PrP [20], is of biological importance. Chimeric PrP may constitute the bridge that allows human prion diseases to manifest in mice. Indeed, chimeric human mouse PrP was required to transmit at low incubation times genetic and sporadic human prion disease to mice [15]. Moreover, while Tgs expressing the GSS 102 mutation in human PrP did not present spontaneous disease, the same mutation in chimeric PrP did present neurodegenerative disease [28]. Whether the structure of chimeric PrP is more favorable for disease transmission or otherwise the chimeric form has the ability to bind a mouse component important for transmission of human prion diseases to mouse models remains to be established. Another novel feature of our Tg line is the generation of de-novo infectious prions that could be transmitted to naïve wt mice. Indeed, E200K CJD is the genetic disease most similar to the sporadic forms, in both clinical appearance, age of onset and pathology [2], [10]. This may imply that E200K de-novo prions are more similar in structure to sporadic ones, which are highly transmissible [9]. Whether the chimeric background of E200K PrP in these mice is also a factor in the transmissibility of disease is unknown at this point, however it is important to state that chimeric mouse human PrP Tg mice are susceptible for infection with both mouse and human prions [28]. While the neuropathology features of our Tg mice were similar to E200K human patients with regard to reduced spongiosis and intracellular PrP accumulation [24], PK resistant PrP in the TgMHu2ME199K mice was detected mostly by the C-terminal pAb RTC, suggesting a considerable fraction of disease related PrP in the brains of these mice accumulates as a truncated form. Indeed, diverse truncated PrP forms were also described in brains of CJD patients [35], including those carrying the E200K mutation [36]. Interestingly, intraneuronal immunoreactivity in these CJD patients predominates in the brainstem and may be associated with alterations in the accumulation of other neurodegeneration-related proteins (e. g. phospho-tau, alpha-synuclein) [24]. Evaluation of concomitant protein pathology in our model is the objective of another ongoing study. Because gCJD is a dominant genetic disorder, we investigated the properties of the TgMHu2ME199K mice not only on a PrP ablated but also on a wt PrP background. We first speculated that the presence of wt PrP may preclude some disease symptoms related to the absence of the elusive PrPC activity or to the putative toxicity of truncated PrP forms, as was shown previously in other systems [29]. However, this is probably not the case, as can be inferred from the fact that no significant differences were seen between both lines of Tg mice in clinical symptoms, kinetics and pathological examination. Finally, while the investigation of a small group of homozygous E200K CJD patients showed a moderate decrease in the age of disease onset for most patients, it also described a patient with a very slow progressive disease (96 months), who died in the absence of PrPSc accumulation [37]. Whether disease onset in one or both lines of TgMHu2ME199K mice may be modulated by oxidative stress or other pathogenic insults is under investigation in our laboratory. In summary, we believe that our TgMHu2ME199K lines will play a central role both in the elucidation of genetic prion disease pathogenic mechanism as well as in the search for anti-prion compounds. The early presence of spontaneous disease followed by their sequential age related deterioration during several months until death will permit to study the long term effect of reagents that may delay disease onset in at risk subjects. Among those to be tested first are substances suggested to have a marginal but still encouraging result in already sick CJD patients, such as doxicyline [38] and flupirtine [7], as well as those believed to present significant therapeutic results in scrapie infected mice or infected cells, such as Quinacrine and Simvastatin [39], [40]. Novel approaches such as passive [41] or active immunization, as well as RNAi inhibition of mutant PrP expression [42], will also be tested in the near future. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Hebrew University Medical school (Permit Number: MD-11746-5). All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering. Transgenic mice harboring the E200K mutation were generated by one of us (ZM) in the Prusiner laboratory as follows. Using site-directed mutagenesis by PCR, the E200K mutation was inserted into the chimeric human/mouse PrP open reading frame (ORF) of an MHu2M construct that was previously prepared as described [15]. The 0. 8 kb SalI-XhoI fragment containing the PrP ORF with the E200K mutation was inserted into cos. ShaTet and further injected into transgenic mice ablated for the PrP gene [43]. Creating and screening of the transgenic mice were done as described [44]. Several lines were produces at the time and at least 2 presented spontaneous prion disease (ZM, personal communication). For the present project, C57BL/6 female mice were impregnated with frozen sperm from one of these lines in the Jackson laboratories. The designated offspring were subsequently crossed either with wt or with PrP ablated C57BL/6 mice (Harlan laboratories, obtained by crossing of PrP0/0 FVB mice [43] with the C57BL/6 strain for 10 generations, and screened for the presence of the TgMHu2ME199K PrP, ablated or wt PrP allele, as required. The mice used in this project are of a mixed C57B/6/FVB background, ranging from 75- to 95% C57B/6. Total RNA from mice brains was isolated using TRI reagent (Sigma, Israel). cDNA was prepared from 2 µg of total RNA using MuLV reverse transcriptase and random hexamers (Promega) according to the manufacturer' s instructions. Quantitative RT-PCR was carried out in 15 µl reactions containing 1 µl of cDNA, 0. 3 µM of the appropriate primers (sigma, Israel), and 7. 5 µl of the SYBR Green master mix (Finnzymes). Gene amplification was carried out using the GeneAmp 7500 Sequence Detection System (Applied Biosystems). Measurements were performed in triplicates and UBC (Ubiquitin C) and TBP (TATA-box binding protein) transcript levels were used to normalize between samples. The primers used were PrP, 5′-CAA GCA GCA CAC GGT CAC C-3′ (forward), 5′-GGC CTG GGA CTC CTT CTG G-3′ (reverse) TBP, 5′-TGT GCA CAG GAG CCA AGA-3′ (forward), 5′-CCC CAC CAT GTT CTG GAT-3′ (reverse); UBC, 5′-CAG CCG TAT ATC TTC CCA GAC T-3′ (forward); 5′-CTC AGA GGG ATG CCA GTA ATC TA-3′ (reverse). The primers used for PrP were chosen so that they can be used for both wt (mouse) PrP as well as for MHu2M PrP. Mutant TgMHu2ME199K mice from both lines (PrP ablated or wt background) were followed twice a week for the appearance of spontaneous neurological disease. Mice were scored for disease severity and progression according to the scale of clinical signs described in Table 2. This scale was designed by us to fit the clinical symptoms observed in the Tg mice and was proven to be parallel to the NNS (neurological severity score). Mice were sacrificed according to the ethical requirements of the Hebrew University Animal authorities (when too sick or paralyzed to reach food and water, or after loosing 20% body weight). As described in table 2 mice were scored for disease severity and progression according to a scale of clinical signs designed by us to fit the clinical symptoms observed in the Tg mice. Hind limbs weakness was first evaluated by closely watching the mouse walking on a flat surface looking for sings of abnormal limb posture or abnormal walking pattern (high or low gait, leg dragging). Next, mice were tested for their ability to walk on a 3 cm beam in a straight line and maintain balance. Finally mice were lifted by their tail to check for leg clasping. Full paralysis was evaluated by total lack of movement in the limb. This scale of scoring was proven to be parallel to the NSS (neurological severity score) [45]. Blindness was tested by the lack of reaction of the mice to a paper slowly placed before its eyes. 10% brain homogenates of asymptomatic or sick TgMHu2ME199K/wt, sick TgMHu2ME199K/ko, as well as from control TgMHu2M and naïve wt mice were each inoculated i. p. or i. c, as designated in the text, into a group of 6 C57BL/6 mice (Harlan laboratories). The inoculated mice were scored twice a week for clinical signs of prion disease until the beginning of symptoms and more closely thereafter. Following termination of each experiment, mice were sacrificed and analyzed for pathology and for the presence of disease related PrP. Four µm thick sections of formalin fixed, paraffin embedded brains of TgMHu2ME199K mice as well as of C57BL/6 mice infected with TgMHu2ME199K brains, in addition to controls and PrP ablated mice were evaluated for the presence of disease related PrP, gliosis and spongiform changes as previously described [24]. A less harsh epitope retrieval method, with the avoidance of formic acid, was also applied in some cases. Brains from TgMHu2ME199K mice on a wt or ablated background, normal mice, control TgMHu2M and scrapie RML infected mice were homogenate at 10% (W/V) in 10 mM Tris-HCl, pH 7. 4 and 0. 3 M sucrose. For Proteinase K digestions, 30 µl of 10% brain homogenates extracted with 2% sarkosyl were incubated with 30 µg/ml Proteinase K for 30 min at 37°C. Samples were subsequently subjected to SDS PAGE and immunoblotted with the diverse α-PrP antibodies, as described in Figure 2 a. Protein precipitation experiments, as the ones observed in Figure 5 c, were performed by ultracentrifugation of Sarkosyl extracted homogenates at 100000 g, and subsequently separating pellets from supernatant. Deglycosylation by PNGase was performed as previously described [46]. Normal and prion infected Sarkosyl extracted brain homogenates were subjected to sucrose gradients as described [47]. Shortly, 140 µl of 10% brain homogenates extracted in the presence of 2% Sarkosyl were overlaid on a sucrose gradient composed of layers of increasing concentrations of sucrose (10–60%). Gradients were then centrifuged for 1 h at 55000 rpm in a Sorval mini-ultracentrifuge and subsequently 11 samples of 120 µl were collected from the top to the bottom. Gradient fractions were then immunoblotted with either α PrP pAb RTC or RVC.

Inherited prion diseases, such as genetic CJD, are dominant disorders linked to mutations in the gene encoding the prion protein, PrP. Since therapeutic intervention in all types of human prion diseases has failed, we propose that therapeutic efforts should be directed mostly to the development of preventive treatments for subjects incubating prion diseases, as is the case for asymptomatic carriers of pathogenic PrP mutations. These subjects will develop disease symptoms at some point in their adult life; therefore they should be treated before clinical deterioration. Candidate treatments will need to be tested for efficacy and safety first in animal models that mimic most properties of genetic CJD. In this work, we describe a new transgenic mouse model for E200K genetic CJD, presenting progressive neurodegenerative disease and age related prion disease pathology and biochemistry, as is the case in the human disease. Brain extracts from these mice also transmitted prion disease to wt mice, as shown before for parallel human samples. We propose that these animals will play a significant role in the development of novel anti-prion prophylactic treatments.

Abstract Introduction Results Discussion Materials and Methods

medicine biology

2011

Fatal Prion Disease in a Mouse Model of Genetic E200K Creutzfeldt-Jakob Disease

Peptide-protein interactions contribute a significant fraction of the protein-protein interactome. Accurate modeling of these interactions is challenging due to the vast conformational space associated with interactions of highly flexible peptides with large receptor surfaces. To address this challenge we developed a fragment based high-resolution peptide-protein docking protocol. By streamlining the Rosetta fragment picker for accurate peptide fragment ensemble generation, the PIPER docking algorithm for exhaustive fragment-receptor rigid-body docking and Rosetta FlexPepDock for flexible full-atom refinement of PIPER docked models, we successfully addressed the challenge of accurate and efficient global peptide-protein docking at high-resolution with remarkable accuracy, as validated on a small but representative set of peptide-protein complex structures well resolved by X-ray crystallography. Our approach opens up the way to high-resolution modeling of many more peptide-protein interactions and to the detailed study of peptide-protein association in general. PIPER-FlexPepDock is freely available to the academic community as a server at http: //piperfpd. furmanlab. cs. huji. ac. il. Proteins are the workhorses inside living cells, and interactions among them are critical for various important biological processes [1]. A significant fraction of these interactions (15–40%) [2] are peptide mediated, where a short stretch of residues from one partner contributes most to its binding to the other. Such short peptidic regions, also termed short linear interacting motifs (SLIMs) are often found embedded inside disordered regions of intrinsically disordered proteins (IDPs) [2,3], or appear as flexible linkers connecting domains [4] and as flexible loops tethered to rigid segments [5]. The development of accurate structure based modeling tools is critical for atomic level understanding of peptide-protein interactions, to allow the manipulation of known interactions, to discover yet unknown peptide-protein interactions and networks, and to provide starting points for the design of novel peptides and related molecules to target specific systems of pharmacological interest [6]. A number of computational tools have been developed to assist the characterization of peptide-protein interactions, including the prediction of peptide binding sites [7–9], refinement of coarse peptide-protein models [10], folding and docking on a known binding site [11] and most challenging of all, global peptide-protein docking with no prior information about the peptide structure and the binding site [12–17]. The challenges associated with the global docking of flexible peptides have been addressed in different ways, by reducing the conformational space to be sampled both for the internal degrees of freedom of the peptide as well as its rigid-body orientations on the receptor surface. For peptide docking within the HADDOCK docking framework [12], the peptide backbone is represented by idealized conformation (s), such as alpha helix, beta strand and polyproline-II, followed by rigid-body, semi-flexible and fully-flexible docking with explicit solvation [18]. The pepATTRACT protocol [13,19] uses the same approach to represent the peptide, followed by coarse-grained rigid-body docking and flexible full-atom refinement. The AnchorDock protocol uses molecular dynamics simulations to generate a set of plausible peptide conformations, which are then docked using anchor-driven simulated annealing molecular dynamics around predicted anchoring spots on the receptor [14]. The CABS-dock protocol uses randomly generated peptide conformations based on either predicted or known secondary structure, randomly orients these peptides over the receptor surface, and refines them using replica exchange Monte Carlo dynamics [15]. The MDockPep protocol [16] uses peptide sequence similarity to extract fragments from high resolution protein structures, which are further refined using MODELLER [20] to generate plausible peptide conformations, and then docked onto the receptor using rigid-body docking and flexible docking with AutoDock Vina [21]. The recently published IDP-LZerD protocol models the binding of long disordered segments to structured proteins using the Rosetta fragment picker protocol [22] to generate fragments of 9-residue overlapping windows followed by LZerD [23] rigid-body docking and molecular dynamics refinement [17]. Finally, we have recently advanced a novel, global motif-based peptide fragment docking approach, PeptiDock [24], in which peptide binding motif information rather than secondary structure propensity is used to extract fragments from the Protein Data Bank (PDB [25]), which are then docked to the receptor using PIPER rigid body docking [26], followed by minimization using CHARMM [27]. Notwithstanding these significant recent advances in global peptide docking, present approaches are still limited in their modeling quality and general applicability, and there is ample room for improvements that would enable the detailed high-resolution study of more peptide-protein interactions with higher accuracy. Here we describe PIPER-FlexPepDock, a successful effort toward the development of such a robust, highly accurate, global peptide-protein docking protocol. By integrating accurate peptide fragment ensemble generation using the Rosetta fragment picker [22], fast and exhaustive fragment-receptor rigid-body docking using PIPER docking [28], and flexible full-atom refinement of coarse PIPER models using Rosetta FlexPepDock [10], we were able to sample both the peptide backbone conformational states, as well as the landscape of the peptide-receptor interactions efficiently and with much higher accuracy than current protocols: on a representative non-redundant dataset of peptide-protein complexes well resolved by X-ray crystallography (Table 1 below), PIPER-FlexPepDock generates for about half models within 2. 5 Å ligand RMSD (2. 0 Å, when restricted to motif regions where available), more than twice as many as for existing peptide docking protocols such as pepATTRACT [13] (among the 10 top-ranked predictions; Table 2 below). Our results highlight the relevance of representing the peptide as a set of fragments that can be exhaustively docked as rigid bodies onto the receptor structure and subsequently refined using an accurate refinement protocol. They reinforce the underlying biophysical model of a conformer ensemble of the free peptide that already samples the bound conformation (at least in the encounter-complex, protein-like environment) and involves only limited induced fit, not unlike the classical association between preformed protein domains. As a result, PIPER-FlexPepDock brings into reach the study and targeted manipulation of a range of additional peptide-mediated interactions not accessible before due to limitations in sampling and/or accuracy. Motivated by our recent advance in global peptide docking using a motif-focused approach [24] we ventured into the development of a more generalized protocol. We initially calibrated our docking approach on a small but representative set of nine peptide–protein complexes (highlighted in bold in Table 1; see also S1A Table). We trimmed the peptide based on the motif defined in ELM, where available. For all complexes high modeling accuracy was achieved for this new global docking approach (within ≤2. 5Å Ligand RMSD models among the top 10 ranking clusters; Table 1). For the full length peptides modeling near-native models were obtained for 5/9 cases, highlighting the benefits for motif (or shorter peptide sequence) focused modeling, due to better fragment quality compared to the corresponding full-length peptides (Table 1). Encouraged by these initial results, we proceeded to the validation of our protocol on a larger representative set of peptide-protein complexes (Table 1 and S1B Table). We assessed the performance of PIPER-FlexPepDock on a larger, non-redundant set consisting of 27 complexes (compiled from the 42 complexes used in previous studies, but non-redundant at the domain level, as defined by CATH [36]; see Methods), among them 12 with reported binding motif. The benchmark is summarized in Table 1 (S1C Table provides results for the redundant set of 42 complexes used in previous studies, as well as additional details, including performance of other approaches for comparison). We compared the results of PIPER-FlexPepDock (unbound-min run) with other existing global peptide-protein docking protocols such as HADDOCK [12], pepATTRACT [13], CABS-dock [15], and MDockPep [16] on our non-redundant set of 27 complexes, as well as on the set of 42 complexes used by these protocols in previous studies [34 complexes were compared with HADDOCK as other 8 cases were not included in their unbound run set). Since full length peptides were modeled using the other protocols, we modeled full length peptides for the motif set cases for valid comparison. The success rate for generating near-native models (i. e. , L-RMSD within 2. 0Å, or 3. 0Å) was significantly better for PIPER-FlexPepDock than any other protocol, even for models of the full peptides (see Fig 4B and Table 2). In order to maximize the impact of our new protocol for global peptide-protein docking and to make it accessible to the modeling of many new peptide-protein complexes, we have set up a user-friendly server open to the scientific community (Fig 5). All that is needed is a structure of the receptor and a sequence of the peptide, but additional information about peptide secondary structure can also be included to narrow the search. The top-ranking resulting models can be downloaded, or inspected by an interactive viewer using the 3Dmol. js libraries [45]. With the presentation of our new PIPER-FlexPepDock algorithm, we have demonstrated that combining fast and exhaustive rigid-body docking (using the FFT-based PIPER docking algorithm) of a representative peptide conformer ensemble (approximated by fragments extracted from solved structures, based on local similarity of sequence and secondary structure), with high-resolution refinement (using Rosetta FlexPepDock) is a successful approach for the generation of models of peptide-receptor structures of remarkable accuracy–significantly better than any other current protocol—starting from the sequence of the peptide and the structure of the receptor. The performance on a representative benchmark of solved peptide-protein complex structures demonstrates both accuracy and robustness of our modeling approach, and opens up the way of modeling many more peptide-protein interactions at much higher resolution and accuracy than any existing global peptide-protein global docking protocol. This study demonstrates that fragments derived from solved protein structures, based on secondary structure and sequence similarity (rather than on sequence binding motifs which are not always available) represent the peptide conformational states with high accuracy, in particular the bound state. Interestingly, it is this same observation regarding the representation of local conformational preference that provided originally the platform for the breakthrough of Rosetta ab initio protein structure prediction [46]. This indicates that while isolated peptides in solution rarely show significant conformational preferences [47], in the encounter complex regime in vicinity of other proteins, their conformational freedom seems to be restricted significantly (similar to local peptide regions within a full protein) and can be represented by fragment libraries, in concordance with previous reports that show similar arrangements of fragments within monomers and peptide-protein interactions [48]. The simplified scoring function and exhaustive sampling with PIPER allows uniform sampling of the fragments onto the receptor on a smoothened energy landscape. The top scoring PIPER models represent the dense sampling into wider energy basins. Though the ranking of models might lack the accuracy at this stage, the following refinement stage performs local sampling to efficiently locate the minimum. Interestingly, this approach is much more effective than the local refinement starting from one representative model: only one FlexPepDock optimization run is necessary starting from each PIPER model, compared to several hundred to thousand runs starting from a representative (defined, e. g. from a PIPER run as implemented in the PeptiDock peptide motif docking algorithm [24]). This is most probably due to the fact that these starting coarse models are trapped in many distinct states, each near a distinct local minimum, simplifying sampling during optimization. The peptide-receptor binding energy landscape can provide a broader understanding of the binding mechanism itself. The exhaustive sampling with accurate refinement provides a high-resolution map of the energy landscape and helps us understand the energetic of the encounter between the peptide and the receptor. In a previous study, we were able to demonstrate that experimentally observed encounter complexes are well reproduced from a global protein docking energy landscape [49], and we anticipate that the corresponding peptide-protein docking energy landscape will provide similar information. The approach described in this study improves significantly both accuracy as well as scope of peptide docking, at least as suggested by its performance on the widely accepted PeptiDB peptide docking benchmark [12,13,15,16]. At the same time, it also highlights the bottlenecks to be overcome for its broader generalization: (1) Accurate modeling of peptide conformational ensemble: Even though the fragments generated using the Rosetta fragment picker protocol sample in general the bound peptide conformation well, challenges remain in the modeling of longer peptides, as well of as peptides with unusual conformations (Fig 2A). This is attributed to the lack of a large pool of longer representative fragments with similar sequences in solved structures. The rigid body PIPER docking step does not include any flexibility, and therefore accurate fragments are very important for efficient further refinement by FlexPepDock to near-native model quality. This challenge could be overcome by incorporating a peptide-folding algorithm as first step for fragment generation, assuming that bound-like conformations would indeed be sampled. (2) Modeling significant receptor backbone flexibility: While for many peptide-protein interactions the receptor is already pre-organized and the binding of the peptide does not induce considerable movement [50], binding may involve significant structural rearrangement of the receptor (e. g. , in the binding of Slam tail peptide to the SH2 domain of the XLP protein SAP, PDB id 1D4T[51]). To model such challenging cases, improved modeling of receptor flexibility is mandatory (using e. g. backrub moves [52] and other advanced comparative modeling approaches [53]). (3) Improved ranking of alternative models: Inspection of failures highlights that despite low quality, many of the failed simulations model the peptide into the correct binding pocket, and identify the binding hotspot regions, similar to our observation in CAPRI community-wide performance [54]. However, the details are not correct, often pointing the wrong peptide residue side chain into a given binding pocket. Such ranking problems might be removed with the advance towards better scoring functions. (4) Extension to flexible interactions: Last but not least, this approach might be restricted to peptide-mediated interactions in which the bound peptide adopts one, defined conformation, since it has been calibrated on well-resolved crystal structures of peptide-protein complexes. Many biologically relevant interactions remain more flexible, and are therefore studied using e. g. NMR experiments. The next challenge will be to extend this approach to the study of such interactions. To summarize, the novel global peptide-docking pipeline presented here allows modeling of peptide-protein interactions with much improved accuracy and scope. With further improvements for modeling of increased receptor flexibility and peptide conformational ensemble generation as described above, we should be able to accurately model any interaction that adapts a stable conformation that can be crystallized, as well as explore common features of interactions beyond. Docking performance and analysis was calibrated and assessed on a benchmark of peptide-protein complexes derived from the PeptiDB database [50], filtered according to the following criteria: The dataset was further divided into two subsets, based on available information about a peptide binding motif (defined in this study based on ELM [29], http: //elm. eu. org): For the motif set (12 complexes) we modeled only the motif part, since it contributes most to binding, and shorter peptides are easier to model. To enable comparison to performance of other protocols, we subsequently also docked the full peptide. For the non-motif set (15 complexes), the full peptide was docked. Initial calibration set: For initial calibration, we selected a smaller subset of 9 complexes (S1A Table). The established protocol was then validated on the remaining complexes, to ensure similar performance and thereby prevent overfitting of the modeling protocol. In the following we provide specific details of the different steps of the PIPER-FlexPepDock protocol. For runline commands, see the Supplementary S1 Text. For each global docking run the 10 top ranking clusters were selected as prediction and evaluated for quality based on ligand RMSD (L-RMSD), calculated between the native and model peptide backbone atoms after optimal superimposition of the receptor, as done in the CAPRI assessment [34,35]. L-RMSD and other measures, such as Fnat and I-RMSD, were calculated using DockQ [62]. The protocol and tests described in this manuscript follow the FlexPepDock protocol, as implemented within the Rosetta weekly release version 2016. 20. 58704. The processing time for the different stages of the protocol depends on both the length of receptor and the peptide sequence. For example the global docking the carboxy-terminal tail of the ErbB2 Receptor GLDVPV onto the free ERBIN PDZ domain (103 residues) the generation of 50 fragments takes ~8 CPU minutes over an AMD Sun cluster with 300 cores. For the same complex a single PIPER fragment docking simulation takes ~2 minutes and a single refinement run of the PIPER docked model takes ~1 minutes on the same system architecture (~ 1. 5 hours to refine all models). The runline commands are provided in the Supplementary S1 Text. The Rosetta software is available for free to the academic community. The details regarding downloading and installation is available at https: //www. rosettacommons. org. PIPER FFT rigid body docking is available as part of the protein-protein docking server ClusPro (PeptiDock at https: //peptidock. cluspro. org).

Peptide-protein interactions are crucial components of various important biological processes in living cells. High-resolution structural information of such interactions provides insight about the underlying biophysical principles governing the interactions, and a starting point for their targeted manipulations. Accurate docking algorithms can help fill the gap between the vast number of these interactions and the small number of experimentally solved structures. However, the accuracies of the existing protocols have been limited, in particular for ab initio docking when no information about the peptide beyond its sequence is available. Here we introduce PIPER-FlexPepDock, a fragment-based global docking protocol for high-resolution modeling of peptide-protein interactions. Integration of accurate and efficient representation of the peptide using fragment ensembles, their fast and exhaustive rigid-body docking, and their subsequent accurate flexible refinement, enables peptide-protein docking of remarkable accuracy. The validation on a representative benchmark set of crystallographically solved high-resolution peptide-protein complexes demonstrates significantly improved performance over all existing docking protocols. This opens up the way to the modeling of many more peptide-protein interactions, and to a more detailed study of peptide-protein association in general.

Abstract Introduction Results Discussion Materials and methods

cell physiology protein interactions applied mathematics simulation and modeling algorithms receptor physiology mathematics protein structure prediction protein structure sequence motif analysis protein structure databases research and analysis methods sequence analysis bioinformatics proteins biological databases molecular biology biochemistry sequence databases cell biology database and informatics methods biology and life sciences physical sciences macromolecular structure analysis

2017

High-resolution global peptide-protein docking using fragments-based PIPER-FlexPepDock

To characterize intracellular energy transfer in the heart, two organ-level methods have frequently been employed: inversion and saturation transfer, and dynamic labeling. Creatine kinase (CK) fluxes obtained by following oxygen labeling have been considerably smaller than the fluxes determined by saturation transfer. It has been proposed that dynamic labeling determines net flux through CK shuttle, whereas saturation transfer measures total unidirectional flux. However, to our knowledge, no sensitivity analysis of flux determination by oxygen labeling has been performed, limiting our ability to compare flux distributions predicted by different methods. Here we analyze oxygen labeling in a physiological heart phosphotransfer network with active CK and adenylate kinase (AdK) shuttles and establish which fluxes determine the labeling state. A mathematical model consisting of a system of ordinary differential equations was composed describing enrichment in each phosphoryl group and inorganic phosphate. By varying flux distributions in the model and calculating the labeling, we analyzed labeling sensitivity to different fluxes in the heart. We observed that the labeling state is predominantly sensitive to total unidirectional CK and AdK fluxes and not to net fluxes. We conclude that measuring dynamic incorporation of into the high-energy phosphotransfer network in heart does not permit unambiguous determination of energetic fluxes with a higher magnitude than the ATP synthase rate when the bidirectionality of fluxes is taken into account. Our analysis suggests that the flux distributions obtained using dynamic labeling, after removing the net flux assumption, are comparable with those from inversion and saturation transfer. In heart, the mechanisms that ensure energy production meets demand over a wide range of workloads, remains unclear. Fundamental to this search is an accurate understanding of the recycling fluxes of ATP, ADP, Pi, and phosphocreatine (PCr) between the mitochondrial inner membrane space and the ATPases on both the myofibrils and sarcoplasmic reticulum. In highly compartmentalized environments, such as heart muscle [1]–[8], determination of energy transfer fluxes is far from trivial. The organ level methods used to determine the fluxes include -NMR inversion and saturation transfer, and the transient labeling of, , , and Pi using, a method we refer to as dynamic labeling. To estimate the fluxes using -NMR inversion and saturation transfer, magnetization transfer has been simulated in the compartmentalized system and fitted against experimental data [9]–[12]. A recent study [12] employed rigorous statistical testing of model solutions against experimental data using stochastic models that described measurement uncertainty. Such analysis was applied to discriminate between energy transfer pathways over a range of cardiac performance levels. These results suggest that at least 40% of the energy is exported via direct ATP transfer at high cardiac performance. This method is unable to determine the split between direct ATP transfer and the creatine kinase (CK) shuttle at lower cardiac performance. While the split has not been determined for all conditions, total CK unidirectional flux was found to be stable if energy demand was changed either by variation of extracellular calcium or by left ventricle balloon volume in isovolumetric contractions. There are several complications that have to be considered when interpreting magnetization transfer experiments. As summarized recently [13], [14], magnetization transfer interpretation should be based on a complete model of the biochemical reactions that could contribute to the measured transfer. For example, oversimplified interpretation of saturation transfer experiments to determine the ATP synthesis flux by analyzing the rate Pi and ATP can lead to overestimated values for ATP synthesis [13]. A recent study overcame this limitation by analyzing -NMR inversion and saturation transfer results using compartmental models and statistical methods on multiple experiments [12]. Dynamic labeling has also been employed to measure the fluxes of high-energy metabolites [15]–[28], although no study to date has measured the cardiac performance dependence of these fluxes [29]. The dynamic incorporation of into, , , and Pi was quantified using mass spectrometry in earlier papers, and more recently using -assisted -NMR [20]. From the labeling data in [28], a linear relationship was found between rate pressure product after ischemia-reperfusion recovery of individual hearts and energy transfer through CK. This relationship was identified due to variation of heart recovery after being exposed to ischemia-reperfusion with and without preconditioning. In control conditions, before exposure to ischemia, the flux through CK was estimated to be which converts to (see Methods for conversion factors). This is considerably smaller than the total CK unidirectional flux in [12] and other -NMR saturation transfer studies [30]–[38]. However, in contrast to -NMR saturation transfer studies, it is proposed that flux estimation on the basis of labeling leads to the estimation of net flux through CK shuttle system [20], [24]. Definitions for net and total flux are provided in Figure 1. Usually, labeling by isotopic tracers depends on the bidirectionality of reactions which allows transfer of the label in the opposite direction to net flux in the system [39]. Considering that the adenylate kinase (AdK) and CK reactions are reversible, it is expected that labeling could be influenced by the bidirectionality of the reactions, as well as the net flux through the shuttles. The incorporation of into the phosphoryl groups of ATP and Pi has been used in many studies including analysis of the ATP synthase mechanism [40] and cyclic nucleotide metabolism [41]. To study cyclic nucleotide metabolism, the rate of incorporation into the groups of guanine nucleotides was measured to calculate the change in cyclic GMP and cyclic AMP flux [41]. Another study by the same group used the rate of incorporation into the groups of ADP and ATP to calculate cyclic AMP fluxes in human platelets [42]. A number of years later the same group developed a technique that is able to determine the rate of ATP hydrolysis (and synthesis) by exchanging with and tracking the time course of incorporation into [43]. In contrast to groups, groups are exchanged much more often, and this technique was deemed to be unreliable if Pi and ATP also participate in reactions that have a higher magnitude of flux than the synthesis rate [43], such as the CK reaction in isolated rat heart [31]. The labeling technique was then applied in a long series of papers that explore the fluxes in the phosphotransfer networks of both heart and skeletal muscle [15]–[28], [44], [45]. While it was stated in [20], [22], [24], [25], [46] that labeling was analyzed using models, the details of these models were not presented. Despite this, it was further stated that the method determines net flux through the individual phosphotransfer pathways. For determination of net flux through the CK shuttle, a relationship between PCr labeling and CK flux was established on the basis of gradual inhibition of CK [16], [27]. However, to our knowledge, no sensitivity analysis of this method of flux determination has been performed which limits our ability to compare flux distributions predicted by labeling, and -NMR inversion and saturation transfer. In addition, we are unaware of any comprehensive kinetic analysis of labeling on a phosphotransfer network in its entirety since the work by Dawis et al. [43]. The aim of this work is to analyze the properties of -provided labeling in a physiologic heart phosphotransfer network with active CK and AdK shuttles and establish which fluxes determine the labeling state. For this, we use an integrated kinetic model able to predict the dynamic labeling state to assess if the labeling state is sensitive to net fluxes when enzymatic and transport fluxes are not assumed to be unidirectional. Prior to conducting the sensitivity analysis a number of model validation steps were performed. First, we tested if the steady state labeling distribution predicted by the model matches the theoretical distribution provided by Dawis et al. [43]. Following this we compared the dynamic predictions of our model to the dynamic predictions in [43]. Figure 2 shows that both the steady state and dynamic labeling state predictions provided by our model are indistinguishable from those provided by the model developed by Dawis et al. Additionally, it is important to ensure that the model we constructed is able to provide predictions that adequately match published dynamic labeling data measured in heart. In most studies, the fluxes derived from organ-level labeling are published without the corresponding labeling dynamics of individual species. However, a recent paper does provide dynamic measurements of the four labeled species of (, , ,) and the three labeled species of (, ,) that result after surgically removing hearts from anesthetized male rats, and immediatly immersing harvested atrial tissue in Krebs-Henseleit solution enriched with 30% [28]. As a test we used the first flux distribution from Table 1 and simulated the labeling state. This flux distribution is based on a flux distribution found using organ-level rat heart -NMR inversion and saturation transfer results from [12] together with an estimation of the AdK flux from activity measurements [47]. We found that without any fitting, the model prediction could explain the atrial tissue data in [28] (see Figure 3). Note that such close similarity between the measurements and the model solution was obtained using a flux distribution that corresponds to significantly different conditions (isovolumetrically beating heart vs isolated atrial tissue). It is clear from Figure 3 that the model does not exactly match the 120 second data point for. This species of Pi is the entry point of into the model and the overshoot in predicted labeling is analogous to that seen in Figure 2B and discussed in [43]. The dynamic overshoot in is less dramatic than in Figure 2B due to the differences in the flux distributions used and inclusion of CK reactions. The absence of the overshoot in the measurements can be due to the gradual labeling dynamics of water (the simulation uses a perfect step change), compartmentation of Pi, and the influence of reactions not considered in the model. The available data does not contain enough information to warrant the changes required for our model to fit this one datapoint. This is especially true because no estimation of measurement error was provided in [28], so it is not possible for us to quantify the goodness of fit. However, while comparing the model solution to this measured data we found that many possible flux distributions provide labeling predictions that also adequately explain the measured data. Following model validation, we performed a sensitivity scan of the model parameters. The most sensitive parameters were found to be the ATP synthase rate as well as the net and exchange fluxes through the AdK shuttle. Figure 4 provides upper and lower bounds for the dynamic labeling state predictions when the cardiac performance and net fluxes in the system are kept constant. Note that the upper and lower bounds are formed from multiple simulation results. We varied six enzymatic exchange fluxes (AdK (2 fluxes, one in cytoplasm and one in mitochondria), CK (2), ATP synthase, ATPase), 12 transport exchange fluxes (Pi (2), water (3), ADP (4), ATP (2), PCr (1) ), and 18 pool parameters (ADP (5), ATP (5), Pi (4), PCr (2), water (2) ). The six enzymatic exchange fluxes were varied over five evenly spaced points each for a total of 15625 combinations. Small and large pool size sets were constructed based on their mean values and standard deviations reported in [12]. Two transport exchange flux sets with high and low values were used. This gives a total of combinations. Additional combinations of pool parameters could increase the upper and lower ranges of sections of these curves, however, the change is not expected to be pronounced because the total pool size of the metabolites in this model are well characterized [12]. Table S2 provides upper and lower ranges for all 36 parameters that were varied. The labeling state of is most sensitive to changes in metabolic pool sizes and exchange fluxes. It was found that AdK exchange fluxes influence the labeling state, and not the labeling state of other species. Figure 5 illustrates how the transient labeling predictions change when the ATP synthesis rate or the total AdK flux is changed. Referring to Figure 5, the ATP synthesis rate is seen to influence the labeling state of all species, however, the total AdK flux mainly influences the labeling state of. Note that the combined sensitivity of all exchange fluxes and pool sizes is roughly the same as the influence of the ATP synthesis rate. The main application of dynamic labeling analysis is for the determination of intracellular flux distributions. In general, to be able to determine the flux distributions, the method must be sensitive to variations in the fluxes. In the case of dynamic labeling analysis, the labeling of species has to be sensitive to the changes in the fluxes of reactions considered. Having determined which model parameters influence labeling state predictions most, we tested if the model is able to distinguish between various flux distributions that are physiologically possible within the context of the model we present. We restrict the physiological sensitivity analysis to a heart performance where predictions of the energetic fluxes in rat heart are available. To simplify the initial phase of the physiological sensitivity analysis, the bidirectionality of transport reactions as well as pool size parameters were held constant. The combined influence of these parameters, together with the physiological parameters, provide similar predictions of the labeling state (see Figure 4), and thus can be treated sequentially in a more general sensitivity analysis. In total, nine flux distributions were studied. These flux distributions were selected to test the labeling sensitivity to the changes of specific fluxes, as explained below. General descriptions of these are provided in Table 1; Table S1 provides a detailed summary of the fluxes. Flux distributions 1 through 6 have roughly equal contributions to the net export of energy via direct ATP transfer and the CK shuttle. Flux distributions 6 and 7 have zero net flux through the AdK shuttle. Flux distributions 7 and 8 have 100% and 95%, respectively, of the energy exported via the CK shuttle, while flux distribution 9 exports 95% via direct ATP transfer. There are several specific questions that are of interest when studying the flux distributions. First, what is the magnitude of the fluxes? In terms of dynamic labeling analysis: How does the labeling state change when the activities of CK and AdK are increased? Comparing the simulated labeling states of flux distributions 1,2, and 4 could provide insight into the effect of increasing the activity of AdK, while comparing flux distributions 2 and 3 could provide insight into the effect of increasing CK activity. Comparison of flux distribution 5 to the above tests reducing the total flux of both enzyme systems (CK and AdK) simultaneously. Another question of interest is: Do net flux and bidirectional enzyme activity have different influences on the labeling state? Comparing flux distributions 2 and 6, as well as 7 and 8, allows one to test if the labeling state is determined by net AdK flux, or bidirectional AdK enzyme activity. Note that in those pairs of the flux distributions, the net flux of AdK shuttle changes. Finally, how does the labeling state change when the net transport of energetic phosphoryl groups occurs via the CK shuttle, or direct ATP transport? Insight into this important question could be provided by comparing flux distributions 2,8, and 9. Additional combinations of these flux distributions allow for additional comparisons. Transient solutions for these nine flux distributions after a step to 30% and 100% are provided in Figure S1. To compare the different influence of bidirectional and net fluxes we introduce two total fluxes, one each for AdK and CK. The total flux through CK is the sum of unidirectional reactions in the mitochondrial intermembrane space (IMS) and cytosol that proceed towards PCr, and the total AdK flux is the sum of unidirectional AdK reactions in the IMS and cytosol that proceed towards ADP. To perform the comparison between fluxes, in our analysis, the total flux is increased by simultaneously increasing the forward and reverse flux in one or both compartmental locations (Method), or by increasing the net flux through the shuttle (Method). Table S1 provides the expressions used to increase total flux using both methods. It should be noted that Method requires changing the net flux through one of the other parallel pathways (AdK shuttle, or direct ATP transport, or the CK shuttle. Excluding Figure 6, where the sensitivity to total CK flux was analyzed, all flux distributions chosen use a total CK flux of taken from [12] (except flux distributions 3 and 5 which have unidirectional CK fluxes). We take this as the maximum CK flux. This flux lies at the lower range determined in other -NMR saturation transfer studies [30]–[38]. Aksentijevi et al. found that the activity of CK is three times higher than AdK [47], so we took the maximum AdK flux to be, represented in flux distribution 1. Pucar et al. reported an AdK flux equivalent to [20], and a lower value of was chosen for all other flux distributions (except 4 and 5 which have unidirectional AdK fluxes). Figure 6 shows the influence of changing the total AdK flux using Method while keeping all other fluxes in the flux distribution constant (see Table 1). In this and following figures, some flux distributions overlap others over the range of the plot (flux ranges provided in Table S1), i. e. in Figure 6, flux distribution 2 is not plotted since it is a subset of flux distribution 1. All solutions for 30% are seen to give similar labeling distributions as the AdK flux is changed. Note that regardless of the differences in the net flux between these flux distributions, the labeling state depends mainly on total AdK flux. After a step change to 30% the labeling states of PCr, , and Pi are only sensitive to total AdK fluxes below. is seen to be sensitive up to at 30s, reducing to at longer sampling times, as evidenced by the variation of labeling induced by changes in total AdK flux (see Figure S2). Labeling with 100% is seen to extend the range of sensitivity to total AdK flux for all species. The use of 100% increases the rate of label incorporation into the phosphotransfer network without changing the ATP synthase or ATPase rates. This increases the ratio of the rate of label uptake to the rate of phosphotransfer reactions. Because the metabolic system given in Figure 1 is compartmentalized, it is helpful to explore if the compartmental location of the AdK flux has an influence on the labeling state. Figure S3 shows the influence of changing the compartmental location of the AdK flux while keeping both the total AdK flux and all other fluxes in the flux distribution constant. Using 30%, the compartmental location of the AdK flux is not seen to influence the labeling state, however, labeling with 100% shows that the labeling state is somewhat sensitive to the compartmental location of the AdK flux, with the largest sensitivity observed in the labeling state of. To explore how the two coupled CK fluxes influence the labeling state, a plot (Figure 7) of total CK flux versus species labeling was produced using Method while keeping the other fluxes in the system constant (see Table S1). The vertical black line shows the CK flux that was calculated by Pucar et. al based on observations of the labeling state [20] (see Methods). While the flux found in [20] was expected to be net flux through CK shuttle, total flux through the CK reaction should be at least as large as the one estimated in [20]. This plot clearly shows an insensitivity of the labeling state to total CK flux above. This means, for example, that the total CK flux determined in [12] provides almost the same labeling state of species as that provided by. This demonstrates that the dynamic method is not sensitive enough to distinguish between any CK flux above. The use of 100% is seen to slightly increase the range of sensitivity (see Figure S4). Note that the differences between solutions are due to differences in total AdK flux. Keeping total and net AdK flux constant as well as total CK flux constant, the percentage of energy exported via direct ATP transfer was varied from zero (maximum net flux through the CK shuttle) to its maximum possible value (zero net flux through the CK shuttle). The labeling states of flux distributions 1,2, 4, and 7 were plotted during this change in Figure 8. For illustration, two unidirectional CK flux distributions (3 and 5) were included in the plot and are discussed below separately. Figure 8A (30%) shows that the labeling state in the four flux distributions with reversible CK does not change appreciably as the flux of energy export is shifted from predominantly CK mediated to direct ATP transport, while Figure 8B (100%) shows a moderate shift in labeling state. Figure S5 provides additional time points for Figure 8. Looking at these we see that in practice, this moderate shift cannot be used to determine which parallel pathway carries the most flux. This is also true because a number of less characterized parameters of this model such as the reversibility of ATP synthase to oxygen exchange and the sizes of metabolic pools have comparable effects on the labeling state (see Figure 4). Considering bidirectional flux distributions 1,2, 4, and 7, in Figures 7 and 8, we see that the labeling states for each flux distribution are almost identical. This demonstrates that the labeling state is predominantly determined by total CK flux and not net CK flux. This property prevents the labeling state from explicitly defining the net flux through the CK shuttle. Similar conclusions can be reached from the analysis of energy export via AdK shuttle, as demonstrated in Figure S6. For illustration, two additional flux distributions (3 and 5) with unidirectional CK fluxes are plotted on Figure 8. In this case, it is not possible to keep the total CK flux constant over the range of the plot, so the range of Figure 8 can be interpreted as a change in total CK flux produced using Method. For these two flux distributions, the labeling state approaches that of the other four flux distributions while moving from lower to higher percentages of energy export via CK (lower to higher total CK flux). This shows that total CK fluxes above produce the same labeling state regardless of the percentage of energy export. To illustrate the properties of the pseudo-linear approximation method, used to determine the CK flux in [16] and more recently in [27], we used the same pseudo-linear approximation on predictions provided by our model. This approach is based on establishing the relationship between PCr labeling and CK flux through inhibition of CK activity by DNFB. For illustration, we used flux distribution 9, plotted in Figure 7, that provides a model prediction of the labeling state of all labeled species of CK as the total CK flux is gradually inhibited. Figure 9 applies the pseudo-linear approximation method to the total labeling curve found by combining the single, double, and triply labeled species of this model prediction. As Figure 9 demonstrates, the pseudo-linear approximation method underestimates total CK flux. Looking at the sensitivity plots in Figures 4 and 5 we see that a wide range of model parameters provide very similar predictions of the labeling state (both in magnitude and structure). Changing different parameters simultaneously is likely to result in a very similar prediction of the labeling state. This identifies that in the range of physiologically relevant cardiac performance and shuttle activity, many parameters in this model are structurally unidentifiable — including net flux (observation (I). Observation (II) does not support the suggestion that the labeling method leads to estimation of net flux through phosphotransfer systems, as proposed in [20], [24]. According to our simulations, total flux through CK and AdK reactions have a major role in determining the labeling state of metabolites (Figures 3 and 6). Thus, our simulation results suggest that the fluxes estimated using the labeling method in [20] are total fluxes and can be directly compared to the fluxes estimated using -NMR saturation studies. Observation (III) limits the use of the labeling method to study fluxes that are smaller than the ATP synthase rate, echoing a statement made by Dawis et al. [43]. The labeling method may be used to measure the flux of reactions that proceed at a rate slower than the ATP synthase rate such as reactions involving the groups of ATP or ADP [41], [42]. With regards to observation (IV), we note that the use of 100% increases the sensitivity of the method, although this is not enough to find the proportion of energy exported via direct ATP transfer and the CK shuttle. Looking at Figure S5B, we see that a ten second labeling experiment using 100% would result in a mass isotopologue distribution that could allow one to determine this split. This short duration experiment would present technical challenges, and the mixing rate of would become a critical component of the model, perhaps requiring the use of an organ level model taking into account heterogeneity within the heart. Taken together, these observations lead to the conclusion that labeling with does not provide sufficient sensitivity to study the large fluxes, such as expected for the CK shuttle, under the conditions simulated herein. However, there are a number of ways that the sensitivity of the method can be improved, for example, by: (I) increasing the rate input of into the phosphotransfer network through the use of with a larger enrichment, (II) performing experiments at higher cardiac performance, and (III) using shorter sampling times. Figure S6 shows that the labeling state is more sensitive to changes in AdK flux using 100% and labeling states predicted for the different flux distributions show a larger variation at shorter time points. The integrated kinetic model presented in this work was constructed to account for the most rapid isotope transformations that occur in the high-energy phosphotransfer reaction network in heart. As a first step in our analysis, a number of tests were conducted to determine if this model is suitable for the analysis of published dynamic labeling data. Figure 2 shows that both steady state and dynamic labeling state predictions provided by our model match the predictions from the model developed by Dawis et al. [43]. Thus, we are able to reproduce earlier published studies with our implementation of the model. The model presented here is considerably more complex because it considers the bidirectionality of reactions as well as compartmentalized metabolic pools. It should be noted, however, that this added complexity is a minimum requirement to separate out the effects of net versus total flux. Following this, we demonstrate that the model solution is consistent with published dynamic traces of labeling (see Figure 3). Intriguingly, in Figure 3, we used a flux distribution determined by -NMR inversion and saturation transfer [12] to calculate the labeling of metabolites in rat heart atrial tissue, and found that a wide range of flux distributions are able to reproduce these measured labeling states. A number of simplifying assumptions were made to construct the model we present in this work. It is explicitly assumed that all metabolic fluxes proceed at a steady metabolic rate. In addition, we do not employ enzyme kinetics in the flux simulations, although this is not seen as a trade off because the resulting model has fewer parameters and many of the enzymatic kinetic parameters are not well characterized. A number of phosphotransfer fluxes were excluded from the model. However, the dynamic labeling method may be useful to determine total fluxes lower than the ATP synthase rate, so adding reactions with fluxes of lower magnitude may allow one to determine total flux in the additional phosphotransfer pathways. This extension of the model could be used to study how these total fluxes, which include the AdK shuttle flux, are altered under diseased conditions. It is expected that adding additional phosphotransfer reactions into the model will result in a dampening of the model dynamics, including the dynamics observed in Figure 3. Unfortunately, the original labeling dynamics used for flux estimations in [20] have not been published. While early dynamic labeling studies include original labeling dynamics of individual species [41]–[43], only sums of species are reported in later studies [15]–[17], [25], [44], [45], [48]. For the heart, most of the reports include only derived data in the form of flux estimates [19]–[21], [23], [24], [26], [49]. Two heart studies include labeling dynamics reported as a sum of species in mouse and rabbit [18], [27]. A recent study reports labeling dynamics in atrial tissues taken from surgically excised male rat hearts [28]. However, no fluxes are reported in [28], thus we cannot compare the flux distributions derived from -NMR inversion and saturation transfer [12] with labeling dynamics. We produced these modeling results using a step change from to a percentage of in the surrounding water. This change provides the most sensitive change in labeling state. In practice, however, the switch between and will be slower and will reduce the sensitivity of the method relative to our predictions. As an alternative approach to the analysis presented in this work, one could compose hypothetical data sets and find the confidence intervals of model parameters. This would give an estimate of the sensitivity of the labeling method. Regardless of the approach used, we expect the conclusions to be the same. The approach used in this work was tailored towards comparison of different flux distributions to see whether the different energy transfer mechanisms proposed in the literature can be distinguished on the basis of labeling data. Without going through the published data presented in all dynamic studies [15]–[27], [44], [45], it is sufficient to state that interpretation of dynamic labeling data requires one to consider the size of metabolic pools and the bidirectionality of metabolic reactions. When comparing the total flux determined using -NMR inversion and saturation transfer analysis () [12] with the net flux determined by labeling analysis in control conditions for hearts before exposure to ischemia-reperfusion () [20], we see from Figure 6 that the same labeling state is predicted with both fluxes. Keeping in mind that the pseudo-linear approximation method may underestimate the total CK flux (see Figure 9), the underlying flux distribution in the corresponding experiments could have been the same, regardless of the fluxes predicted in the control cases in [12] and [20]. However, we should stress that the flux distributions in [12] and [20] could be different due to differences in substrates used in those studies. Importantly, our modeling results resolve all known discrepancies between the results of the dynamic labeling method, and -NMR inversion and saturation transfer. No fundamental difference was found in the nature of the fluxes being measured (net versus total), and indistinguishable labeling states were predicted using fluxes with different magnitudes. Because our model is non-discriminatory with respect to CK fluxes, we suggest that interpretation of dynamic labeling data would result in flux predictions that are compatible with -NMR inversion and saturation transfer results. It has been shown that information regarding the compartmentation of metabolites and the bidirectional nature of metabolic fluxes is contained in the dynamic component of labeling data [39]. Combining a range of sampling points from short and long labeling experiments may increase the sensitivity of isotopologue modeling because many fewer flux distributions will have same labeling dynamics that match all measured data compared with only one sampling point. Finding plausible flux distributions from dynamic data sets requires the use of an integrative kinetic model. The model composed in this work included a system of 132 ordinary differential equations that were generated using a specialized program (see Text S1). While composition of such a model is not trivial, we find it an obligatory step for the analysis of labeling dynamics. For the phosphotransfer network in the heart, sampling at an earlier time, in addition to 30 seconds, would enhance the sensitivity of the method. Adding an additional sampling point at a longer time during the approach to isotopic equilibrium would provide a better means to extract pool size information from the isotopic transient. However, the dynamic labeling method requires the sacrifice of multiple animals per time point, and adding additional sampling times will greatly increase both the ethical and monetary costs. In [20], the performance of Langendorff perfused rat hearts was relatively low, as evidenced by the rate pressure product (RPP) equal to. For comparison, the performance range used to study energy transfer in [12] was, in terms of RPP, from 1700 to. While the highest RPP value in [12] corresponded to the case which should be considered as an extreme condition representing pathology and at the limit of the isovolumic perfused heart [12], [29]. However, no signs of pathological conditions at 63300 and were observed. These levels of cardiac performance are considerably larger than the level of cardiac performance used in [20] suggesting that higher cardiac performance levels are attainable using the same isovolumic preparation. This would increase the flux through ATPase reactions and improve the sensitivity of the dynamic labeling method. In this work (Figure 5A), the upper range of simulated cardiac performance is, which roughly corresponds to. While our results show that dynamic labeling data is unable to determine CK fluxes with a higher magnitude than the ATP synthase rate, or the split between the CK shuttle and direct ATP transfer in normoxic hearts, this method is sensitive to total AdK flux because total AdK flux is expected to be lower than the ATP synthase flux [47]. Our sensitivity analysis shows that changes in total AdK flux produce significant changes in the labeling state of. This opens up the possibility of combining the dynamic labeling approach with -NMR inversion and saturation transfer. By using the same rigorous statistical testing of the model solutions against experimental data as in [12], with both types of data, it would be possible to combine the strengths of both methods with the promise of determining intracellular energy transfer flux distribution in the beating heart. It is for this reason that we view these two organ-level methods as complementary. Our results widen the discussion that attempts to reveal the mechanisms that ensure the homeostasis of metabolites during cardiac function (or not) [50]. Understanding these mechanisms will require the use of integrative kinetic models that consider all possible operating modes of this metabolic network and all possible functional purposes of all enzymes, metabolites, and dynamic effects involved in the transfer of ATP from the IMS to the myofibrils and sarcoplasmic reticulum Ca2+ ATPase pumps. The metabolic network in Figure 1 is the simplest possible model that is able to separate out the effects of net versus total flux in the CK and AdK shuttles. The reactions included in the network are a subset of the reactions known to transfer high-energy phosphoryl groups. Reactions catalyzed by the enzymes glyceraldehyde-3-phosphate dehydrogenase, 3-phosphoglycerate kinase, pyruvate kinase, hexokinase, succinyl-CoA synthase, and guanylate kinase have been excluded to simplify the system. The activities of the four glycolytic enzymes mentioned are equal to or less than the activity of AdK [47], [49]. With the exception of guanylate kinase, the net flux through all of these reactions is constrained by the stoichiometry of a larger metabolic network and is not expected to contribute appreciably to ATP regeneration. An estimate of glycolytic net flux is provided by isotopologue studies which have calculated that the anaplerotic flux into the citric acid cycle derived from glycolysis is between 3 and 12% [51] of net citric acid cycle flux which varies between 4 and in normoxic heart [52]. Because the citric acid cycle flux is much smaller than the ATP synthase flux considered here (), and is constrained by the stoichiometry of the entire metabolic network, we have excluded these reactions from the analysis. However, the reversibility of these fluxes coupled to large pools could facilitate temporary regeneration of the ATP pool in the failing heart, a phenomena recently observed by Aksentijević et al. [47]. Their reversibility could slow down the dynamics of incorporation somewhat by increasing the size of the pool of metabolites that become labeled. This phenomenon is analogous to adding a compartmentalized side pool which slows down the labeling transient as described in [39]. Because this study is concerned with only the net fluxes in normoxic heart we need not consider these reactions here. Because the model developed in this work tracks the transient exchange of each oxygen atom in the phosphotransfer network with the surrounding water environment, it is necessary to know how each oxygen atom transfers during the course of each reaction. It is known that exchange of phosphate oxygens with those of water does not occur in glycolysis [53, and references therein] and requires ATPases. Only the ATPase and ATP synthase reactions are able to transfer between and any of the four oxygen atoms of inorganic phosphate. Pi is symmetric and enrichment in each oxygen position occurs at the same rate. The enrichment observed in the three oxygen atoms of are also identical. Enzyme bound states for both ATPase and ATP synthase were included in the model because it is known that even under physiological conditions of oxidative phosphorylation multiple reversals of ATP formation occur before ATP is released to the media [54], and multiple reversals are known to occur during actomyosin catalysis [55]. No appreciable amount of positional oxygen exchange is observed between the and oxygens in ATP or the and oxygens in ADP [56]. Taken together, these properties ensure that all three oxygen atoms in every phosphoryl group of all species have an equal probability of being isotopically labeled. The derivation of the model equations assumes this behavior. The model was constructed by: (I) generating the full set of individual isotopic transformations, (II) combining these transformations into a mass balance around each isotopologue in the system, (III) composing mass isotopologue pool relations while taking into account oxygen atom mappings, and (IV) composing mass isotopologue balances by collecting the isotopologue balances according to the pool relations. The result is a system of 132 ordinary differential equations (ODEs). The intermediate equations for all of these steps are provided in supporting Text S1. The mass isotopologue equations contain variables for (I) the pool size of all 18 metabolic species and (II) the forward flux for all 20 reactions as well as the reverse flux for the 17 bidirectional reactions, giving a total of 37 unidirectional flux variables. A program was written in Python to generate these equations. This program implements symbolic manipulation tools specifically designed to carry out steps (I) through (IV), and is available as a Python module for generation of mass isotopologue equations in http: //code. google. com/p/iocbio/wiki/IOCBioOxygenIsotopeEquationGenerator. The 18 metabolic pools are assumed to be constant during the labeling processes. Pool size measurements for total ATP, PCr, and Pi were taken from Vendelin et al. for activation by Ca 1. 8 mM [12]. The sizes of these pools are 7. 551. 13,16. 42. 44, and 1. 410. 78, respectively. In [12], the pool of ADP was too small to measure, and was taken as 1% the ATP pool. The pools of these four species are split between the various compartments in the model. In the IMS, the fraction of the total pool of ATP, PCr, and ADP was taken to be 1% of the total amount of each metabolite. The fraction of the total pool of ATP, ADP, and Pi in the enzyme bound ATPase and ATP synthase states was taken to be 0. 05% of the total amount of each metabolite. The fraction of ATP, ADP, and Pi in the mitochondrial matrix was taken to be 12. 5% from measurements of the ATP pool [12]. The pool size of enzyme bound water was taken to be the same size as the enzyme bound ATP pool. The 20 net flux variables in the model are constrained to metabolic steady state by three independent net flux variables. We chose these to be the net rate of ATP synthase, the net flux of ATP between the IMS and the cytosol, and the net flux of PCr between the IMS and the cytosol. A set of 17 relations between these and all other net flux variables was found using a method we recently developed [57]. The forward and reverse fluxes for the 17 bidirectional reactions were constructed by combining the net flux with an exchange flux, as described by Wiechert and Graaf [58]. Figure 1B in the paper by Pucar et al. reports a CK flux of 330 [20]. All metabolite data are expressed in of intracellular water, assuming 2. 72 intracellular water to total protein content, as in Vendelin et al. [12], so the CK flux in [20] converts to. Likewise, Figure 2B in [20] reports an AdK flux of 45. 6 which converts to. The citric acid cycle flux is reported by Des Rosier et al. to be between 0. 1 and 4 in [52]. As above, assuming 2. 72, and 160 as in Vendelin et al. [12], these values convert to between 4 and. To solve the system of ODEs we used a variable-coefficient ODE solver with the Backward Differentiation Formula method [59] provided by SciPy (http: //www. scipy. org). Our system of ODEs was implemented in C for computational efficiency and exposed to Python using f2py [60] for efficient prototyping.

In heart, the movement of energy metabolites between force-producing myosin, other ATPases, and mitochondria is vital for its function and closely related to heart pathologies. In addition to diffusion, transport of ATP, ADP, Pi, and phosphocreatine occurs along parallel pathways such as the adenylate kinase and creatine kinase shuttles. Two organ-level methods have been developed to study the relative flux through these pathways. However, their results differ. It was recently demonstrated that studies often suffer from the exclusion of compartmentation from their metabolic models. One study overcame this limitation by using compartmental models and statistical methods on multiple experiments. Here, we analyzed the sensitivity of the other method - dynamic labeling of phosphoryl groups and inorganic phosphate. For that, we composed a mathematical model tracking enrichment of the metabolites and evaluated sensitivity of labeling to different flux distribution scenarios. Our study shows that the dynamic method provides a measure of total flux, and not net flux as presumed previously, making the fluxes predicted from both methods consistent. Importantly, conclusions derived on the basis of labeling analysis, particularly those regarding the net flux through the shuttles in control and pathological cases, need to be reevaluated.

Abstract Introduction Results Discussion Methods

animal models medicine applied chemistry nuclear magnetic resonance model organisms chemistry biology computational biology metabolic networks biophysics rat cardiovascular chemical properties

2012

Sensitivity Analysis of Flux Determination in Heart by H218O -provided Labeling Using a Dynamic Isotopologue Model of Energy Transfer Pathways

The transferrin receptor of bloodstream form Trypanosoma brucei is a heterodimer encoded by expression site associated genes 6 and 7. This low-abundance glycoprotein with a single glycosylphosphatidylinositol membrane anchor and eight potential N-glycosylation sites is located in the flagellar pocket. The receptor is essential for the parasite, providing its only source of iron by scavenging host transferrin from the bloodstream. Here, we demonstrate that both receptor subunits contain endoglycosidase H-sensitive and endoglycosidase H-resistant N-glycans. Lectin blotting of the purified receptor and structural analysis of the released N-glycans revealed oligomannose and paucimannose structures but, contrary to previous suggestions, no poly-N-acetyllactosamine structures were found. Overlay experiments suggest that the receptor can bind to other trypanosome glycoproteins, which may explain this discrepancy. Nevertheless, these data suggest that a current model, in which poly-N-acetyllactosamine glycans are directly involved in receptor-mediated endocytosis in bloodstream form Trypanosoma brucei, should be revised. Sequential endoglycosidase H and peptide-N-glycosidase F treatment, followed by tryptic peptide analysis, allowed the mapping of oligomannose and paucimannose structures to four of the receptor N-glycosylation sites. These results are discussed with respect to the current model for protein N-glycosylation in the parasite. Finally, the glycosylation data allowed the creation of a molecular model for the parasite transferrin receptor. This model, when placed in the context of a model for the dense variant surface glycoprotein coat in which it is embedded, suggests that receptor N-glycosylation may play an important role in providing sufficient space for the approach and binding of transferrin to the receptor, without significantly disrupting the continuity of the protective variant surface glycoprotein coat. The tsetse-transmitted Trypanosoma brucei group of parasites cause human African trypanosomiasis and nagana in cattle and constitute a serious health problem for people and livestock in 36 countries of sub-Saharan Africa. T. brucei exists in the mammalian host as the bloodstream form trypomastigote and in the midgut of the tsetse fly vector as the procyclic form. The major surface molecules of the bloodstream form parasite are the glycosylphosphatidylinositol (GPI) anchored [1]–[4] and N-glycosylated [3]–[6] variant surface glycoproteins (VSGs), 5×106 homodimers of which form a dense monolayer over the whole trypanosome [4]. The ability of individual trypanosomes to switch expression from one VSG gene to another gives rise to antigenic variation by which the parasite population survives the host acquired immune response [7]. Other less abundant glycoproteins are arranged either apparently randomly within the VSG coat, like the invariant glycoproteins ISG65 and ISG75 [8], [9], while others have specific surface locations, like Fla1 which locates to the flagellar adhesion zone [10] and the transferrin receptor which locates to the flagellar pocket [11]. Still other glycoproteins are located primarily in intracellular sites, like lysosomal p67 [12], Golgi and lysosomal tGLP1 [13], endoplasmic reticulum GPIdeAc [14] and endosomal TbMBAP1 [15]. The surface of the procyclic form parasite is dominated by 3×106 copies of the GPI-anchored and N-glycosylated procyclin glycoproteins [4], [16], [17], about 1×106 free GPI glycolipids [18], [19] and a high-molecular weight glycoconjugate complex [20], [21]. While this life cycle stage shares some glycoproteins with the bloodstream form, like p67, tGLP1 and Fla1, others are clearly bloodstream form specific, like ISG65, ISG75, TbBMAP1 and the expression site associated gene (ESAG) 6 and ESAG7 subunits of the heterodimeric T. brucei transferrin receptor (TfR). Some of these glycoproteins are encoded by polygene families, causing sequence heterogeneity in the populations expressed by the trypanosomes. In the case of the TfR ESAG6/ESAG7 subunits, the ESAG6 and ESAG7 genes are associated with telomeric VSG expression sites such that one dominant ESAG6/ESAG7 pair dominates according to which site (and VSG variant) is being expressed [22]. However, there is also some transcriptional breakthrough from other expression sites, as the ESAG6 and ESAG7 genes are immediately adjacent to the expression site promoters, providing some sequence heterogeneity in all TfR preparations [23]. There is functional significance with respect to which ESAG6/ESAG7 pair is expressed due to their different affinities for transferrins from different mammalian species [24], [25]. While there are quite complete data on the GPI anchor and N-glycan structures and N-glycosylation site occupancies of specific VSGs and procyclins [1]–[6], [17], [26] and on the structures of the total N-glycan repertoires of the bloodstream form [27], [28] and procyclic form [29], [30] of the parasite, there is a paucity of data of the glycosylation status of other specific T. brucei glycoproteins. In this paper, we describe the N-glycosylation status of the ESAG6 and ESAG7 subunits of the transferrin receptor (TfR) and, together with our previous description of the GPI anchor of the ESAG6 subunit [31], provide a relatively complete description of the glycosylation status of this low abundance (approximately 3000 copies per cell [32]) but nutritionally essential [33] glycoprotein. The results are discussed in the context of proposed mechanisms of protein N-glycosylation [34]–[37] and endocytosis [38] in T. brucei. We also build a molecular model of the glycosylated ESAG6/ESAG7 transferrin receptor, surrounded by models of glycosylated VSG molecules, to visualize how this receptor sits in the VSG coat on the flagellar pocket membrane and how it might bind its transferrin ligand. The transferrin receptor (TfR) was purified by affinity chromatography on immobilized transferrin, following the method first described by Steverding and Overath [39]. An aliquot was analyzed by SDS-PAGE and silver staining, which showed the characteristic ESAG6 and ESAG7 subunits (Figure 1A). The identities of the ESAG6 and ESAG7 components were confirmed by excision of the individual bands, in gel tryptic digestion and proteomic analysis (data not shown). Endoglycosidase digests confirmed that both ESAG6 and ESAG7 carry N-linked oligosaccharides. Thus, digestion with both peptide N-glycosidase F (PNGase F), an enzyme that cleaves essentially all types of N-linked glycan, and Endoglycosidase H (Endo H), an enzyme that cleaves only oligomannose-type N-glycans, reduced the apparent molecular weights of both proteins, as judged by SDS-PAGE and Western blotting with anti-TfR antibodies (Figure 1B). However, PNGase F reduced the apparent molecular weights of both proteins more than Endo H (Figure 1B, compare lanes 1 and 3), suggesting that both proteins contain a mixture of Endo H-sensitive (i. e. , oligomannose) and Endo H-resistant (i. e. , paucimannose and/or complex) N-glycans. The heterogeneity still apparent in ESAG6 following complete de-N-glycosylation with PNGase F is presumably due to the reported heterogeneity in the α-galactose side chains of the GPI anchor attached to this TfR subunit [31]. Aliquots of purified TfR were separated by SDS-PAGE, blotted onto nitrocellulose and probed with anti-TfR antibody (Figure 2, lane 1) and by lectins. Consistent with the presence of oligomannose N-glycans, both ESAG6 and ESAG7 subunits gave a positive reaction with concanavalin A (ConA) (Figure 2, lane 2), as did the bovine ribonuclease B positive control glycoprotein (Figure 2, lane 4). These reactions were abolished when α-methyl-mannose was included in the blotting buffer (Figure 2, lanes 3 and 5), demonstrating the carbohydrate specificity of the ConA blots. However, neither of the ESAG subunits gave a significant reaction with the poly-LacNAc-specific tomato lectin (Figure S1) or, more importantly, with the far more permissive [40] N-acetyllactosamine (LacNAc) specific lectin from Erythrina cristigalli (ErCr) (Figure 2, lane 6) or with the terminal β-galactose-specific lectin ricin (Figure 2, lane 10). These experiments were performed under conditions where a strong reaction was seen against the positive control glycoprotein bovine asialotransferrin (Figure 2, lanes 8 and 12) and where the reactions with the ErCr and ricin lectins against the positive control were abolished by the inclusion of lactose or galactose plus lactose, respectively, in the blotting buffer (Figure 2, lanes 9 and 13). These data suggest that the Endo H-resistant N-glycans of ESAG6 and ESAG7 are not of the poly-LacNAc-containing complex type nor, indeed, even of the LacNAc-containing complex type and are, therefore, most likely of the paucimannose type. The lectin blotting experiments, described above, suggested that ESAG6 and ESAG7 contain oligomannose and paucimannose N-glycans. However, there remained the formal possibility that the Endo H-resistant N-glycan fraction might include complex N-glycans fully capped with terminal α-Gal residues, which could abrogate ricin and ErCr lectin binding to the sub-terminal β-Gal residues and LacNAc units, respectively, and for which there is precedent in some VSG N-linked glycans [5]. Therefore, to analyze the N-glycan structures further, total N-glycans were released from TfR with PNGase F, radiolabeled by reduction with NaB[3H]4 and analyzed by high-performance thin layer chromatography (HPTLC) alongside radiolabeled N-glycan standards (Figure 3A). A ladder of bands was observed, stretching from the position of Man9GlcNAc2 to Man5GlcNAc2, with two additional bands of higher Rf, possibly corresponding to Man4GlcNAc2 and Man3GlcNAc2 paucimannose species. Significantly, there were no bands with Rf values consistent with complex N-glycans capped with terminal α-Gal residues or with poly-LacNAc-containing N-glycans, like those found in VSG variant MITat1. 7 [5] (Figure S2). The radioactive material at the origin of the TLC plate in (Figure 3A) is present in all NaB[3H]4-labeled samples, including commercial glycan standards (Figure S2). A sample of the mixture of labeled N-glycans was separated by Dionex high-pH anion exchange chromatography (HPAEC) and three major radioactive peaks were recovered (Figure S3). These were individually analyzed by HPTLC alongside authentic radiolabeled N-glycan standards and it was found that peak b and peak c co-migrated with Man5GlcNAc2 by HPTLC, while peak a migrated ahead of Man5GlcNAc2 and was assigned as a putative Man4GlcNAc2 structure (Figure 3B). Consistent with the latter assignment, digestion of the peak a material with the Aspergillus saitoi Manα1-2Man-specific α-mannosidase (ASαM) caused an increase in Rf equivalent to the removal of a single hexose (Figure 3C, compare lanes 1 and 2). In contrast, the majority of the material in the peak b fraction was resistant to ASαM (Figure 3C, compare lanes 3 and 4), suggesting that this is a tri-antennary Man5GlcNAc2 structure of the conventional oligomannose series. By inference, we assign the peak c material as the bi-antennary Man5GlcNAc2 structure of the paucimannose series and, indeed, a small component of the peak b material does digest with ASαM to lose two hexose residues, suggesting this is a small amount of bi-antennary Man5GlcNAc2 contamination from the adjacent peak c (Figure 3C, lane 4). Unfortunately, there was insufficient radiolabeled purified peak c material on which to perform a separate ASαM digest. The proposed structures of the main N-glycan species are shown in (Figure 3A). These structures are consistent with the data in (Figure 3A–C) and also draw on our prior knowledge of the structures of the oligomannose and paucimannose series in bloodstream form T. brucei [5], [6], [28]. The aforementioned endoglycosidase digestion results, lectin blots and N-glycan structural analyses strongly suggest that ESAG6 and ESAG7 contain both oligomannose and paucimannose N-glycans, but not complex N-glycans. Previous work has shown that bloodstream form T. brucei expresses two classes of oligosaccharyltransferase (OST] activity [34]–[37]: One that transfers Man5GlcNAc2 from Man5GlcNAc2-PP-Dol to N-glycosylation sequons in relatively acidic environments and another that transfers Man9GlcNAc2 from Man9GlcNAc2-PP-Dol to the remaining N-glycosylation sequons. These activities are encoded by the TbSTT3A and TbSTT3B genes, respectively [34]. We therefore subjected purified TfR to Endo H digestion followed by PNGase F digestion, resolved the double-digested ESAG6 and ESAG7 by SDS-PAGE and performed in-gel tryptic digestion and analyzed the resultant peptides by LC-MS/MS. Using this protocol [34], peptides encompassing Endo H-sensitive (oligomannose) N-glycosylation sites appear with a 203 Da shift, from the single GlcNAc residue left attached to the Asn residue by Endo H, and peptides encompassing Endo H-resistant (paucimannose) N-glycosylation sites appear with a 1 Da shift, from the conversion of Asn to Asp by PNGase F. Using this technique, we were able to positively identify three of the five N-glycosylation sites of ESAG6 as occupied, one (Asn94) with Endo H-resistant paucimannose N-glycans and two (Asn10 and Asn344) with Endo H-sensitive oligomannose N-glycans. The pI values of these Asn-Xaa-Ser/Thr sequons ±5 amino acid residues are consistent with their modification by TbSTT3A and TbSTT3B OST activities, respectively [34] (Table 1). Peptides encompassing the remaining two putative N-glycosylation sites, at Asn219 and Asn234, were not observed but their surrounding sequences would suggest that they are both modified by TbSTT3B OST and are likely to carry oligomannose structures (Table 1). In the comparable ESAG7 analysis, we positively identify one (Asn10) of the three N-glycosylation sites as occupied with Endo H-sensitive oligomannose N-glycans, consistent with its modification by TbSTT3B. Peptides encompassing the remaining two putative N-glycosylation sites, at Asn94 and Asn218, were not observed but their surrounding sequences would suggest that Asn94 is modified by TbSTT3A OST and likely to carry paucimannose structures and Asn218 is modified by TbSTT3B OST and likely to carry oligomannose structures (Table 1). Representations of the glycosylation of the ESAG6 and ESAG7 subunits of the TfR are shown in (Figure 4). The proteomics analysis of the TfR components (described above) also indicated that the principal ESAG6 and ESAG7 sequences present the purified TfR preparation corresponded to those deposited under accession numbers CAQ57442. 1 and CAQ57441. 1, respectively. Nolan and colleagues that have reported that TfR can be isolated from a trypanosome lysate with tomato lectin-Sepharose [38]. However, we did not identify any tomato lectin (TL) binding poly-LacNAc-containing N-glycans in either subunit of trypanosome TfR. We therefore entertained the possibility that TfR binds indirectly to TL through interaction with other glycoprotein (s) that do bear poly-LacNAc-containing N-glycans. To investigate this, we took osmotically lysed cells, depleted of VSG and TfR by the action of endogenous GPI-PLC on their GPI anchors, and isolated the total ricin-binding glycoprotein fraction, that includes the TL binding glycoproteins as a significant sub-set [27], and separated and immobilized them by SDS-PAGE and Western blot. The presence of TL-binding glycoproteins was confirmed by probing the blot with TL (Figure 5, lane 3) and the carbohydrate-specificity of this signal was confirmed by inhibition with chitin hydrolysate (Figure 5, lane 4). Identical blots were probed with anti-TfR antibodies before and after pre-incubation with purified TfR. Without pre-incubation with purified TfR, the anti-TfR blots were devoid of significant signal (Figure 5, lane 1), whereas with pre-incubation with purified TfR the anti-TfR blots showed two clear bands at apparent molecular weights of around 55 kDa and 97 kDa. From these data we conclude that TfR is able to bind to other glycoproteins that, in turn, can bind to ricin and therefore possibly also to TL. Based on the widely accepted assumption that T. brucei TfR has a similar tertiary structure and quaternary structure to the N-terminal domain of VSG [41], [42], for which there are crystallographic data [43], we have made a homology model of the ESAG6/ESAG7 heterodimer of TfR and added to this representative N-linked glycan structures, according to the data and predictions presented in this paper (Table 1 and Figure 4), and a GPI anchor [31]. VSG MITat1. 2 was modeled based on the crystal structure of the N-terminal domain [43], the NMR structure of the C-terminal domain [44], and representative N-linked glycan and GPI anchor structures [2], [5], [35], [36]. The N-terminal and C-terminal domains were placed with relatively compact linkers between the two domains and between the C-terminal domain and the GPI anchor. With extended linkers the two domains could be displaced significantly further from the membrane. Human transferrin was modeled based on the structure of iron-bound human transferrin in complex with the human transferrin receptor [45] and representative N-linked [46] and O-linked [47] glycans. The comparison between the models of TfR and VSG MITat1. 2 is shown (Figure 6A). A model of TfR surrounded by VSG molecules at their expected surface density [48] is also shown (Figure 6B). Into this model we have placed a model of glycosylated human transferrin, in the same orientation in which it docks to the human receptor [45] (Figure 6C). Although the TfR model is based on the specific ESAG6 and ESAG7 species found in our TfR preparation (accession numbers CAQ57442. 1 and CAQ57441. 1), the highly conserved amino acid sequences and glycosylation sites of the T. brucei brucei ESAG6 and ESAG7 families (Table S1 and Table S2) suggests that it would be reasonable to assume that this is a general model for all T. brucei brucei ESAG6/ESAG7 heterodimers. As well as contributing to a three dimensional model of T. brucei TfR, the experimental data on N-glycosylation site occupancy for three of the five N-glycosylation sites of ESAG6 and one of the three for ESAG7 presented in this paper (Table 1; Figure 4) provide support for the model of a unique mechanism of protein N-glycosylation in T. brucei [34]–[37]. According to this model, T. brucei N-glycosylation sequons in relatively acidic environments (like Asn94 of ESAG6) co-translationally receive exclusively (Endo H-resistant) biantennary Man5GlcNAc2 through the action of an oligosaccharyltransferase (OST) encoded by the TbSTT3A gene whereas the remaining sites (like Asn10 and Asn344 of ESAG6 and Asn10 of ESAG7) are acted upon post-translationally by an OST encoded by the TbSTT3B gene and receive exclusively (Endo H-sensitive) triantennary oligomannose Man9GlcNAc2. Once transferred to protein, the biantennary Man5GlcNAc2 structure on the acidic sites may then be processed to paucimannose (Man4GlcNAc2 and Man3GlcNAc2) structures with the latter, in some cases, further elaborated to complex glycan structures. Apparently this further processing to complex glycans does not occur on ESAG6 or ESAG7, where Man4GlcNAc2 appears to be the predominant endo H-resistant structure. The triantennary oligomannose Man9GlcNAc2 structures at the non-acidic sites can only be maximally processed to the triantennary oligomannose Man5GlcNAc2 structure, which appears to be the predominant endo H-sensitive structure on ESAG6 and ESAG7. Another recent analysis of the single N-glycan of VSG MITat. 1. 8 also supported the model in [34]. In this case, a single acidic N-glycosylation site at Asn59 was found to be occupied exclusively by a biantennery complex N-glycan structure of Galβ1-4GlcNAcβ1-2Manα1-3 (Galβ1-4GlcNAcβ1-2Manα1-6) Manβ1-4GlcNAcβ1-4GlcNAc. Presumably, and in contrast to the VSG MITat. 1. 8 example, the acidic TfR N-glycosylation sites fail to be processed beyond the trimming of up to two α1-2-linked mannose residues due to steric constraints, reducing access by α-mannosidases and preventing subsequent access by βGlcNAc-transferases. It was suggested by Nolan and colleagues that T. brucei TfR contains poly-LacNAc glycans because ESAG6 and ESAG7 in whole cell detergent lysates bound to tomato lectin (TL) beads [38]. These authors further suggested a tentative model for endocytosis in trypanosomes, postulating an interaction between poly-LacNAc N-glycans on TfR (and other receptors) and a protein in the flagellar pocket with an extracellular TL lectin-like domain and a cytoplasmic domain that interacts with the machinery of the endocytic pathway. This model was supported by an approximately 5-fold reduction in transferrin endocytic rate when trypanosomes were incubated in 15 mM each of tri-N-acetyl-chitotriose and tetra-N-acetyl-chitotetraose. However, our data show that TfR does not contain any poly-LacNAc structures. Therefore, a direct link between receptor-linked poly-LacNAc glycans and endocytic machinery can be ruled out. However, we have shown here that TfR is able to bind to immobilized ricin-binding glycoproteins. Since the TL-binding glycoproteins of T. brucei are a sub-set of the ricin-binding fraction [27], these data may explain why TfR was found in the TL-binding fraction [38], i. e. , through the non-covalent association of TfR with other glycoproteins. It should be pointed out that the association seen in the TfR overlay experiment could be through protein-protein and/or protein-carbohydrate interaction (s) and that, relevant to possible protein-carbohydrate interactions, the glycoproteins in the ricin-binding fraction contain oligomannose and paucimannose glycans as well as conventional complex and poly-LacNAc containing N-glycans. The latter two classes of glycan bind directly to ricin while the former are present because many glycoproteins contain a mixture of both oligomannose and/or paucimannose and complex and/or poly-LacNAc glycans attached to different glycosylation sites in the same polypeptide. While the indirect association of TfR with other glycoproteins could still be relevant for poly-LacNAc-mediated endocytosis in theory, the normal in vitro growth rate of bloodstream form trypanosomes under TbSTT3A RNAi knockdown [34], when the synthesis of almost all complex N-glycans (including poly-LacNAc glycans) is abrogated, also brings the model of poly-LacNAc-mediated endocytosis into question. We therefore suggest that we should return to a null hypothesis: That transferrin, captured by the parasite TfR embedded in the VSG coat, is endocytosed constitutively in clathrin-coated vesicles and that the extremely rapid turnover of the flagellar pocket membrane in bloodstream form T. brucei [49], [50] provides a sufficient rate of uptake of this (and other) essential macromolecular nutrients from host serum. The molecular modeling of TfR alongside VSG shows that TfR is predicted to sit low in the VSG coat. However, the N-glycans of TfR significantly increase the surface area occupied by TfR compared with VSG. This may be physiologically relevant since the TfR glycans may contribute to protecting the underlying plasma membrane from lytic host serum components while providing sufficient space to allow access of the 80 kDa bi-lobed transferrin glycoprotein, regardless of the relative orientations of the receptor and the ligand (which are currently unknown) when binding takes place. Thus, the widest diameter of transferrin is significantly larger than that of aglycosyl TfR but similar to that of glycosylated TfR. Previously, we had speculated that since the single dimyristoyl-GPI anchor of the ESAG6/ESAG7 TfR heterodimer (as compared to the twin dimyristoyl-GPI anchors of VSG homodimers) would lead to relatively weak association of TfR with the flagellar pocket membrane [51], this might allow TfR to leave the membrane and dock with transferrin in the fluid phase of the flagellar pocket [31]. While the molecular modeling presented here does not altogether rule that model out, it does suggest that the TfR does not de facto have to leave the membrane to dock with its ligand. Finally, one would predict that transferrin/TfR accessibility at the flagellar pocket membrane is under extreme spatial constraint to prevent complement activation by the underlying plasma membrane. In other words, one would predict that TfR should be able make sufficient space within the VSG coat to allow transferrin to approach and be captured but without exposing significantly more underlying plasma membrane than found throughout the rest of the VSG coat. This tuning of the space occupied by TfR appears to be satisfied by N-glycosylation of both of its subunits and it may explain why TfR has so many N-glycosylation sites (eight) compared to the structurally-related VSGs, which generally have only two or four N-glycans per VSG dimer [4]. Rodents were used to propagate sufficient T. brucei parasites for the purification of sufficient transferrin receptor for high-sensitivity structural analyses. The animal procedures were carried out according the United Kingdom Animals (Scientific Procedures) Act 1986 and according to specific protocols approved by The University of Dundee Ethics Committee and as defined and approved in the UK Home Office Project License PPL 60/3836 held by MAJF. The transferrin receptor was purified from blood stream form trypanosomes as previously described by Mehlert and Ferguson [31] using affinity chromatography with transferrin-Sepharose which was first described in [39]. Exoglycosidase digests were carried out using both N-glycanase F and Endoglycosidase H as described in Izquierdo et al [34]. The exoglycosidase digests were analyzed by reducing SDS-PAGE with 4–12% gradient gels (Invitrogen), using MOPs buffer and then Western blotting onto nitrocellulose (GE Healthcare) as in [28]. After blocking and incubating in rabbit polyclonal anti-transferrin receptor (kindly supplied by Dietmer Steverding) at the dilution of 1 in 1000 then washing several times in blocking buffer, the membranes were incubated in Anti-rabbit HRP at a dilution of 1 in 20,000. After further washing visualization of the bands was achieved using ECL reagents (GE Healthcare). SDS PAGE and Western blotting was carried out as above and then the membranes were stained using lectins as described in [34]. All lectin-biotin conjugates were obtained from Vector laboratories. Concanavalin A conjugated to biotin was used at a dilution of 1 in 3,000 (with or without 0. 5 M α-methyl-mannose). Ricin-biotin was used at a dilution of 1 in 3,000 (with or without 10 mg/ml galactose and 10 mg/ml lactose), tomato lectin-biotin conjugate diluted was used at a dilution of 1 in 10,000 (with or without chitin hydrolysate, Vector Laboratories, at a dilution of 1 in 10), ErCr lectin was used at a dilution of 1 in 3,000 (with or without 200 mM lactose). The blots were washed extensively after being incubated with the lectin solutions and were incubated in streptavidin-HRP obtained from Sigma Aldrich and diluted to 1 in 10,000. Bands were visualized using ECL reagents as above. The N-glycans of the trypanosomal heterodimeric transferrin receptor were released by PNGase-F and labeled with sodium borotritiide following the method described in [28]. After extensive cleanup steps to remove any contaminating tritiated material [28] the 3H-labeled glycans were analyzed by HPTLC [Merck silica gel 60] and fractionated by HPAEC as described in [28] and fractions were pooled according to the amount of radioactivity after 10% was used for scintillation counting. Some of the pools were digested using the broad specificity alpha mannosidase extracted from Canavalia ensiformis (jack beans) (Sigma-Aldrich) and the α1-2 specific alpha mannosidase extracted from Aspergillus saitoi (Prozyme), as described in [2]. After digestion the samples were desalted using a mixed bed column as described in [2] and then analyzed again by HTPLC as above. The HTPLC plates were run 3 times in butanol ∶ methanol ∶ water, 4 ∶ 4 ∶ 3 (v/v), with drying between each run, then dried, sprayed with En3Hance (Perkin Elmer) and fluorographed with intensifying screens for up to 8 weeks at −80°C. Samples of transferrin receptor were digested with Endoglycosidase H followed by PNGaseF digestion (Roche), then analyzed by SDS-PAGE as above. Following staining with Simply Blue (Sigma-Aldrich) the bands corresponding to ESAG6 and 7 were cut out and subjected to proteomic analysis. An aliquot of the tryptic digest was analyzed by LC-MS on an LTQ Orbitrap XL (Thermo) using a Dionex 3000 Nano-LC as in [34]. The resulting data were analyzed using Mascot and the T. brucei geneDB protein database using variable modifications of N-acetylated glucosamine modification of Asn, which would signify an Endoglycosidase H sensitive site, and deamidation of Asn to Asp, which would signify an Endoglycosidase H resistant site, as described in [34]. A ricin-binding total glycoprotein fraction from bloodstream form T. brucei [27] was subjected to SDS-PAGE and transfer to nitrocellulose. These blots were probed with tomato lectin (with and without chitin hydrolysate inhibitor), as described above, or with or without purified TfR (approximately 0. 2 µg/ml in phosphate buffered saline). The latter blots were subsequently probed with anti-TfR antibody with ECL detection, as described above. Molecular modeling was performed on a Silicon Graphics Fuel workstation using InsightII and Discover software (Accelrys Inc. , San Diego, USA). Figures were produced using PyMol (The PyMOL Molecular Graphics System, Schrödinger, LLC). Protein structures used for modeling were obtained from the pdb database [52]. The homology model of T. brucei TfR was based on crystal structure of VSG MITat1. 2 (pdb code - 1vsg [43]). The sequence alignment between ESAG6, ESAG7 and VSG was based on [44], modified to take account of the protein tertiary structure and the additional disulphide bonds present. The formation of the disulphide bonds in ESAG6 and ESAG7 between residues equivalent to residues 62 and 286 in MITat1. 2 required a distortion of the helix starting at residue 61 and a rearrangement of the loop-containing residue 286. The additional disulphide bond in ESAG7 between residues equivalent to residues 203 and 220 in MITat1. 2 could be accommodated with no alteration in the secondary or tertiary protein structure. The model of VSG MITat1. 2 was based on the crystal structure of the N-terminal domain (pdb code – 1vsg [43]) and the NMR structure of the C-terminal domain (pdb code – 1xu6 [44]). The C-terminal domain was placed directly below the N-terminal domain [44] to allow for the dense packing of the N-terminal domains on the trypanosome surface [48]. The linkers between the two domains and between the C-terminal domain and the GPI anchor were modeled as relatively compact random loops. The model of human transferrin was based on the structure of iron-bound transferrin in complex with the human transferrin receptor (pdb code – 1suv [45]), N-linked and O-linked glycan structures and GPI anchors were added to all models as appropriate. The structure of the glycans were generated using the database of glycosidic linkage conformations [52] and in vacuo energy minimisation to relieve unfavorable steric interactions. The Asn-GlcNAc linkage conformations were based on the observed range of crystallographic values [53], [54] the torsion angles around the Asn Cα-Cβ and Cβ-Cγ bonds then being adjusted to eliminate unfavorable steric interactions between the glycans and the protein surface. The following GenBank protein sequence accession numbers were used in this study: CAQ57442. 1 and CAQ57441. 1. The following Protein Data Bank (pdb) files were used in this study: 1vsg, 1xu6,1suv.

The tsetse fly transmitted parasite that causes human African trypanosomiasis, or sleeping sickness, scavenges iron from the bloodstream of the infected individual so that it can live, multiply and ultimately cause disease. To do this, it places a glycoprotein (a protein with carbohydrate chains attached) called the transferrin receptor on its surface to capture circulating human transferrin, an iron transport protein. It then internalizes transferrin receptor/transferrin complex and digests the transferrin part, releasing the iron for its own use. By analyzing the parasite transferrin receptor, we have been able to describe the carbohydrate chains of the transferrin receptor and thus complete a molecular model of this important glycoprotein. We have further built models of how we expect this low abundance glycoprotein will sit in the surface coat of the parasite, which is made of millions of copies of another glycoprotein. The results provide a ‘molecule' s eye view’ of how the carbohydrate chains of the transferrin receptor provide the space necessary for the transferrin to bind to it without disrupting the protective coat.

Abstract Introduction Results Discussion Materials and Methods

biochemistry analytical chemistry chemistry biology

2012

Modeling of the N-Glycosylated Transferrin Receptor Suggests How Transferrin Binding Can Occur within the Surface Coat of Trypanosoma brucei

The identification of cell cycle–related genes is still a difficult task, even for organisms with relatively few genes such as the fission yeast. Several gene expression studies have been published on S. pombe showing similarities but also discrepancies in their results. We introduce a network in which the weight of each link is a function of the phase difference between the expression peaks of two genes. The analysis of the stability of the clustering through the computation of an entropy parameter reveals a structure made of four clusters, the first one corresponding to a robustly connected M–G1 component, the second to genes in the S phase, and the third and fourth to two G2 components. They are separated by bottleneck structures that appear to correspond to cell cycle checkpoints. We identify a number of genes that are located on these bottlenecks. They represent a novel group of cell cycle regulatory genes. They all show interesting functions, and they are supposed to be involved in the regulation of the transition from one phase to the next. We therefore present a comparison of the available studies on the fission yeast cell cycle and a general statistical bioinformatics methodology to find bottlenecks and gene community structures based on recent developments in network theory. The cell cycle is a highly controlled ordered set of events, culminating in cell division into two daughter cells. The cell division requires doubling of the genome (DNA) during the synthesis phase (S phase) and halving of that genome during mitosis (M phase). The period between M and S is called G1; that between S and M is G2. Microarray technologies have been used to identify cell cycle genes in several organisms (human, Saccharomyces cerevisiae, Schizosaccharomyces pombe, and Arabidopsis thaliana) [1,2]. Datasets are generated using different synchronization conditions and time measurements [3]. Among them, centrifugal elutriation produces a homogeneous population of small cells early in their cell cycle, while temperature-sensitive mutants show arrest in specific cell cycle stages at a restrictive temperature. The mRNA is extracted at a number of time points following synchronization. After measuring the expression level for all genes, those expressed in a periodic manner are identified using several different methods, such as Fourier analysis [4,5]. The result is the assignment of a cell cycle phase to each gene that has been detected as periodically regulated. The cell cycle of the fission yeast S. pombe lasts approximately 3 h. Its structure is the same as in all other eukaryotes. However, S. pombe is the only yeast that divides by fission, a symmetrical process in which the old cell grows until it divides, with the formation of a central mitotic spindle, into two equal new cells. As a consequence, it is characterized by a very long G2 phase of overall increase of the cell mass that covers 70% of the cell cycle. The M phase is marked by chromosome condensation and segregation to opposite ends of the cell. Then the cell goes rapidly through the G1 phase with the synthesis and accumulation of active proteins required for DNA replication. Therefore, by the time cytokinesis occurs, the S phase is completed and an entire complement of chromosomal DNA is synthesized. Recently, three independent studies have made available gene expression data on the cell cycle of fission yeast [6–8]. They measured gene expression as a function of time in both wild-type elutriation and cdc25 block-and-release experiments, and they identified different datasets (Table 1). A total number of almost 1,400 genes are found to oscillate in the three studies. About 10% of these genes are identified as periodically regulated in all the three studies and less than 30% in at least two of them. The definition of cell cycle–regulated genes is far from being rigorous. The identity and the numbers of genes in the periodic datasets strongly depend on the approach and on how conservative one wants to be. Instead of looking at the single gene, we define a periodic cell cycle network and study its cluster structure to find universal properties that are stable despite differences in the datasets. Both Rustici et al. [6] and Peng et al. [7] identified four clusters of periodic genes, corresponding roughly to the four main phases of the cell cycle, while Oliva et al. [8] proposed eight different clusters. Nevertheless, the distribution of the phases only reveals two clear expression waves. We consider the periodic cell cycle network corresponding to the intersection of the three datasets, and we study the clustering and its stability [9,10]. At first, two main components appear. The first one groups all genes in the M, G1, and S phases, and the second corresponds to the entire G2 phase. They fit the pattern shown in the distribution of the phases. Further search for hierarchical substructures of these two clusters shows that the M and G1 phases form a robustly connected single component, while the G2 phase can be divided into two clusters, and the S phase forms a separated component of its own. The stability measure indicates that a structure made of four clusters represents the more reliable pattern of the distribution of the periodic genes on the cell cycle. These clusters are separated by bottleneck structures corresponding to cell cycle checkpoints. We will discuss a set of genes located on these bottlenecks. Genome-wide microarray expression data as well as a list of periodically regulated genes from each study are available online, along with phase and amplitude values assigned to each gene [11,12]. We considered data obtained from elutriation experiments in the three studies [4]. We analyzed the distribution of the phases and amplitudes in each periodic set. We then considered the distribution of phase differences as a more reliable comparative parameter among the three studies. After having studied the histograms, we made use of kernel density plots to remove insignificant bumps and reveal real peaks. Histograms strongly depend on the choice of the bin grid and on the starting point. Kernel density estimators are smoother than histograms and converge faster to the true density [13–15]. The choice of a proper bandwidth is still an important issue, and it should represent a compromise between smoothing enough and not smoothing too much to smear real peaks away. We computed the histograms averaging over a large number of shifts of starting points and considering very small bins with data-dependent bandwidth. Changes in the bandwidth do not affect our qualitative analysis. The three periodic datasets show differences in size. Searching for an explanation for this discrepancy, we computed the cyclic Fourier component obtained by the time series of each gene in the genome. We then compared it with the one obtained from randomly reshuffled expression data to generate a p-value for the periodicity of the corresponding gene. This indicator represents the probability that the observed oscillation occurs by chance. The smallest p-values correspond to the most cyclic genes. The amplitude of the oscillation also contributes to the p-value in such a way that genes with greater amplitude have a smaller p-value. We studied the normalized distribution P (p) of the p-values for the three studies. We defined a network represented by a complete graph (each node is connected to all other nodes in the graph) where each node corresponds to a gene whose expression was identified as periodically regulated during the cell cycle in the corresponding study. In all datasets, a gene is assigned a phase ϕi and an amplitude Ai at the expression peak. The most useful parameter for comparison is the phase difference, a measure of the expression peaks distance between genes in the cell cycle. The link between node i and node j is thus assigned a weight ωij given by the expression where ϕi is the phase of node i and β is a tuning parameter. We studied the degree distribution and clustering coefficient of the resulting network. As we are dealing with a complete, weighted graph, we considered appropriate definitions of the weighted degree (strength) and of the weighted clustering coefficient [16]. We also considered the binary network obtained by fixing a threshold t and keeping only links with ωij ≤ t. We studied the degree distribution and the correlation between the degree and the phase for each gene. We applied the Markov clustering algorithm (MCL) [17,18] to study the cluster structure of the periodic cell cycle network. Unlike most clustering algorithms, the MCL does not require the number of expected clusters to be specified beforehand (this condition can be very limiting and time-consuming when there is no specific a priori information regarding the network structure), and it can easily identify possible hierarchies of substructures. Note that the MCL has been used in approaching several bioinformatics classification problems [19–22]. The basic idea underlying the algorithm is that dense clusters correspond to regions with a larger number of paths. A random walk has a higher probability to stay inside the cluster than to leave it soon. The crucial point lies in deliberately boosting this effect by an iterative alternation of expansion and inflation steps. The algorithm iterates three steps. Given a network with n vertexes, it takes the corresponding n × n adjacency matrix A and normalizes each column to obtain a stochastic matrix M. It takes the kth power Mk of this matrix (expansion) and then the rth power mijr of every element (inflation). In the case of a weighted graph, such as the periodic cell cycle network, the probability of the random walk is proportional to the weight of the link. In our analysis, the expansion parameter k is always taken equal to 2, while the granularity of the clustering is controlled by tuning the inflation parameter r. In addition to the parameter r, we also introduced a control parameter β. This parameter allows us to speed up the process. In a first analysis with β = 1, the algorithm needed high values of r to identify the first two clusters and then went on very slowly. This behavior can be explained by the fact that our periodic cell cycle network is a complete graph. In what follows we will always consider β = 10. To study the robustness of the results given by the MCL, we considered the stability of the clustering as related to the identification of unstable nodes [23]. A node is unstable if it typically lies at the borders of different clusters, so that the algorithm has some difficulty in assigning it to either of its basins of attraction. To measure the stability of the clustering patterns, we added random noise on the weights of all links in the network and studied the clustering after many realizations. Let Pij denote the probability that the link between node i and node j connects two nodes inside the same cluster (P is equal to 1 for a link that is always kept and 0 for a link that is always cut by the algorithm). By fixing a threshold θ (typically θ = 0. 8) and eliminating all the links with Pij ≤ θ, we obtain a certain number (greater than or equal to the number of original clusters) of disconnected components. Nodes belonging to small components that cannot be identified with any of the original clusters can be defined as unstable. Figure 1 shows a very simple network with a cluster structure made of three components. Through different random noise realizations, the green node is alternatively assigned to either of its basins of attraction. The resulting probabilities Pij of the two links that connect the node to the rest of the network are ≤ 0. 8. The node is thus identified as a single component that does not correspond to any cluster, and it is defined as unstable. In the case of the periodic cell cycle network, the weights on the links are modified as ωij (1 + Δij) where Δij are Gaussian deviates with 0 mean and standard deviation 0. 5. Results do not change if we increase the noise strength. Using the probabilities Pij, we introduce the average clustering entropy per edge: The sum is over all edges, and the entropy is normalized by the total number of edges L. If the network is totally unstable (Pij = ½ for all edges), then S = 1; if the network is perfectly stable (Pij = 1 or 0 for all edges), then S = 0. In the case of the MCL algorithm (and of any other clustering algorithm defined through a parameter), we can either consider single values of the function S at fixed values of r, or we can study the landscape of the clustering entropy as a function of the clustering parameter. To compare the results given by the different studies and to analyze the structure of the cell cycle given by a more reliable core of periodic genes, we studied the intersection of the three datasets. We observed that even if a gene is identified as periodically regulated by the three studies, the assigned phase values ϕi can be substantially different. We considered a distance matrix A whose elements aij are given by the phase difference ϕi − ϕj between the expression peaks of genes i and j (a symmetric matrix with all zeros on the diagonal). Having three distance matrices, one for each dataset (only genes in the intersection of the three experiments are considered, in order to obtain three matrices of the same size), we applied the Mantel test, which computes a correlation between two n × n distance or similarity matrices. It is based on the normalized cross-product: where aij and bij are the generic elements of the two matrices A and B we want to compare, ā and b̄ are the corresponding mean values, and sa and sb the standard deviations. The null hypothesis (NH) is that the observed correlation between the two distance matrices could have been obtained by any random arrangement. The significance is evaluated via permutation procedures. The rows and columns of one of the two matrices are randomly rearranged, and the resulting correlation is compared with the observed one. We also computed a general error on the phase values as the distance between two successive points of the time series, assuming that inside the interval between them it is impossible to precisely assign the phase value. According to this error, we studied the agreement on the phase values. After this preliminary analysis on the agreement of the three datasets, we considered the network corresponding to the intersection. We studied the clustering and its stability. The computation of the strength of the nodes (the sum of the weights of all links of a node) allows us to identify genes that are located at the borders of the clusters [16]. Their strength is significantly smaller than the mean value of the network. As a preliminary comparative study of the different datasets, we analyzed the distributions of the phase and amplitude of the periodic genes in the three datasets. The amplitude distribution is very similar across the studies and well fitted by a bell-shaped distribution. It does not give further information on data. The phase distribution is more interesting (Figure 2, left). The overall behavior is universal, with two main peaks separated by low-expression regions. The first one corresponds to the transition from phase S to phase G2 and the second from phase G2 to phase M. We also introduced the phase distribution of the set of ~800 genes identified as periodic in the budding yeast S. cerevisiae [24] for comparison. It shows a single expression wave corresponding to the S and G1 phases of the cell cycle. Discrepancies can be observed in the position and extension of the two peaks across the three studies on S. pombe. Differences in the synchronization technology and phase assignment method are probably at the origin of these deviations. We thus consider the phase difference Δϕ as a more reliable parameter for comparison. The corresponding distributions are much more similar (Figure 2, right). The common minimum corresponds to the low expression regions between the two peaks in the phase distributions. Slight deviations in the head of the distribution are a consequence of normalization over datasets of different size. Differences in the tail depend on a lack of uniformity in the abundance of genes across the four phases of the cell cycle (Table 1). The number of genes identified as periodic in the three studies are quite different. Rustici et al. [6] propose 407 periodically regulated genes, while Peng et al. [7] and Oliva et al. [8] indicate bigger sets with ~750 genes each (Table 1). The correct number of periodic genes is still unknown, and whether it is better to focus on a small number or to consider all the results as equally meaningful is an open question. In Figure 3 we show the normalized distributions of p-values relative to the cyclic spectral component for all genes from the different studies (time series with two or more consecutive missing data points have been ignored). They depend on time series properties, such as the number of points and intervals that differ from one study to the other. Nevertheless, in all of the three studies, they show inverse power-law behavior. This result tells us that if we consider exclusively the information contained in the time series, there is no characteristic threshold that can be used to separate periodic from nonperiodic genes. In the experiments, the choice of the proper periodic dataset strongly depends on the false discovery rate and on the visual doublecheck of the time series. Some degree of arbitrariness remains in the choice of the cut, and it is co-responsible for the discrepancies between the three studies. The fact that Peng et al. [7] and Oliva et al. [8] presented similar datasets in size is not significant. In their study, Oliva et al. [8] concluded that there is no way to distinguish between periodic and nonperiodic genes. They ranked more than 2,000 genes according to their periodicity indicator, and they finally focused on a set of 750 to establish reasonable comparisons with the other already published datasets on the budding and fission yeasts. In Figure 4 we give a graphical representation of the periodic cell cycle network in the three studies. The relevant structure of this kind of network is given by links with higher weight. These links connect genes that are expressed (and probably regulated) at the same time on the cell cycle. Figure 4 shows the binary networks obtained by keeping only links with ωij < t (in the present case we fixed t = 18,000; lower thresholds only affect the thickness of the circular graph). They reflect the time progression of the cell cycle, with the correct sequence of phases. The length of each phase does not correspond to the real extent of the phase in the cell cycle, but rather reflects the corresponding number of periodic genes. The diameter of each node represents the amplitude assigned to the corresponding gene. We observe that in all experiments high-amplitude genes are mostly concentrated in the three shorter phases (M, G1, S). We stress that the threshold t was only introduced for graphical purposes and that all analysis were made on the complete, weighted networks. The weighted degree distribution highlights that most nodes have high strength, reflecting the completeness and overall uniformity of the network. The correlation between the strength and the phase of each node tells us that genes lying on the low expression regions that separate the two peaks of the phase distribution correspond to nodes with strength significantly smaller than the mean value of the network. Moreover, genes belonging to the M, G1, and S phases (first peak) have greater strength than those belonging to the G2 phase (second peak). The three networks are characterized by a very high clustering coefficient (C ≈ 0. 5). Their community structure appears to be robust, as they all split in a relatively small number of groups of nodes (<10) for increasing values of the granularity parameter r (Figure 5, top). The progression of the clustering is smooth. At first, all networks are separated into two large clusters (see Figure 4) that the MCL is able to identify at r ≈ 1. 25. The first one corresponds to the set of genes belonging to M, G1, and S phases of the cell cycle, while the second one collects all genes belonging to the G2 phase. Such clusters reflect the two peaks of the phase distribution (see Figure 2). In all experiments, they appear to be separated by bottleneck structures, which correspond to transitions from one phase to another (more precisely, from phase S to phase G2 and from phase G2 to phase M) and seem to be characterized by the presence of a smaller amount of periodic gene expression (in good agreement with the distribution of gene phases in Figure 2). By increasing the value of the granularity parameter r, the two main clusters are respectively split into subclusters, suggesting the presence of a hierarchical organization. To assess the significance of this scenario, we studied the stability of the cluster patterns in terms of the presence of unstable nodes and of the behavior of the clustering entropy [23]. The cluster structure of the network of periodic genes in S. pombe is highly stable. We identified no more than two or three unstable nodes, depending on the granularity of the clustering. As one might expect, these nodes correspond to genes lying at the bottleneck structures visible in Figure 4 (i. e. , genes belonging to periods of transition between different phases in the cell cycle). Furthermore, there is a correspondence between the number of clusters and the trend of the clustering entropy S as a function of the parameter r (Figure 5). The jump from a partitioning level to the following shows up as a peak in the entropy landscape. In Figure 5 (top) we see that the number of clusters sometimes remains constant for a large interval of values of the parameter. In these cases, increasing r from the first-cut value (actually corresponding to a peak in the entropy) results in a decrease of the entropy until it reaches a minimum. This minimum represents a more stable configuration of the clustering. This picture holds until the entropy reaches saturation. In each of the three studies, all genes in the genome were ranked according to a periodicity indicator. Comparing the three ranked lists, it is possible to observe that the agreement between the three datasets is much stronger between top-ranked genes, which means genes that are found to be more strongly regulated [8]. It is therefore interesting to study the minimal list given by the intersection of the three datasets. It comprises only genes that have been identified as periodic by all groups and that are thus placed on top of the three ranked lists. The intersection is given by a set of 156 genes, that is, about 10% of the entire pool. The Mantel test returns a value cAB ≈ 0. 8, showing a good correlation between the distance matrices of the three experiments. To assess the statistical significance of this result, we compared it with the NH. We obtained a p-value that scales as p (n) = e−n in which n is the number of pairwise rearrangements of the rows and columns (calculated over 105 randomizations). This means that the mere reshuffling of 20% of the matrix gives p ~ e−10. The actual value is thus significant against the NH. The distribution of the number of genes on the four phases of the cell cycle is now different (Table 1). Less then one-third of the shared genes belong to the G2 phase, with most genes belonging to the M–G1–S cluster, and, more precisely, half of them to the G1 phase. We found that even if a gene is periodic in more than one group, the corresponding phase values can be different. The best agreement between the phase values is for genes in the M–G1–S cluster. Less than 20% of the genes show an agreement of the three phase values within the error, while ~60% of the genes show good agreement at least between two values. The clustering structure of the intersection shows the same two main clusters as in the separated networks (Figure 6). Further analysis of the hierarchical substructure reveals that genes in the M and G1 phases are strongly connected and cannot be separated, while the S phase forms an independent component. Moreover, the G2 phase shows at least two separated subclusters. The resulting four clusters are separated by bottleneck structures. Two of them correspond to the phase transitions already observed in the networks of the entire datasets. The third one is less evident, and it corresponds to the transition between the G1 and S phases. This clustering pattern is very robust. No more than two or three nodes can be identified as unstable, and they are located on the G1–S bottleneck. We computed the strength of the nodes to identify genes that lay at the borders of the clusters. A list of these genes and their functions is shown in Table 2. We have carried out an extensive comparative analysis of the results from three independent groups working on cell cycle data. The power-law distribution of the p-values (Figure 3) in the three datasets shows that a large number of genes are above the 1% and 5% confidence levels. There is no apparent change in the exponent of the power law in these regions. This implies that the information contained in the time series is not enough to establish a clear division between strictly periodic and nonperiodic genes. It is now known that the cell cycle is central to a large number of subnetworks dedicated to other cell activities. Disruptions in pathways leading to DNA repair, signaling, membrane lipid and protein formation, and protein degradation may affect the survival potentialities of the daughter cells. This argument suggests that a large number of genes may be loosely involved in the cell cycle even if they are not main players of it. The numbers and identity of genes in the periodic sets thus strongly depend on false discovery rate and interpretation of the data [25]. At first, a statistical method such as Fourier analysis is applied in a rather blind way. Then, cell cycle profiles are filtered for minimal amplitude and doublechecked by eye. The visual inspection of the gene expression profiles has its advantages, but obviously a different reproducibility from that of a statistical analysis. Different methods may extract different types of information and may be difficult to compare. The network defined in Clustering and Entropy Measure provides a representation of the entire periodic cell cycle with no need to focus on the single genes or phase values. Figure 4 shows that, despite the differences in the datasets, some properties of the general structure of the network are universal across the three studies. Although G2 is the longest phase in the cell cycle, it is not the most densely populated. It spreads over ~70% of the cell cycle, but in all studies, it contains no more than a half of the identified periodic genes (see Table 1). This is not unexpected. In S. pombe, cell division occurs at the end of S phase, implying that the G2 phase represents a long period of mass increment. Most genes are expressed during the entire process and thus do not have a defined expression phase. Some genes, which are probably involved in more specific tasks, are periodically regulated. We studied the amplitude of each gene at the expression peak as a function of the corresponding phase, and we observed that genes are more strongly regulated during the M, G1, and S phases than in the G2 phase. The study of the intersection of the three datasets shows that genes and phase agreement is not so good on the G2 phase. This suggests that G2-phase genes are more likely to be false positives and probably represent an overestimation of the number of genes that are truly periodic. Moreover, in Rustici et al. [6] (dataset with 407 periodic genes) G2-phase periodic genes represent a little more than one-third of the entire pool, while in the other experiments (both showing a much bigger pool of ~750 periodic genes), the G2-phase genes correspond to about half the dataset (Table 1). The distributions of the phase differences in the three datasets are very similar, and they clearly identify two waves of periodic expression (Figure 2). They are peaked in the M–G1–S phases and the G2 phase, respectively, and they are separated by short, quiet periods in which very few periodic genes are expressed. Nevertheless, both Rustici et al. [6] and Peng et al. [7] identified four clusters of periodic genes, roughly corresponding to the four main phases of the cell cycle, while Oliva et al. [8] proposed eight different clusters. This clustering is not consistent with the distribution of periodic genes on the cell cycle. To study the topological clustering of the cell cycle, we applied the MCL to the periodic cell cycle network. In all studies, the first hierarchical level of the clustering shows two main clusters, one corresponding to the M–G1–S phases and the other to the G2 phase, separated by bottleneck structures with very few genes. This pattern reflects the behavior of the phase distribution. The study of the lower hierarchical levels reveals that genes in the M and G1 phase form a strongly connected cluster that cannot be further divided by the algorithm. On the other hand, genes in the S phase are grouped in an independent component, and the G2 phase can be partitioned into at least two clusters. This pattern is universal across the different studies, suggesting that the topology of the network contains information on the biological processes involved in the cell cycle. The stability of the clustering shows the robustness of the structure against the presence of false positives and false negatives in the datasets. The entropy landscape does not change much across the experiments. It always reaches saturation when the network splits into six or seven clusters. In Figure 5, we see that Peng et al. [7] and Rustici et al. [6] show a common stable minimum of the entropy at r ≈ 2. In both studies, this minimum corresponds to a separation into five clusters. A similar result holds for Oliva et al. [8] with a slightly different value of the clustering parameter (r ≈ 2. 2). In the network of the intersection, the entropy landscape shows a minimum that corresponds to a separation into four or five clusters. We thus suggest that a basic structure made up of four clusters (with eventually a fifth one in the G2 phase) could be the most reliable picture of the clustering pattern. Genes in the M and G1 phases form the first component, genes in the S phase form the second component, and then genes in the G2 phase form the last two components. For the sake of comparison, we also applied a different clustering algorithm to the periodic cell cycle network. In recent years, several methods have been proposed to reveal the community structure of very heterogeneous networks. Among them, only a few can successfully handle a complete and weighted graph. One possible choice is an algorithm based on modularity (M) optimization (a measure of the difference between the number of links inside a given module and the expected value for a randomized graph of the same size and degree distribution) [26]. We considered a recent method based on simulated annealing to obtain clustering by direct maximization of M [27]. The results are very similar to the more reliable picture obtained by the MCL (as described in the previous paragraph). The application to the periodic cell cycle network in Rustici et al. [6] and Oliva et al. [8] returns a structure made of four clusters: one corresponding to the M phase and part of the G1 phase, one corresponding to the end of the G1 phase and the S phase, and two modules inside the G2 phase. In the case of Peng et al. [7], a fifth cluster corresponding to the G2–M phase is identified. The bottleneck structures identified by the MCL are well respected. The main difference seems to be the partitioning of genes in the G1 phase between the two clusters corresponding, respectively, to the M phase and the S phase. To explain this behavior, we refer to a recent work on resolution limits in community detection [28]. The authors give evidence that modularity optimization may fail to identify modules smaller than a certain scale, depending on the total number of links in the network and on the number of connections between the clusters. More precisely, even a module whose size is on the order of the size of the entire network may not be resolved if it has a number of external links on the order of the number of connections inside the module itself. In the exploratory data analysis we showed that only the two main communities of the periodic cell cycle network are revealed by the phase distribution. However, the MCL is able to identify a cluster substructure. Our discussion of the results points out that the module corresponding to the S phase is the last one isolated by the algorithm, and that the bottleneck between the G1 and S phases is the less evident and more unstable one. The number of links connecting this module to the M–G1 cluster is on the order of the number of internal links. According to these arguments, the modularity algorithm would rather split the big M–G1–S cluster into two symmetric subclusters than separate the smaller S phase from the larger M–G1 component. The analysis of the stability of the clustering in the network of the intersection reveals its robustness. There are no unstable nodes on the two main bottlenecks, the G2–M and the S–G2, and only one or two unstable nodes on the border between the G1 and S phases and between the two clusters in the G2 phase. These results confirm the significance of these structures and their role in the biology of the cell cycle. The bottlenecks are strongly correlated to cell cycle checkpoints. These are cellular pathways, induced by DNA damage, that block cell cycle progression or slow the rate at which the phase proceeds. According to the cell cycle stages, there are at least three DNA damage checkpoints: G1–S (G1) checkpoint, intra–S phase checkpoint, and G2–M checkpoint. We thus investigated those genes that are located at the borders between different clusters that correspond to cell cycle checkpoints. We ranked the nodes according to their strength, starting from the one with the smallest value, and we kept those that are on top of the ranking in at least two datasets. These nodes represent genes that are located on the bottlenecks corresponding to cell cycle checkpoints. This means that they are periodically expressed during the transition from one phase to the next. A list of these genes and their functions is shown in Table 2. Most of them have important functions, and we propose them as potential new cell cycle regulators involved in the control of the transition from one phase to the next [29]. The approach described in this paper is an example of comparative analysis and can be applied to other, similar complementary datasets. Moreover, the periodic cell cycle network can be built from any gene expression dataset. The study of the clustering and the stability measure reveal the more reliable community structure of this network. The identification of nodes lying at the borders of different clusters can contribute to the isolation of genes potentially involved in cell cycle regulation. As a future development, we will consider applying this method to gene expression data on the human cell.

Because of the diversity in technological and analytical approaches, published microarray studies on a given organism show similarities as well as differences. While a great amount of data is now available, there is a general need for comprehensive methodologies that would allow us to analyze and compare all these data. We propose a general statistical bioinformatics approach based on recent developments in network theory, and we present an application to three different cell cycle–regulated genes datasets on the fission yeast. We introduce the periodic cell cycle network built upon microarray data on gene expression, and we study the properties and the stability of its community structure. We show that the periodic cell cycle network of the fission yeast is characterized by four clusters separated by bottleneck structures corresponding to cell cycle checkpoints. We identify a set of genes located on these bottlenecks, and we propose them as potential new cell cycle regulators involved in the control of the transition from one phase to the next. Our approach can be applied to other similar complementary datasets or to any gene expression datasets to reveal the community structure of the corresponding network and to isolate genes potentially involved in cell cycle regulation.

Abstract Introduction Methods Results Discussion

cell biology genetics and genomics eukaryotes computational biology

2007

Bottleneck Genes and Community Structure in the Cell Cycle Network of S. pombe

In the Cameroon, previous efforts to identify Buruli ulcer (BU) through the mobilization of community health workers (CHWs) yielded poor results. In this paper, we describe the successful creation of a BU community of practice (BUCOP) in Bankim, Cameroon composed of hospital staff, former patients, CHWs, and traditional healers. All seven stages of a well-defined formative research process were conducted during three phases of research carried out by a team of social scientists working closely with Bankim hospital staff. Phase one ethnographic research generated interventions tested in a phase two proof of concept study followed by a three- year pilot project. In phase three the pilot project was evaluated. An outcome evaluation documented a significant rise in BU detection, especially category I cases, and a shift in case referral. Trained CHW and traditional healers initially referred most suspected cases of BU to Bankim hospital. Over time, household members exposed to an innovative and culturally sensitive outreach education program referred the greatest number of suspected cases. Laboratory confirmation of suspected BU cases referred by community stakeholders was above 30%. An impact and process evaluation found that sustained collaboration between health staff, CHWs, and traditional healers had been achieved. CHWs came to play a more active role in organizing BU outreach activities, which increased their social status. Traditional healers found they gained more from collaboration than they lost from referral. Setting up lines of communication, and promoting collaboration and trust between community stakeholders and health staff is essential to the control of neglected tropical diseases. It is also essential to health system strengthening and emerging disease preparedness. The BUCOP model described in this paper holds great promise for bringing communities together to solve pressing health problems in a culturally sensitive manner. Buruli ulcer (BU) is one of several neglected tropical skin diseases that afflict the rural population of sub-Saharan Africa, especially the poor living in areas with limited access to health infrastructure [1,2]. BU stands out as one of the most disabling of all neglected tropical diseases. The large majority of cases of BU have been identified in West Africa, particularly the countries of Benin, Cameroon, the Democratic Republic of the Congo, Cote D’Ivoire, Ghana, and Nigeria [3]. BU is caused by Mycobacterium ulcerans (MU), a microorganism belonging to the same genus of bacteria as tuberculosis and leprosy. BU has a known cause and cure, but an unknown route (s) of transmission and poorly understood incubation period [4,5, 6 7]. BU manifests as necrotizing cutaneous lesions. Thirty-five percent of lesions are located on the upper limbs, 55% on the lower limbs, and 10% on other body parts [5]. In Africa, about 48% of those affected with BU are children under 15 years of age [4], and males and females are affected equally. If not treated early and in a timely manner, BU often, but not always, progresses to an advanced state requiring prolonged wound care and skin grafting. If left untreated or treated late, BU does not kill, but may render the afflicted permanently disabled. Early diagnosis and treatment are the only ways to minimize morbidity and prevent disability [8]. Notably, at present most cases (68%) of BU are diagnosed at a late stage—categories II and III of the disease [9]. Beyond non-identification in its early stages, and delay in seeking care, there is some evidence suggesting that distinct phenotypes of BU may be more likely to progress to severe forms [10]. Prior to 2005, when effective antibiotic treatment was discovered for treating early stages of BU, all cases required surgery. Antibiotic treatment with streptomycin injections and oral rifampicin for 56 days proved to be highly successful in early (category I) BU cases. Once treated with appropriate antibiotics, there was a very low rate of BU relapse [11]. Clinical trials of oral treatment for the early stages of BU using rifampicin and clarithromycin have been found promising and recently approved for use by WHO [3]. The key to community management of BU is identifying cases in the early stages of the disease and enrolling the afflicted in treatment programs with minimal delay and sustained adherence to treatment. In this paper, we describe a pilot project conducted in Bankim District, Cameroon that proved to be highly successful in establishing a BU community of practice (BUCOP) (Fig 1). A community of practice (COP) is an assemblage of stakeholders committed to a common objective, a common basic understanding of a focal problem, and mutual respect for what each stakeholder contributes to a process of problem solving [12,13]. In the case of BU, this entails health staff, community health workers (CHWs), and traditional healers sharing a common understanding of the signs of BU, collaboration in encouraging the afflicted to seek and continue BU treatment, open lines of communication between stakeholders, and mutual respect for what each contributes to a process of healing that includes, but extends beyond the management of BU as a disease. BU has been identified in 64 of Cameroon’s 179 districts. Bankim District, the focus of this paper, is located in the northwest Adamawa region of Cameroon bordering Nigeria. As noted in Fig 2, the district has one of the three highest prevalence rates for BU in the country [14]. The central treatment and referral hospital for BU in Cameroon is the Ayos District Hospital located in the southern part of the country. Until recently, BU patients from all over Cameroon had to travel considerable distances to be treated at Ayos, a journey many were reluctant to make. Bankim is located over 475 KM away from Ayos and over 10 hours by local transport. Over the last decade, Cameroon’s National BU Control Program has trained clinic staff to provide treatment for BU in its early stages in many regions of the country and established five diagnostic and advanced treatment centers. Another two BU treatment centers are soon to be functional [14]. Bankim is located in a remote region of Cameroon with environmental factors favoring the presence of the MU microorganism responsible for BU. The district is situated in the Mape River Valley, where a dam was built to generate hydro-electric power more than 25 years ago. The Mape Dam splits the area into isolated islands and scattered villages. In the last two decades, increased irrigation has enabled rice cultivation. Rice farming has been identified as a possible risk factor for BU transmission [15,16]. Inhabitants of the region also engage in the growing of maize, cassava, and peanuts as well as various forms of hunting and fishing. Much agriculture is done on plots of land some distance from villages during the months of January to May, and many inhabitants seek employment in Nigeria from November to April. Population movement is both fluid and seasonal. Health services in Bankim district include a district hospital and 5 satellite clinics. The hospital, at the onset of the project, was staffed by one doctor, 2 nurses, 4 nurse assistants, and one lab technician. The 5 satellite clinics were staffed by one nurse and 1–3 assistants. Two of the 5 satellite clinics had a lab technician conducting basic laboratory analysis. Each large village in Bankim has one community health worker/ volunteer (CHW) who assists hospital staff with outreach activities when requested to do so. There are 86 primary schools in the district and 10 secondary schools. School attendance waxes and wanes depending on season and agricultural activities. Bankim Health District is a challenging place to initiate a community outreach program for BU due to both its rugged terrain and the wide variety of ethnic groups inhabiting and moving in and out of the region. These groups speak a variety of languages and dialects in addition to French and Pidgin English. The region is inhabited by Tikars, Yambas, Mambilas, Kwanjas, Fulanis as well as ethnic groups hailing from the neighboring Western, North West, Central, and Adamawa regions of Cameroon, and Nigeria. This required BU outreach activities to be carried out in multiple languages and for the research team to seek the approval and support of Christian and Muslim clerics, influential traditional healers, and local chiefs. There are seven paramount chiefs responsible for the welfare of Bankim district. The project staff had to gain the permission from each paramount chief before they could initiate outreach activates in their domain. A proof of concept study was first initiated in the jurisdiction of the paramount chief of Bankim town, a forward thinking, but cautious leader. Once the project was deemed feasible, other chiefs agreed to sanction community based activities conducted during the pilot phase of the project. This community-based intervention employed a seven-stage formative research process [17] summarized in (Table 1) and adapted for BU. The formative research process covers all aspects of an intervention from the collection of baseline data and problem recognition to the generation and weighing of possible interventions from the vantage point of different stakeholders to project implementation, monitoring and evaluation. The three phases of the project are summarized in Table 2. Phase one focused on baseline data collection, problem identification, and the generation and assessment of intervention options. Phase two had two parts. Part one entailed a proof of concept study to test the feasibility of promising interventions on a small scale. Part two applied lessons learned in the proof of concept study to a large pilot study. In phase three, outcome, process and impact evaluations of the pilot intervention were carried out. In the first phase of the project, a team of three Cameroonian anthropology graduate students and their research advisers conducted interviews with current (N = 69) and former BU patients (N = 22) as well as health staff (N = 15), traditional healers (13), and CHWs (19). These interviews probed predisposing, enabling, and health service related factors influencing health care seeking for chronic ulcers, and reasons for treatment delay and drop out among BU patients offered free treatment. The team identified barriers to treatment adherence related to: 1) cultural perceptions of why wounds do not heal in a timely manner; 2) fear of hospital treatment and trust in hospital staff, and 3) pragmatic issues such as poor transportation, housing, and the availability of food for patients and caretakers when treatment requires hospitalization. The three anthropologists were then embedded in separate communities to investigate the current role of community health workers and healers in chronic wound management, existing BU detection activities, patterns of treatment referral, and household wound care decision making in different seasons. Health staff interaction with BU patients, CHWs, and healers were also observed in the community, at the district hospital, and in local clinics. Research methods employed included participant observation, key informant interviews, prospective and retrospective case studies of BU patients, semi-structured interviews incorporating “what if” scenario, and observations of social interactions between health staff and community members. At the end of four months of intensive ethnographic research, the anthropological team presented their findings at a workshop attended by the doctor in charge of Bankim district hospital, the heads of the Cameroon National BU Program, and representatives of the NGO Fairmed providing support for neglected tropical diseases (NTD) programs in the region. In keeping with stages 2–4 of formative research, the team also presented data on what different stakeholders saw as possible means of more proactively involving community members in BU detection and the kinds of support that might reduce treatment delay and drop out. Two major concerns were raised at the workshop. The first concern entailed the need to pay respect to local culture while at the same time addressing cultural beliefs and practices that posed barriers to BU detection and treatment. This required gaining the trust of local leaders and traditional healers and enlisting their support in a new BU outreach initiative. The key question posed was: Would it be possible to involve traditional healers in community-based BU outreach such that they become part of the solution, rather than a major part of the problem of BU treatment refusal, delay, and drop out? The second concern raised was pragmatic. What could be done to reduce the difficulties faced by impoverished patients faced with having to travel long distances to clinics for daily outpatient treatment or required to remain at Bankim hospital as an inpatient for months? The question posed was: would providing transport to the clinic, and food and lodging when necessary, increase the local population’s willingness to seek BU treatment early and adhere to treatment guidelines? Phase two of research entailed a small- scale proof of concept (POC) study testing the feasibility of a package of proposed interventions to enhance BU outreach and establish a BU community of practice. The twin objectives of the intervention were to raise consciousness about BU using mass outreach events, and to use these events as an opportunity to establish collaborative relationships between clinic staff, chiefs, CHW, traditional healers, and recent BU patients who had a positive treatment experience. A second part of the POC was testing different types of patient support. The POC study produced positive outcomes (reviewed in the Results section) warranting a larger scale pilot study in Bankim district. In the second part of phase two, a three year pilot study was launched after four modifications were made. The first modification entailed the use of a new WHO promotional video for BU in outreach programs prior to presentation of the educational PowerPoint. Videos are very popular in rural Cameroon. Although the new video was not designed for educational purposes, the theme of hope portrayed was in line with the BUCOP outreach program. Research revealed that while community members did not comprehend much of the language contained in the video, they were happy to see images of patients who had recovered from BU after treatment. The messages in the video did not duplicate nor clash with the messages presented in the educational power point presentation. A second modification was using CHWs to engage in simultaneous translation into local languages. Hospital staff would present PowerPoint slides in French while a community health worker would translate the messages into local languages (Fulani, Pidgin English, Kwanja, Mambila, Yamba, and Tikar). Teams of presenters adopted a familiar presentation style commonly used in Pentecostal churches with messages passed back and forth between languages in a free and easy style. Pilot research revealed that although repetitive for speakers of multiple languages, translation was responded to positively by audiences even though it increased the duration of the program. A third modification involved expansion of the number of traditional healers involved in the BUCOP. The healers participating in the POC were carefully screened as exemplars to model best practices. In the pilot study, three other healer groups located in the district were invited to participate in the BUCOP. Members of these healer groups were trained using the BU PowerPoint presentation as a common reference point for instruction. Healer groups met every month to discuss cases and each group monitored members’ adherence to a BUCOP contract. Traditional healers were also invited to participate in community outreach activities in their locales, often working along with CHWs. As in the POC study, they were paid a small honorarium for their efforts. A fourth modification entailed an upgrade of the halfway houses and hospital wards. During the POC, concern was expressed about unhygienic conditions in the temporary halfway houses. Two halfway houses with bore wells were constructed in rural areas and visited by health staff traveling by motorcycle. At Bankim hospital, both a BU ward and a well laid out dressing room were constructed. A second surgeon was stationed at the hospital such that most skin grafts could be carried out in Bankim rather than being referred to Ayos hospital. As noted in Table 2, the role of the team of anthropologists shifted over the course of the pilot project. During the POC and the first year of the pilot study, the anthropologists played an active role assisting in the implementation of the intervention package and monitoring interventions, enabling mid-course correction. Anthropologists acted as change agents and were consulted by clinic staff when problems arose. Gradually, implementation of all intervention activities was turned over to hospital staff, CHWs, and leaders of traditional healer groups. By the second year of the pilot study, anthropologists assumed the role of participant observers, monitoring outreach and referral activities, and documenting cases of successful partnerships as well as the ways in which members of the BUCOP solved problems. Then, in the third year of the pilot, the social scientists left Bankim for 12 months to see if BUCOP activities would be sustained without their presence as cultural brokers. Phase three of the project took place one year after the social science team left Bankim. The team returned and conducted outcome, process, and impact evaluations to assess the effectiveness of the intervention in terms of BU detection, treatment adherence, and BUCOP stakeholder collaboration without the presence of social scientists as change agents. Attention was focused on whether COP member partnerships and lines of communication were sustained. During this evaluation phase of the project hospital records were reviewed and 44 interviews and twelve focus groups were carried out with 22 CHWs, eighteen healers, nine health staff, nine former and eight recent patients, and five government administrators responsible for health activities in the district. The three anthropologists who had carried out stage one formative research investigated shifts that had occurred in community stake holder relationships with clinic staff as well as each other, task sharing, and changes in the social status of health staff, CHWs, and traditional healers. Broad impacts of the project were assessed as well. Chief among these was whether collaborative relations established by the BUCOP were being leveraged and extended to other health initiatives, and whether the BUCOP model constituted a viable means of promoting trust between health center staff and community stakeholders. Ethical clearance and research authorization for the project was secured from the National Ethics Committee of the Cameroon Ministry of Health. The District Medical and Sub-Divisional Officers granted authorization for local entrance into the region and community leaders beginning with the Paramount Chief of the Tikars approved of the project. Informed consent was obtained orally from all adult participants in the project after being assured that their participation was voluntary, that the information they shared was confidential, and that they had the right to decline to be interviewed at any point during the project. Oral consent was necessitated given both low rates of literacy and the need to communicate details about the project in local languages, some of which are only spoken. Four sets of observations made during phase one formative research may be briefly highlighted: 1) local perceptions of chronic ulcers encompassing BU, 2) health care seeking patterns for chronic ulcers, 3) health staff, CHW and healer interactions; and 4) problems in BU identification and treatment warranting intervention. Chronic ulcers that do not heal, a hallmark of BU, are often but not always attributed to the local disease category Mbouati (Atom in other parts of Cameroon) [18,19,20] a spirit affliction that is also the sign of special power accorded to the afflicted. Informants across all ethnic groups voiced the opinion that both Mbouati and BU co-existed in the region, some believing that only healers could determine the difference between the two and transform Mbouti into a chronic physical ulcer (nbong: a widely used Tika term) amenable to successful treatment. A common perception was that dual illness causality could be responsible for ulcers that do not heal. Such ulcers could either be caused or complicated by a combination of natural and supernatural factors [20]. There was also widespread speculation that in recent years the type of ulcer health staff call BU had increased in the region following the dam project and the introduction of rice cultivation. Traditional healers are very popular in Bankim and often turned to as a first source of treatment for skin lesions, especially if they do not heal and are linked to witchcraft or Mbouati. Healers commonly use herbs, incantations, talismans, and “vaccination” (cutting, burning wounds, etc.) to treat chronic ulcers. While healers work independently, many are members of healer groups, which answer to the paramount chief of Bankim, recognized to be the chief of healers in the district. Healers have close ties to village chiefs and are called upon to offer blessings and protection to the community. The hospital was not a place that villagers commonly readily turned to for treatment of BU for three reasons. Community members were afraid of hospital based BU treatment as it was associated with operations and amputation. Second, they had little interest in being referred to Ayos hospital as it was far off and in a place foreign to them. Third, despite BU treatment being free, people feared the indirect costs of treatment and hospitalization. There was also some confusion about what kind of wounds were being treated free. While medicines for BU were supplied free of cost by the Cameroonian NTD program, other chronic ulcers that look like BU are not treated free. Healers did not have close working relations with clinic staff and did not refer cases to them. In the five years prior to the research project, not one case of BU had been referred to Bankim Hospital or any satellite clinic by a traditional healer. When healers visited Bankim hospital at the request of patients, they did so secretly. Healers also had little contact with CHWs. CHWs attended meetings at the district hospital when called to do so, but their role was passive. They were given standard WHO educational materials depicting the signs of BU, but were not involved in actively educating community members about BU. For the most part, CHWs only identified possible BU cases when the disease control officer (DCO) from Bankim hospital personally visited their village by motorbike and directly asked CHWs to be shown villagers with chronic wounds. In the five years prior to the research project, CHWs identified only 48 potential BU cases and all of these cases were in advanced stages. BU cases were detected by chance by health staff during vaccination campaigns, and by the DCO when assisting a foreign research team in identifying cases for a clinical trial testing thermotherapy as a possible means of treatment [21]. CHWs did not see themselves as having a clear role in organizing BU outreach activities and they were not in close communication with clinic staff. Health staff saw BU detection and treatment as the responsibility of the DCO and surgeon who headed Bankim hospital. Four types of interventions were called for on the basis of formative research. First, a new outreach education program (described below) was needed to raise community awareness about BU and its treatment, address rumors undermining confidence in clinic based care, and foster hope as a means of diffusing fear about BU treatment. Second, CHWs and traditional healers needed to be mobilized, and given a proactive role in both BU case detection and referral as well as patient follow up and psychosocial support. Third, visiting a clinic for 56 days of treatment was challenging for members of many households due to travel difficulties and indirect costs. It was clear that education alone was not going to solve the problem of treatment delay and drop out. Transport, and when necessary, lodging and the feeding of patients and their caretakers was required. Fourth, more advanced BU patients unwilling to travel to a referral hospital like Ayos needed to be treated at Bankim Hospital. Upgrades in BU care needed to be made at the hospital and outreach programs needed to inform the local population that high quality treatment for BU was now available locally at the Bankim district hospital. An intervention package addressing these four intervention priorities was developed after options were weighed in keeping with stage four formative research. The first component of the package, seen as the cornerstone for building a BUCOP, was the introduction of a culturally sensitive community-based BU outreach program requiring clinic staff and community stakeholders to work closely together. An innovative education program was already in the process of being developed by teams of West African social scientists participating in the Stop Buruli Consortium, including the team from Cameroon. The education program developed and tailored for each consortium country (Benin, Cameroon, Ghana) is the subject of a forthcoming publication. In brief, it took the form of an image-rich PowerPoint presentation on BU delivered by local teams equipped with portable generators, laptop computers, LCD projectors and sound systems. The outreach program adopted a question–answer format enabling new issues to be added as they arose. Outreach meetings were interactive, not passive, and questions were invited from community members in attendance. As such, the educational presentation was the product of an iterative process. The social scientists investigated how best to respond to questions in a way that was at once scientific and understandable to local audiences. Messages and visuals were tested and changed as needed. Table 3 briefly summarizes the major sections of the PowerPoint presentation (available at https: //www. fairmed. cm/defis/maladies-tropicales-negligees). Different messages were designed to inform and educate the community about BU, reassure community members about the quality of care available at clinics, offer hope of a cure, or display stakeholder collaboration. Over the course of the POC the education program was developed, and pretested. Community outreach education meetings were held in the evenings in eight communities. CHWs were responsible for organizing meetings and inviting chiefs, local healers, and former patients to attend. Programs were treated as social performances where roles in the BUCOP were enacted. Chiefs and healers were seated in places conveying respect, and they were invited to voice their support for the program using a microphone, itself a symbol of power. Social scientists trained the DCO to deliver the PowerPoint education program during the POC and were on hand to assist in responding to questions from the community. After the program was complete, hospital staff were then on hand to screen community members for wounds they suspected might be BU. During the POC study, 21 suspected cases of BU were identified by health staff either during or a few days following outreach meetings, (see Table 4 below). Patient’s samples were collected either using wound swabs or fine needle aspiration (for oedematous lesions). Of these, 19 (90. 5%) were confirmed to be BU by Ziehl-Neelsen staining and/or polymerase chain reaction tests PCR), of which 10 (53%) were category I–early category II BU cases. In villages too remote to reach with audio and visual equipment, CHWs delivered key messages from the PowerPoint presentation orally, using posters and other visual aids depicting the signs of BU, and holding interactive question–answer sessions. Of cases referred to health staff by CHWs, 40% were deemed unlikely to be BU based on visual inspection. Of the remaining cases referred to health staff, 44 (62%) were confirmed by laboratory test to be BU and treated. Nineteen of these cases were either category I or early category II BU. Following POC outreach educational activities, many cases of chronic ulcers and skin lesions were brought to the attention of CHWs or health staff by community members themselves. Ninety-eight of these cases were suspected to be BU by health staff, of which 45. 9% were confirmed a by laboratory test and treated as BU. Of these cases, 25 were category I or early category II BU. As a point of comparison, between 2009 and 2010 no cases of BU like symptoms had been self-referred to clinic staff by community members. During the POC study, former BU patients were also encouraged to refer suspected cases of BU. They referred 17 cases to health staff, of which 11 (64. 7%) were confirmed to be BU and treated. To gain a sense of just how effective the POC was in detecting cases of BU, it is useful to compare POC results with the results of a house to house NTD survey conducted in Bankim district between late March and mid-April 2010. During the NTD survey researchers visited 9,344 households (48,962 people). The survey only identified 25 suspected cases of BU, of which six were confirmed to be BU by PCR [4]. A second component of the POC study was provision of free transport to clinics for category I and early category II BU patients not requiring hospitalization and living 3–7 KM from clinics. Motorcycle taxis were hired to deliver BU patients to the hospital for the duration of treatment. For those who lived too far to make this feasible, housing was secured for them near clinics in facilities termed halfway houses. Halfway houses were tested in both Bankim town and two rural areas. In Bankim town, patients staying at halfway houses received treatment and hospital staff routinely monitored their wounds. In the rural areas, local clinic staff traveled 8–10 km daily by motorcycle to administer treatment to patients at halfway houses. Patients and caretakers remaining at halfway houses were supplied a food ration. Seventeen patients took up residence in halfway houses during the POC. Anthropologists monitored social interaction between patients of different ethnic groups and found that they bonded around the shared experience of BU and supported one another during treatment. The third component of the POC was assessing whether traditional healers would participate in BU outreach activities and become proactive members of a BUCOP. In brief, after obtaining support from the paramount chief of Bankim town, a meeting was held with a group of popular traditional healers who were members of a healer association. Ten traditional healers were selected to participate based on their willingness to collaborate with clinic staff in addressing both biomedical and traditional medical aspects of BU treatment. Clinic staff explained to healers that in order for their treatment to be effective, patients needed to be referred to them quickly and wounds needed to be left undisturbed and not treated with traditional medicines. Clinic staff acknowledged that they had no expertise in the mystical aspects of treatment such as removing spiritual affliction, protecting the patient when vulnerable to forces of malevolence, nullifying obstructions to the healing process, or dealing with patient fears of malevolent spirits during treatment. Healers were asked to attend to the spiritual and psychosocial aspects of treatment, while clinic staff used medications and bandaging to take care of the physical manifestations of the disease. A contract was proposed and signed by members of this group of traditional healers. Two key components of the contract were that traditional healers promised to refer all patients with possible signs of BU to Bankim clinic staff within ten days of seeing them and not treat the skin of these patients. Clinic staff gave healers free access to Bankim hospital, satellite clinics, and halfway houses to offer patients psychosocial support and spiritual protection. Healers were also offered a small amount of money to pay for accompanying patients to a clinic for screening. The social science team monitored traditional healers’ adherence to the contract. All ten traditional healers adhered to the contract, and six of the ten healers referred 49 suspected BU patients to Bankim hospital. Of the 49 suspected cases referred, 13 (26. 5%) were confirmed to be BU by laboratory test. In the case of traditional healers, all cases referred were tested, even when staff thought the case was unlikely to be BU. This policy was followed after a healer challenged a case clinic staff dismissed as not being BU based on visual inspection. The healer was proven correct in her detection of BU. The treatment adherence rate during the POC for BU patients detected was 94% compared to a rate of 54% in Bankim Hospital in 2009–2010. Increased adherence was due to both enhanced patient support by health staff, CHWs, and traditional healers and the provision of resources better enabling patients to remain in treatment. Notably, all BU patients referred by healers and visited by them in the hospital completed treatment. During the pilot project an additional 89 CHWs and 55 healers were trained in BU detection. Forty-four well attended community outreach programs were conducted reaching approximately 15,500 people. In Tables 4 and 5 we present data on BU case referral and confirmation by stakeholders during both the POC and pilot interventions. The number of cases referred designates cases referred to health staff and deemed to be possible cases of BU by visual inspection. Confirmed cases were by Ziehl-Neelsen staining and/or polymerase chain reaction tests (PCRs). During the POC approximately 40% of total cases brought to the attention of health staff were dismissed as some other kind of skin lesion, abscess, or ulcer. During the pilot study this percentage dropped to 30% of all cases reported. Table 6 presents data on the category of confirmed BU cases referred by different stakeholders in the BUCOP. Table 7 presents data on adherence to treatment by confirmed cases of BU that initiated therapy. This data is present by the type of stakeholder who referred the case. Several findings stand out as notable. First, prior to the project no cases of BU were referred to health staff by community stakeholders with the exception of cases identified when a disease control officer occasionally visited a community searching for NTD cases. During the project, 90% of all confirmed cases of BU were identified by community stakeholders. Second, as a result of the project not only was there a significant rise in the number of BU cases referred by community stakeholders, but a significant number of category I BU cases were detected and treated. Five hundred and twenty-two suspected cases of BU that health staff thought warranted laboratory confirmation were referred by community members. Out of these 522 cases, 266 cases (51%) were confirmed to be BU. Twenty-one percent of these cases were category I. Third, more suspected BU cases were referred by family members following outreach educational programs than any other BUCOP stakeholder. Their referrals accounted for 40% of all referrals of suspected cases of BU during the POC, and 46% of all referrals during the pilot study. In many instances, family members checked with CHWs and traditional healers participating in the BUCOP to ask their opinion about the lesion or to request that they accompany them to a clinic. In the same period, 23% of suspected cases were detected and referred by CHWs, of which 17% were confirmed of which 19%% were category I. Health staff only detected 39 cases of BU during their routine activities, of which only five cases (13%) were category I. A fourth finding was that traditional healers actively participated in both the POC and pilot project and kept to the terms of the contract they signed. Traditional healers referred 19% of all suspected cases of BU during the pilot project, 15% of all confirmed cases, and 18% of confirmed category I cases. Fifth, the confirmation rate of suspected cases sent for testing was impressive. While 30% to 40% of lesions brought to health staff for inspection were dismissed as other skin aliments, a high percentage of the remaining 60–70% of cases were found to be BU. Of cases tested, 43% of those referred by CHWs, 32% by traditional healers, 52% by former patients, and 52% by family members were found to be positive. The rate of confirmation for traditional healers was lower because unlike the other stakeholders every case referred was sent for testing. If 60%–70% of their cases had been sent for testing after staff screening, their confirmation rate would have been similar to other BUCOP stakeholders. Sixth, treatment adherence rates for confirmed cases referred by all stakeholders were > 90%. One can compare this to a 54% BU treatment adherence rate two years prior to the project. The goal of the pilot project was not just to mobilize individual stakeholders to become more actively involved in BU detection and referral, but to create an interactive and supportive BUCOP. Phase three evaluation research revealed that a functioning BUCOP had indeed emerged in Bankim. Lines of communication between CHWs, traditional healers, and health staff were well established and collaboration in BU outreach activities were ongoing. The status of CHWs increased markedly as a result of CHWs becoming actively involved in arranging outreach education and screening programs for their communities. A closer working relationship with clinic staff empowered them to play a more proactive role in both referring and following up suspected BU cases. The status of former BU patients also changed. The community appreciated their testimonials during large outreach meetings. Former patients came to be seen less as victims and more as survivors having BU treatment experience they were willing to share. Clinic staff noted being surprised by the large number of BU cases referred by community stakeholders as a result of the project. They came to value the BUCOP, stating that it both enabled them to have a much closer working relationship with CHWs and traditional healers, and it enhanced the reputation of the hospital. One testament to growing trust in the hospital was consultation by members of the Fulani ethnic group. Members of this group had previously been reluctant to seek treatment for BU. Outreach education delivered in pidgin by a trained Fulani community health worker (also a traditional healer) and a few treatment success stories have paved the way for greater contact with this community. Additionally, the director of Bankim hospital has been praised by Cameroonian health officials for forging close working relationships with community stakeholders. Better working relations between health staff and CHW, and increased acknowledgment of each stakeholder’s contribution to BU program success served as important non–financial incentives enhancing stakeholder motivation [22]. A major question posed during phase one of the project was whether traditional healers would be willing to actively participate in a BUCOP. The impact evaluation found that healers are presently seen by clinic staff and health officials as having an important role in community-based BU management. Not only have traditional healers referred cases and honored their treatment contract, but they have routinely assisted CHWs during outreach activities, offered psychosocial support to patients who are hospitalized, and encouraged patients who dropped out of therapy to return to treatment. Health staff are now invited to traditional healer group meetings to discuss cases, and healers are invited to hospital meetings when outreach activities are being planned. Health officials were initially reluctant to give traditional healers identification badges designating them as members of the BUCOP. By the second year of the pilot project, officials felt that healers had proven their commitment to the BUCOP. Traditional healers were offered badges along with referral cards and health officials left it up to traditional healer groups to both monitor member adherence to the contract and to sponsor new healers who wished to join the BUCOP after receiving training. A key issue investigated was what healers gained and lost from participation in the BUCOP. Some members of the NTD community initially expressed the opinion that traditional healer participation might be primarily motivated by funds received when referring patients. This opinion proved to be a false. Research revealed that healers lost far more financially then they gained by referring patients for treatment. At the time of the impact evaluation, funds for healers to accompany patients to the clinic were exhausted. Yet healers continued to refer as well as visit patients at the clinic without charging any fee in cash or in kind. During healer groups, cases were discussed and when one healer did not have the funds to take a patient to the clinic or visit them, another member of the group often offered to do so. So what did healers get out of the becoming members of the BUCOP? Research revealed that the social and symbolic capital healers gained outweighed any financial loss they incurred for referring patients. Healers who visited Bankim hospital were warmly received by health staff who saw them as an asset in reassuring patients about their treatment. Healers took pride in offering spiritual protection to patients so that BU medications could act effectively, and for offering patients psychosocial and spiritual support while under treatment. When BU patients were cured, healers shared credit with health staff for treatment success. While healers received no payment for their BU related activities, they did receive gifts when patients were cured. Moreover, other patients residing at the hospital requested their assistance. In short, their reputation increased and was not diminished by collaboration with clinic staff. An additional impact of the BU project was its contribution to the integration and control of other neglected tropical skin diseases. Although not intended to do so, BU outreach programs attracted a large number of cases of yaws, a disease not seen at clinics in the district for over a decade and assumed to be eradicated in this region of the Cameroon. Cases of yaws identified during BU outreach programs alerted health staff of the presence of the disease and they then conducted follow up school based screenings in these communities. Eight hundred and fifteen cases of confirmed cases of yaws were successfully treated [23]. We do not claim that the relationships established in the Bankim BUCOP can be formed in all contexts. Context must be taken into consideration. For example, traditional healers in Bankim are organized into groups which exercise some modicum of authority over their members. Such is not the case in all African contexts. Mobilizing individual healers would prove more challenging than mobilizing groups. A challenge that may be briefly mentioned is that when the reputation of a health facility rises so do patient load and resource need. In Bankim, the increased reputation of the hospital has drawn patients from outside the district and even neighboring Nigeria. Resources to maintain quality of care will need to be sustained or the reputation of the hospital will decline. In a recent review of the impact of two decades of health social science research on NTDs, Bardosh [24] notes that while this research has generated important insights into health care seeking and community response to disease centered programs, it has not been effectively used in program development and implementation. This pioneering study illustrates how a multistage formative research process can contribute to the creation of a BUCOP. We would argue that the COP model of clinic staff–community stakeholder collaboration presented here has great potential for other community-based disease outreach programs in Africa and beyond. It’s inclusion of community stakeholders like healers, volunteers, and former patients extends the COP model recently advocated for establishing productive relationships between local experts, NGOs, ministries of health and government health staff, and foreign advisers [25,26]. In Bankim, now that collaborative relationships have been established, they are being leveraged for other public health endeavors such as vaccination programs. Establishing a COP also has great potential for emerging disease preparedness. Should a disease like Ebola strike Bankim, collaborative relationships between clinic staff, CHWs, chiefs, and healers will enable a rapid coordinated response. If the 2015 Ebola outbreak in West Africa taught us anything, it is recognition that close ties to community stakeholders constitute an essential part of health system strengthening [27,28,29,30,31,32]. Establishing trust and lines of communication with community leaders enables swift action and increased opportunities for local problem solving.

Buruli ulcer (BU) is a neglected tropical disease primarily found in West Africa largely effecting the rural poor. BU has a known cause and cure, but an unknown route of transmission and a poorly understood incubation period. If not treated early and in a timely manner, BU often progresses to an advanced state requiring surgery and prolonged wound care. In the Cameroon, previous efforts to mobilize community health workers and educate community members to identify cases of BU yielded poor results. In this paper, we describe steps undertaken to create a successful BU community of practice (BUCOP) composed of community stakeholders working in concert with clinic staff. The success of the BUCOP was measured in terms of numbers of suspected BU cases referred and confirmed, a decline in treatment drop out, and sustained collaboration among stakeholders both during and following the pilot project. Pilot project success is attributed to an innovative and culturally sensitive approach to BU outreach education, increased levels of patient assistance, and mutual respect among BUCOP members for what each stakeholder contributed to BU detection, treatment, psychosocial support, and spiritual protection.

Abstract Introduction Methods Results and discussion

medicine and health sciences pathology and laboratory medicine health services research sociology tropical diseases geographical locations social sciences health care bacterial diseases research design scientists signs and symptoms ulcers neglected tropical diseases africa science and technology workforce research and analysis methods infectious diseases buruli ulcer cameroon research monitoring pilot studies research assessment people and places professions diagnostic medicine science policy careers in research social research population groupings

2018

Developing a Buruli ulcer community of practice in Bankim, Cameroon: A model for Buruli ulcer outreach in Africa

Polyamines are known to play important roles in the proliferation and differentiation of many types of cells. Although considerable amounts of polyamines are synthesized and stored in the testes, their roles remain unknown. Ornithine decarboxylase antizymes (OAZs) control the intracellular concentration of polyamines in a feedback manner. OAZ1 and OAZ2 are expressed ubiquitously, whereas OAZ-t/OAZ3 is expressed specifically in germline cells during spermiogenesis. OAZ-t reportedly binds to ornithine decarboxylase (ODC) and inactivates ODC activity. In a prior study, polyamines were capable of inducing a frameshift at the frameshift sequence of OAZ-t mRNA, resulting in the translation of OAZ-t. To investigate the physiological role of OAZ-t, we generated OAZ-t–disrupted mutant mice. Homozygous OAZ-t mutant males were infertile, although the polyamine concentrations of epididymides and testes were normal in these mice, and females were fertile. Sperm were successfully recovered from the epididymides of the mutant mice, but the heads and tails of the sperm cells were easily separated in culture medium during incubation. Results indicated that OAZ-t is essential for the formation of a rigid junction between the head and tail during spermatogenesis. The detached tails and heads were alive, and most of the headless tails showed straight forward movement. Although the tailless sperm failed to acrosome-react, the heads were capable of fertilizing eggs via intracytoplasmic sperm injection. OAZ-t likely plays a key role in haploid germ cell differentiation via the local concentration of polyamines. As many as 15% of human couples [1] are infertile, and male infertility is associated with about half of these cases. A decrease in sperm production has recently been reported [1]. Although advances in medical technology have allowed some infertile couples to have children, more than half of all infertility is idiopathic [1]. Because unresolved environmental problems such as global pollution might be causing endocrine disruption, a thorough understanding of the basic mechanisms of germ cell differentiation is critical for development of infertility treatments. To elucidate the molecular mechanisms of spermiogenesis, we isolated many cDNA clones specifically expressed in haploid germ cells using a subtracted haploid germ cell-specific cDNA library [2]. One of them (TISP15) encoded the Ornithine decarboxylase antizyme (OAZ) known to control the intracellular concentration of polyamines [3], [4]. Full-length TISP15, also known as OAZ in testis (OAZ-t/OAZ3), was specifically expressed in haploid germ cells [4], [5]. Polyamines, such as putrescine, spermidine, and spermine, are essential for cell proliferation and differentiation via binding to nucleic acids as cations [6], [7]. The actual function of polyamines is not entirely clear although significant amounts of polyamines are synthesized and stored in the testes [3], [8]. The biosynthesis of polyamines is regulated strictly by many proteins via the key enzyme of ornithine decarboxylase (ODC). OAZ is a major regulator of ODC [9]. Upon stimulation with polyamines, OAZ protein is translated by programmed +1 frameshifting to inhibit ODC activity specifically, and the OAZ–ODC complex drives the rapid degradation of ODC by the 26S proteasome [10]–[15]. OAZ belongs to a conserved gene family with at least three members in the vertebrate lineage. OAZ1 and OAZ2 are expressed ubiquitously in all somatic tissues [9], [16], [17]. In male germ cell, the RNA expression of somatic OAZ1 was decreased during the later stages of haploid germ cell differentiation [4]. Further analysis of OAZ-t revealed that polyamines are capable of inducing a frameshift at the frameshift sequence in OAZ-t mRNA [4], resulting in the translation of OAZ-t, as is the case for somatic OAZ1 [10]. The transfection of OAZ-t cDNA inhibits ODC activity in HEK293 cells [11]. OAZ-t may play important roles in the regulation of polyamine concentration in spermiogensis. To clarify the roles of OAZ-t specifically expressed in haploid germ cells, we produced the OAZ-t-disrupted mice and analyzed the effect of the disappearance of OAZ-t. A targeting vector was constructed (Figure 1A) and homologous recombination was used to generate embryonic stem (ES) cell clones that were heterozygous for the OAZ-t mutation. To produce chimeric mice, transgenic ES cells were injected into blastocysts that were subsequently implanted into pseudopregnant mice. Correct recombination was confirmed by Southern blotting (Figure 1B) and PCR (Figure 1C). No OAZ-t expression was detected in the testes of the homozygous null OAZ-t mutant mice by northern (Figure 1D) or western blotting (Figure 1E). Crossing of heterozygous mutant pairs produced the expected numbers of wild-type, heterozygous, and homozygous offspring, according to classical Mendelian inheritance patterns. Matings between homozygous OAZ-t knockout males and wild-type females did not result in any successful pregnancies over a period of more than three months of continuous cohabitation, although vaginal plugs were observed in the paired wild-type females (Table 1). All heterozygous OAZ-t males and homozygous females were fertile (Table 1). Neither the homozygous null mutant nor the heterozygous males exhibited a significant differentiation in body mass (Table S1). Female body mass and the weights of various organs, including the testes and seminal vesicles in the adult OAZ-t homozygous mutant mice, were identical to those in the heterozygous mice (Table S1). The serum testosterone levels and polyamine contents in the adult OAZ-t homozygous mutant male mice were identical to those in the wild-type mice (Table S1 and Table S2). Histological analyses of the testes by light microscopy showed normal morphology (Figure 2A and 2B). In mice, the spermatogenic cycle that occurs in each tubule of the seminiferous epithelium is divided into 12 stages, and the germ cells in the seminiferous tubules are enclosed by Sertoli cells [18]. Spermatogonia, spermatocytes, and spermatids were systematically arranged in the seminiferous tubules of the heterozygous mutant and wild-type testis: spermatogonia were found in the tubule walls, whereas spermatids were located in the tubule centers and spermatocytes were observed between the two (Figure 2A). Tubules with an abnormal arrangement of cells undergoing spermatogenesis were rarely observed in the homozygous mutant mice (Figure 2B). To identify apoptotic cells, we performed terminal deoxynucleotidyltransferase-mediated dUTP nick end-labeling (TUNEL) staining using an in situ apoptosis detection kit (Takara, Shiga, Japan) according to the manufacturer' s instructions. There was no statistical difference in signal between testicular sections prepared from the homozygous and heterozygous mutant mice (Figure 2D and 2E). Fully differentiated sperm were observed in the seminiferous tubules by light microscopy and there was no difference in weight between the homozygous and heterozygous mutant testes (Table S2). Electron microscopic analysis revealed that flagellar formation and nuclear condensation occurred normally in spermatids until step nine (data not shown) and in elongated spermatids in the testes of homozygous mutants (Figure 3A and 3B). However, the direction and location of each flagellum was arranged incorrectly at the caudal pole of the nucleus during maturation in the epididymis (Figure 3C and 3D, and Figure S1). The mitochondria and the outer dense fibers were arranged normally, with few mitochondria to drop out in the cytoplasm. Separation of the sperm head and tail was observed in spermatozoa in the cauda epididymis (Figure 4). In wild-type sperm, the components connecting the sperm head to the flagellum were observed as described in earlier studies [19]–[22]. The basal plate attaching to the outer membrane of the nuclear envelope was identified in wild-type and mutant sperm (Figure 4A and 4B). The capitulum, consisting of electron-dense material, was observed between the basal plate and striated columns, which continued to the axoneme (Figure 4A). In the mutant mouse, the capitulum and striated columns were not apparent in the cytoplasm of the separated head (Figure 4B), but they were observed in the separated tail (Figure 4C). These results strongly suggest that separation occurred between the basal plate and capitulum. Disengagement of the tail from the head was accompanied by plasma membrane, and both stumps were sealed (Figure 4B–4D). Although similar numbers of sperm were recovered from the cauda epididymides of the homozygous and heterozygous mutants, almost all of the sperm heads from the homozygous null mice were detached from tails during incubation in culture medium (Figure 5). Meanwhile, the headless tails showed surprisingly normal energetic swimming ability (Videos S1 and S2). They maintained their swimming ability even after 15 h of incubation. The movement of the separated tails looks normal, although no sign of hyperactivation is evident. We also examined the viability of the tailless sperm heads by staining with propidium iodide (PI). The heads could be considered as maintaining membrane integrity because they were resistant to PI staining (Table 2). It is well known that sperm have no fertilizing ability upon ejaculation, undergoing physiological (capacitation) and morphological change (acrosome reaction) before acquiring the ability to fuse with eggs [23]. Acrosome reaction was known to be artificially induced by a treatment of sperm with calcium ionophore A23187. Therefore, we examined whether or not the tailless heads which were found to be “alive” could respond to the ionophore and undergo induced acrosome reaction. As shown in Table 2, these tailless sperm heads showed no response to the ionophore and acrosome reaction did not take place. The role of OAZ-t in acrosome reaction is not clear at present. However, if we recall research indicating the existence of acrosome reaction-related molecules such as AKAPs [24], [25] and CatSpers [26] in tails, it is possible to assume that the induction of acrosome reaction in the head requires signals from tails [27]. The homozygous mutant sperm were not able to fertilize eggs by in vitro fertilization (IVF) assays (data not shown). Therefore, we injected sperm heads derived from heterozygous and homozygous OAZ-t mutant mice into cytoplasm of unfertilized eggs (ICSI). Twenty-two and fifteen two-cell-stage embryos were obtained from the heterozygous and homozygous OAZ-t mutant sperm, respectively. They were transferred to the oviducts of pseudopregnant females and three healthy pups were sired by each genotype. Thus the infertile nature of OAZ-t null sperm is not derived from the defects in the quality of the head itself. Previous study showed that the polyamine concentration in the germ cells increased after meiotic division, whereas the level of ODC activity declined [28], [29]. It has been proposed that Sertoli cells provide polyamines to germ cells [30]. OAZ-t plays to regulation of polyamine concentration in spermiogenesis instead of OAZ1 and 2. OAZ1 binds to ODC with about a three-fold higher potency than OAZ2 [31]. OAZ1 accelerates proteasomal ODC degradation, whereas OAZ2 does not [31]. OAZ1, OAZ2, and OAZ3/OAZ-t indeed differ in their effect on ODC activity in vitro or in bacteria [4], [16], [32], [33]. The concentrations of polyamines in testis and sperm were not affected by the disruption of OAZ-t. Since the activity of ODC was not regulated only by OAZ-t but by other regulatory proteins such as OAZ inhibitor (AZI) [33], it was assumed that the polyamines amount was kept normal by other factors. OAZ-t was dispensable in regulation of total cellular polyamine concentration. A previous study showed that exogenous primary amines induced head-tail dissociation as a result of the separation of the inner and outer nuclear envelope membranes adjacent to the tail basal plates [21]. These results indicated that the concentration of primary amines affected construction at the head−tail junction of sperm. Because polyamines are alkanes and include primary amines, the segregation of sperm heads and tails may be caused by a change in the local concentration of polyamines. The orthologue of the OAZ-t gene is reported in human. One team investigated the relationship between OAZ-t polymorphism and male infertility. The researchers found one Pro164Ser mutation in one of the azoospermic patients but it is not clear if this substitution affects OAZ-t function. The researchers did not claim an association of OAZ-t polymorphism to human male infertility [34]. In previous studies, decaudated tails and decapitated heads lacking the implantation fossa and basal plate at the caudal pole of the nucleus were observed in infertile patients [35], [36]. This phenotype is consistent with the null mutation of OAZ-t in mice. OAZ-t may thus be one of the genes responsible for decaudated tails and decapitated heads in human sperm, although the neckpieces of human and mouse sperm are not identical. Our OAZ-t-disrupted mouse line may offer insight into the mechanism of spermatogenesis. All animal experiments conformed to the Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Committee of Laboratory Animal Experimentation (Nagasaki International University, Nagasaki, and the Research Institute for Microbial Diseases, Osaka). The mice were kept under controlled temperature and lighting conditions throughout the experiments and were provided with food and water ad libitum. The OAZ-t targeting construct was created by the amplification of a homologous 4. 0-kb 5′-arm and 9. 0-kb 3′-arm using 129Sv genomic DNA as the template. The primers used to amplify the arms were designed to incorporate synthetic enzyme sites at both ends. The amplified fragments were digested to create sticky ends and the clone was sequentially ligated into the poly-linker cloning sites on either side of the neomycin resistance gene in the targeting vector backbone. The targeting vector contained the neomycin resistance gene and a thymidine kinase gene, both under the control of the PGK promoter. The vector plasmid was linearized by NotI digestion prior to electroporation into W9. 5 ES cells. Of 720 G418 gancyclovir-resistant clones screened, two were found to have undergone homologous recombination correctly by Southern blot analysis. The four targeted cell lines were injected into C57BL/6J blastocysts, resulting in the birth of male chimeric mice. Highly chimeric males were mated with C57BL/6J wild-type females to generate F1 offspring, half of which were heterozygous for the targeted allele. Of the two ES cell lines injected, both lines produced a high percentage of chimeras that entered the germline. Heterozygous F1 males were then crossed to C57BL/6 females to obtain heterozygous F2 animals. The heterozygous F2 animals were bred to produce homozygous mutants and to check for Mendelian inheritance. The mice were bred and maintained in our laboratory animal facilities and used in accordance with the guidelines for the care and use of laboratory animals set forth by the Japanese Association for Laboratory Animal Science. Genomic DNA was extracted from the tails of the mice using standard procedures. Southern blotting was conducted to determine the site of integration for the gene trap sequence in the oaz-t locus and to genotype the mice. A 3′ external probe was generated by PCR (primers 5′-CATGATGTCACTGACTCTTTCC-3′ and 5′-CAATGGAAGATGGAAGAATATG-3′) from mouse genomic DNA. Genomic DNA samples (10 µg) were digested with SacI and electrophoresed on 0. 8% agarose gels. All hybridizations were performed using standard protocols. The mice were genotyped by PCR using two sets of primers (Figure 1C) as follows: one set of primers (5′-ATCTGGACGAAGAGCATCAGGGG-3′ and 5′-CCTCAGAAGAACTCGTCAAGAAG-3′) to amplify the Neo gene and one set of primers (5′-TCAGGCCTTGGATCAAGGCAACCG-3′ and 5′- CATACTCCAGTGTTGCTGTCAAGC -3′) for the oaz-t gene. To examine the expression of oaz-t, northern blotting was performed according to the manufacturer' s instructions using PerfectHyb (Toyobo, Osaka, Japan) [4]. Western blotting was performed according to a previously-described protocol [4]. For morphological observation, testes were fixed in Bouin' s solution, embedded in paraffin, and sectioned at a thickness of 8 µm. Deparaffinized sections were stained with hematoxylin and eosin. Sperm from the cauda epididymis were cultured in TYH medium (119 mM NaCl, 4. 8 mM KCl, 1. 7 mM CaCl2,1. 2 mM KH2PO4,1. 0 mM MgSO4,25 mM NaHCO3,5. 6 mM glucose, 0. 5 mM sodium pyruvate, and 4 mg/ml BSA) for 30 min, spotted onto glass slides, and dried. Testes and the caput, corpus, and cauda epididymis of wild-type and mutant mice were fixed in fixative consisting of 4% paraformaldehyde, 2% glutaraldehyde, 0. 05 M HEPES-KOH buffer (pH 7. 4) and 0. 02% CaCl2 for 2 h at room temperature. Spermatozoa from the cauda epididymis were suspended in PBS and centrifuged at 1000× g for 5 min. The pellets were fixed with the same fixative as mentioned above. All fixed samples were post-fixed with 1% reduced osmium for 1 h, dehydrated in a series of graded ethanol solutions, and embedded in Epon. Thin sections were stained with lead citrate and examined with a Hitachi H7650 electron microscope. The status of the acrosome was evaluated by staining with FITC–PNA (Sigma-Aldrich), which binds the outer acrosomal membrane. Sperm samples were dried on glass slides and fixed with 70% methanol at −20°C for 5 min after incubation in TYH medium at 37°C in a humidified incubator containing 5% CO2/95% air. A23187 (Sigma-Aldrich) was added at a final concentration of 10 µM to induce the acrosome reaction [37]. Fluorescence-activated cell sorting (FACS) analysis was used to monitor the activity of the sperm following PI staining. ICSI was performed as described [38]. Briefly, sperm collected from the epididymides of the mice were suspended in 12% polyvinylpyrrolidone (360 kDa; PVP) and decapitated with a Piezo pulse (Prime Tech Ltd. , Tokyo, Japan). The detached heads were then introduced into the cytoplasm of unfertilized cumulus-free eggs. After being incubated in kSOM for 24 h [39], the eggs were transplanted at the two-cell stage to pseudopregnant females. Differences between the experimental and control conditions were compared using one-way analysis of variance with Fisher' s protected least significant difference test. Significant differences (P<0. 01) are discussed here.

Polyamines are essential for cell proliferation and differentiation, but their role in these processes is unknown. Ornithine decarboxylase antizymes (OAZs) are enzymes that control the concentration of polyamines in cells. To elucidate the role of one of these enzymes, OAZ-t, in the regulation of polyamine concentration during sperm formation, we generated mutant mice in which the OAZ-t gene was disrupted. When we observed sperm from the mice lacking a functional Oaz-t gene, we found that the sperm heads separated easily from the tails, indicating that OAZ-t is essential for the formation of a rigid junction between the head and tail during sperm development. Many of the headless tails could continue swimming, but they were unable to participate in the signaling processes required for successful fertilization. However, tailless heads could produce healthy pups when injected into unfertilized eggs. Such a phenotype has not been previously found. The mutant mice evoked rare cases of infertile human patients whose sperm behaves in a proper fashion. Our study underscores the importance of research into the processes of spermatogenesis and fertilization.

Abstract Introduction Results Discussion Materials and Methods

developmental biology/germ cells

2009

OAZ-t/OAZ3 Is Essential for Rigid Connection of Sperm Tails to Heads in Mouse

Mosquito biting frequency and how bites are distributed among different people can have significant epidemiologic effects. An improved understanding of mosquito vector-human interactions would refine knowledge of the entomological processes supporting pathogen transmission and could reveal targets for minimizing risk and breaking pathogen transmission cycles. We used human DNA blood meal profiling of the dengue virus (DENV) vector, Aedes aegypti, to quantify its contact with human hosts and to infer epidemiologic implications of its blood feeding behavior. We determined the number of different people bitten, biting frequency by host age, size, mosquito age, and the number of times each person was bitten. Of 3,677 engorged mosquitoes collected and 1,186 complete DNA profiles, only 420 meals matched people from the study area, indicating that Ae. aegypti feed on people moving transiently through communities to conduct daily business. 10–13% of engorged mosquitoes fed on more than one person. No biting rate differences were detected between high- and low-dengue transmission seasons. We estimate that 43–46% of engorged mosquitoes bit more than one person within each gonotrophic cycle. Most multiple meals were from residents of the mosquito collection house or neighbors. People ≤25 years old were bitten less often than older people. Some hosts were fed on frequently, with three hosts bitten nine times. Interaction networks for mosquitoes and humans revealed biologically significant blood feeding hotspots, including community marketplaces. High multiple-feeding rates and feeding on community visitors are likely important features in the efficient transmission and rapid spread of DENV. These results help explain why reducing vector populations alone is difficult for dengue prevention and support the argument for additional studies of mosquito feeding behavior, which when integrated with a greater understanding of human behavior will refine estimates of risk and strategies for dengue control. Dengue is the most important arboviral diseases of humans worldwide. It occurs throughout most tropical regions. An estimated 390 million people are infected each year and approximately 96 million people suffer from clinically apparent disease annually [1], [2]. Aedes aegypti is the principal mosquito vector of the four dengue virus serotypes, lives in close association with humans, feeds preferentially on human blood [3]–[5] and has a tendency to ingest multiple blood meals during each gonotrophic cycle [4], [6] facilitating efficient transmission of human blood-borne pathogens. Although a tetravalent dengue vaccine is under development [7], a vaccine and anti-viral drugs are not currently commercially available. As a consequence, current dengue prevention programs are limited to control of the mosquito vector [8]. In this study, we used DNA fingerprinting to detect the individual human hosts from whom female Ae. aegypti took blood meals. Genetic markers have been applied to a variety of studies on mosquito feeding patterns (reviewed by Kent [9]). Coulson [10]was the first to investigate DNA fingerprinting for identifying individual human hosts. Others used a similar approach to address questions about mosquito feeding behavior and bed net efficacy [11]–[15]. In 2000, Chow-Shaffer et al. [16] used variable number tandem repeats (VNTRs) and short tandem repeats (STRs) in a pilot study to fingerprint human DNA in blood engorged Ae. aegypti collected in Thailand. De Benedictus et al. [14] applied the same approach to Ae. aegypti collected in Puerto Rico to analyze feeding patterns on 84 human hosts living in 22 houses and reported that 18% of blood meals were identified as coming from one of two people in a 36 hr time period. Herein, we report the fine-scale details of Ae. aegypti blood feeding patterns on individual human hosts in a dengue endemic community. In order to better understand Ae. aegypti-human host interactions that underlie local DENV amplification and spread as well as human risk for infection, we used six human microsatellite markers to reconstruct blood feeding patterns of Ae. aegypti collected over multiple seasons, villages and years in west central Thailand. Our study objectives, were to (1) estimate the frequency at which Ae. aegypti bites different people in a 24 hr period; (2) determine whether human age, gender or house of residence predict the frequency at which different people are bitten and (3) evaluate the effect of advancing mosquito age on blood feeding patterns. In large outdoor field cages, we tested hypotheses concerning the impact of human host size and body position on biting behavior. In the laboratory we determined the accuracy of identifying mixed and degraded human DNA in mosquitoes that fed on more than one host. Our results indicate that frequent and heterogeneous biting by Ae. aegypti on residents and transient visitors and mosquito feeding/transmission hotspots are important entomologic features of dengue epidemiology. Our study was conducted in four villages in northwestern Thailand: Pai Lom (16°45′N, 98°33′E) and Lao Bao (16°45′N, 98°34′E) located in Mae Pa district, 5 km north of Mae Sot in Tak Province; Mae Kasa (16°53′N, 98°37′E), located 20 km north of Mae Sot and Mae Dow (16°53′N, 98°37′E), located 20 km south of Mae Sot. Our field laboratory, a vacant village home, was located approximately 1 km from Pai Lom and Lao Bao. Descriptions of field sites, temperature and humidity during collection periods were previously described by Harrington et al. [17], [18]. Experiments were conducted during both the cool dry season (February 2000,2001,2002, and 2003) and warm rainy season (July 2000,2001, and 2002). These times of the year correspond to periods of low (dry) and high (rainy) dengue transmission in Thailand [19]. Mosquitoes were collected from inside houses using CDC backpack aspirators. Aspirator cartons were placed in plastic bags on wet ice and transported to the field laboratory where mosquitoes were anesthetized with CO2, chilled and sorted by species. During February 2000 and July 2000, abdomens of engorged female Ae. aegypti were saved for DNA analysis by smearing on filter paper, drying and placing in a sterile microcentrifuge tube. Collections from January/February 2001–January/February 2003 were preserved by homogenizing abdomens in 400 µl lysis buffer (1% SDS, 50 mM EDTA, 10 mM Tris-HCL, di H2O) in individual sterile microcentrifuge tubes and transported to the University of California at Davis or Cornell University for further analysis. The right wing was removed from each female and saved for body size estimation. Forceps were sterilized and air dried between each mosquito to prevent cross contamination of samples. Legs from each mosquito were removed with clean forceps and placed in a hexane washed vial for cuticular hydrocarbon (CH) age grading [20]. After obtaining informed consent from study subjects, human DNA samples were collected by gently swabbing the inner cheek with a sterile wooden applicator stick. Four swabs were taken and swirled gently in lysis buffer. Each human sample was provided with a unique code indicating the person, village and date of collection. During each subsequent collection period from 2000–2003 we recorded which individuals were present, who left and who joined the study community. People from whom incomplete profiles were obtained on previous visits were re-swabbed. Participant data was numerically coded to protect the identity of subjects. Children were provided with vitamins and milk as compensation following collection of samples. This research project was conducted with the approval of, and in accordance with, Institutional Review Boards (IRB) at the University of California at Davis (200210073), Walter Reed Army Institute of Research (752), Thai Ministry of Health Ethical Review Committee for Research in Human Subjects, and Cornell University (FWA00004513). All adult subjects provided written informed consent, and a parent or guardian of any child participant provided informed consent on their behalf. Mosquito blood meals were extracted at UC Davis or Cornell University following methods described previously [14]. The amount of human DNA in samples was measured and distinguished from mosquito DNA using the Quantiblot Human DNA Quantitation system from Perkin Elmer (Wellesley, MA) following the manufacturer' s instructions. Six human loci and a gender identification locus (AMEL) were amplified with PCR using Geneprint primers (Promega Corporation, Madison, WI). These loci were selected because they have been well characterized [21], have a high number of alleles, and relatively small PCR product lengths for greater detection probability through the mosquito blood meal digestion process. The AMEL gender identification locus (212 X, 218 Y bp) was employed, as well as the CSF1PO (291–327 bp), THO1 (179–203 bp), TPOX (224–252 bp), D16S539 (264–304 bp), D7S820 (215–247 bp) and D13S317 (165–197 bp) loci. DNA was amplified in a DNA engine Dyad thermocycler (MJ Research, Waltham, MA). PCR products were run on 4–6% acrylamide-bis denaturing gels and visualized with silver staining as described previously [14]. After drying, gels were examined on a light box and alleles in each mosquito blood meal were assigned a number by visual comparison to a reference 100 bp DNA ladder (Promega, Madison, WI, USA). Mosquito blood meals were included in the final analysis only if amplification was successful with at least 5 of the 6 loci. All samples with incomplete profiles after a second PCR reaction were discarded. Two independent methods were used to estimate mosquito age. The first was mark–release–recapture, as described by Harrington et al. [22]. Briefly, mosquitoes were collected as pupae from natural immature development sites in the study villages. After emergence, adult females of known age were transferred to small cardboard cartons and dusted with florescent powder (DayGlo Color Corp, OH, USA). A unique color was used for each release day. Mosquitoes were released in houses within the study community after obtaining informed consent from residents of each house. Marked mosquitoes were subsequently recaptured with CDC backpack aspirators (John W. Hock Co, Gainesville, FL USA). Collected mosquitoes were transported to the field laboratory where they were anesthetized, identified to species, and examined under a dissecting microscope for florescent dust markings. Blood engorged marked and recaptured mosquitoes were processed as described above and their age was assigned based on known days since eclosion. A second age-grading method utilized CH analysis as described by Gerade et al. [20]. Briefly, legs were removed from each specimen using clean forceps, placed in dry n-hexane washed vials, and stored until further processing at the University of Massachusetts. Hydrocarbons were extracted and analysis of legs from each mosquito was conducted against an internal standard of octadecane as described by Gerade et al. [20]. Two experiments were conducted to determine the time intervals over which DNA could be detected in a mosquito that ingested one or two (from different people) blood meals. Large enclosures were constructed over vacant houses in Pai Lom as described previously [22]. Field cages encompassed an entire house and yard (∼10 m wide ×10 m wide ×4 m high). Three-day-old female Ae. aegypti (eclosed from field collected pupae) were marked with colored dust (as described above) and released inside the field cage during July 2002. Mosquitoes were released over 4 consecutive days. A total of 353 (112,60,132, and 49 over days 1–4, respectively) three-day-old non-blood fed females were released inside and outside a house in the enclosure each evening. On the following day, four hosts (study authors and collaborators) entered the enclosure and remained inside for 30 min; two people were inside the house (one sitting and one lying down) and two people were outside in the same sitting or lying positions. Host location and position was rotated each day. Engorged females were collected with CDC backpack aspirators from the house and yard each day after exposure to hosts. Human DNA was extracted, amplified, profiled and matched to each participant as described above. The experiment was repeated with the same methods and 3 of the same 4 participants and one new participant during January 2003. A total of 863 (137,75,77,324,155, and 95 over days 1–6 consecutively) female Ae. aegypti were released into the enclosure as described above for July 2002. In order to understand whether there were spatial patterns of feeding that deviated from random (e. g. “hotspots” or “cold spots”), we used our DNA fingerprinting results to build interaction networks between mosquitoes and human hosts in the villages of Lao Bao and Pai Lom during each sampling period. Each house was considered as a node. Connections were made between nodes (houses) based on mosquito blood meals, linking the house where the mosquito was collected with the house (or houses) were human host (s) lived. We used a traditional method to characterize networks by evaluating the derived network' s topological and structural properties and comparing them with those of random networks [23]. We compared the degree distribution of each network, which was based on the number of connections that a node has; i. e. , its degree. The greater the number of connections, the greater the degree. For this analysis, the majority of nodes, therefore, had approximately the same degree (close to the average k of the network). To determine if there were “hotspots” or “cold spots”, we compared the degree distribution of the observed mosquito-human biting networks, with the degree distribution obtained from 999,999 Monte Carlo randomizations (to avoid artifacts) using χ2 analysis at a 5% significance level. To test the hypothesis that mosquitoes remained in or close to the houses where they were collected and people moving from house to house were bitten [18], [24], [25], we examined the spatial autocorrelation of the relationship between mosquitoes and human hosts based on observed mosquito bites from people in each house. Blood meals where the human host house of residence matched that of the mosquito collection house were defined as “resident” meals. “Non-resident meals” were designated when mosquitoes fed on someone who did not live in the mosquito collection house. Frequencies of blood meals matched to two different people were compared for experiments that examined time limits of host DNA detection. DNA profiles for mosquitoes and human cheek swabs were compared in a common data base. Questionable or partial profiles were re-amplified and re-run on gels for confirmation. All data were analyzed using two custom programs: Mosquito Matcher and/or Blood Match, which are available from the authors upon request. Both programs allowed matching analysis of data in two different excel spreadsheets. In this way the mosquito blood meal profiles could be matched to the human DNA profiles by village, season, and year. Match ID within Mosquito Matcher allowed matching of human DNA sets with each other to identify non-unique single profiles among the human population and non-unique double profiles in a theoretical mixed blood meal. The frequency of single and multiple blood meals were initially analyzed by village of residence, season, year, and mosquito age with cross tabulations, χ2 test of independence and t-test. A logistic regression model was then used to model single and multiple blood feeding as a function of these variables simultaneously. Logistic regression was conducted to test the effect of village of origin, host age and sex on the probability of a person being bitten or not. A negative binomial regression model was used to test the effect of village of origin, age and sex on the number of times of a person was bitten. To analyze the effect of host age on probability of being bitten and number of times a person was bitten, age was sorted into two different types of categories. The probability of being bitten was compared among people placed in age classes by each decade from 0 years to an arbitrary upper age limit of 110. The data were also compared for people aged 0–11 representing the age group with a high probability of DENV non-immune individuals [26]. Data for the proportion of mosquitoes captured in the same house where the person they bit lived were compared across season, year and village with cross tabulations and χ2 tests of independence. A logistic regression model was then used to test the effect of all variables simultaneously. In field cage experiments to compare the effect of body mass and position on host feeding patterns, a body mass parameter was calculated for each host by multiplying the height of the host (m) by their weight (kg). The body mass parameter was compared with feeding frequency across replicates. Regression analysis of the proportion of mosquitoes fed by body mass was performed. Spatial autocorrelation for the network analysis was tested locally using G statistics [27]. For a distance d, a matrix of neighbor was generated and then local clustering was calculated using Gi (d). Due to differences on inter-house distance, the grain of the analysis was 10 m for Lao Bao and 5 m for Pai Lom; i. e. , the variable d was increased every 10 and 5 m, respectively. The significance of G was evaluated using 9,999 Monte Carlo randomizations at a 5% significance level. We used this approach to determine whether popular daytime aggregation sites represent high biting risk to humans, such as homes with attached stores where residents frequented to purchase goods (local markets), were blood feeding hot spots. All statistical analyses for comparison of blood meal frequency, season, village and year, as well as the effect of host body mass and host position, were performed in (SPSS Statistics 17. 0, SPSS Inc. , Chicago, IL). Statistical analyses on the interaction networks and local clustering were performed in R [28]. Cheek swab DNA samples were completely profiled for 676 residents from the four study villages and all study collaborators and mosquito collectors (n = 28), who periodically visited study villages. Each individual profile was unique with the exception of two identical twin boys in one village and a mother and daughter with identical profiles that lived in another village. To understand our ability to detect multiple feeding (two different people in one blood meal), we analyzed all the hypothetical combinations of two people using MatchID following the methods of DeBenedictus et al. [14]. A high percentage (85–94%) of these hypothetical combinations were unique (Table 1). A total of 3,677 blood engorged mosquitoes were collected for DNA fingerprinting analysis over the course of the study. Of these specimens, we obtained complete profiles for 1,186, with the remaining samples likely too degraded to profile completely. Of the 1,186 blood meals completely profiled, 430 (36%) matched the profile of a person (s) living in a study village or a study collaborator/mosquito collector (n = 10). Hourly temperature during the time series experiment in January 2001 ranged from 18 to 36°C, with an average temperature of 26±0. 22 SE. At time <1 hr, mosquito blood meals contained an average of 73. 6±25. 6 ng of human DNA. DNA concentration decreased only slightly for the next 24 hr (54 ng±15 at 24 hr/8. 9 degree days (DD) ) and then decreased rapidly thereafter (Fig. 1). Only trace amounts of DNA were amplified in blood meals sampled after 30 hr/11. 1DD. Results for amplification success of alleles at all loci were consistent with concentration data. Complete DNA profiles from human blood were detected up to 30 hr after feeding. DNA in samples taken at 36/13. 3 and 42/15. 5 hr/DD was likely too degraded for amplification. When mosquitoes took replete blood meals from two hosts in a sequence separated by 24 hr (8 DD, n = 28), only the second host DNA was detected in the blood meal, even when the order of hosts was reversed. To simulate interrupted feeding (partial blood meals from two different hosts) mosquitoes were offered incomplete blood meals first from person A and a second blood meal from person B. With interrupted meals, partial profiles of both hosts were detected, in a small number of mosquitoes. The first host was detected in 3 of 18 (17%) mosquitoes when meals were separated by 0 hr and in 2 of 57 (4%) mosquitoes when meals were separated by 6 hr (2DD). Feeding frequencies on four different people were directly related to host body mass in our field cage experiments. In July 2000, the majority of mosquitoes (38%) fed on the person with the largest body mass parameter (226) followed by 30% on the person with the next highest mass (187). The two smaller people (with 114 and 109 mass parameters) each were bitten 22 times (11%). In the second replicate (January 2003), 55% of all blood meals were again from the largest person, followed by 32% from the second largest person, and 12% and 2% were from people with 114 and 95 body mass, respectively. These results reveal a direct relationship between increasing host height and weight and the number of times bitten (Fig. 2) (July: adjusted R2 = 0. 81, F = 13. 6, P = 0. 06; January adjusted R2 = 0. 96, F = 75. 9, P = 0. 01). A total of 47 females with complete DNA profiles were age-graded using cuticular hydrocarbon ratios. A total of 43 females ingested a single blood meal and 4 took double blood meals. No significant difference in age by blood meal number was detected. The mean age for females ingesting blood from 1 person was 5. 4 days (±0. 35, range <1–12). The mean age for females that fed on more than one person was 5. 4 days (±0. 45, range <1–13). The number of “house resident” and “non- resident” meals per season and village is presented in Table 7. Overall, the majority of mosquitoes were captured in the same house where the person they bit lived (64. 6%). More mosquitoes were captured in the same house as where their blood hosts lived during the rainy (69. 7%) than dry season (45. 1%, χ2 = 19. 2, df = 1, P<0. 0001). Significant differences also were detected by year and village. The greatest percentage of mosquitoes were captured in the same house where their hosts lived during 2001 and 2002 (72. 7% and 65. 6%, respectively, χ2 = 25. 7, df = 3, P<0. 0001). Greater than 62% of mosquitoes were captured in the same house as the person they bit in the villages of Mae Kasa, Lao Bao, and Pai Lom than Mae Dow, where the smallest number of profiled mosquitoes were collected (49. 3%) (χ2 = 19. 7, df = 3, P<0. 0001). Logistic regression analysis revealed a significant effect of year (Wald = 0. 000, df = 1, P<0. 0001) and village (Wald = 0. 00, df = 1, P<0. 0001). Observed biting networks are presented in Fig. 4. Analysis of the degree distribution (Table 8) revealed no significant clustering of feeding patterns for interaction networks in Pai Lom over all sampling events. Significant house-level clustering patterns were observed, however, for Lao Bao for the 2002 rainy high dengue transmission season and the 2003 dry low dengue transmission season. In both cases more households were disconnected with more localized feeding patterns on house residents and less connections to feeding on hosts in other houses. Clustering patterns for mosquito biting was associated with the presence of hot and cold spots in the villages (Fig. 5). These demonstrated more (hot) and less (cold) than expected resident and non-resident blood meals, respectively. Hot and cold spots were detected at several scales during most of the sampling events in both Lao Bao and Pai Lom (Table S1. A–D). Values in Table 8 indicated with parenthesis and a star symbol were those that included the village local market. Values report the interval of distances (m) where significant clustering of bites was detected. For example, in Table 8 B for Lao Bao during the 2001 warm rainy season, hot spots were detected with neighboring houses if the distance among them was 30–60 m, 80 m, and 100–130 m. For 60 m, the clustering included the local market. Although statistically significant, reported clusters at scales bigger than 60 for Lao Bao may not be accurate because of statistical constraints due to the size of the village which was smaller than the two larger cluster sizes. For Lao Bao, local clustering for “resident” bites was observed only during the rainy seasons (2001 and 2002), and in both cases hot spots included the local market. Local clustering for “non-resident” bites was detected for all the sampling events, and the local market was included when neighborhoods were calculated at 80 m or less. For the cool dry low dengue transmission season of 2001 in Pai Lom, the local market was included in the households detected as cold spots for both the “resident” bites (at scales 95–120 m), and the “non-resident” bites (at 65–125 and 160–185 m). In addition, spatial linear regression analysis confirmed a global trend of increasing mosquito biting activity northeast of the local market (p = 0. 005); i. e. , increasing latitude (p = 0. 03), decreasing longitude (p = 0. 02). Ae. aegypti feeds preferentially on humans, and is adapted to living in close proximity to humans, often resting and blood feeding within human dwellings [3], [29]–[31]. This mosquito vector is permissive to infection and transmission of DEN viruses, although vector competence varies with viral, environmental and mosquito genetic factors [32], [33]. The burden of dengue in countries such as Thailand where we conducted our study, is high [26]. Despite this high force of transmission, Ae. aegypti is often found in surprisingly low densities in and around human households and mosquito DENV infection rates during epidemics are typically low (3–7% in Singapore; <1. 0% in Khamphaeng Phet, Thailand; 4. 0% in Southern India [26], [34], [35], but can vary on fine spatiotemporal scales [36]. This entomological paradox raises important questions about mosquito feeding behavior and the factors that drive DENV transmission. The goal of our study was to investigate whether heterogeneities in Ae. aegypti blood feeding behavior are consistent with epidemiologically meaningful patterns, evidenced by seasonal shifts in multiple feeding patterns, variation by host age, sex, and location of blood feeding hotspots within communities. To do this we evaluated mosquito human feeding patterns over a long time period that included multiple seasons and years in a dengue endemic setting. Previous studies, conducted on a smaller scale, reported heterogeneities in human feeding patterns for Ae. aegypti [6], [14], [16], Culex quinquefasciatus [13], Anopheles gambiae, and An. funestus [12], [37] with some results similar to ours, such as frequent feeding patterns on individual hosts [14], [38] and a relationship between host body size and biting frequency [37]. The majority of blood meals in our study were from single hosts that were only detected in mosquitoes once or twice, but some hosts were fed on frequently. For example, three people were bitten nine times over the course of our study. No clear patterns emerge regarding why certain people in our study were fed on so frequently, even though we examined a diversity of ages and body sizes. These heterogeneities in biting trends suggest a role for chemical or environmental cues in human host attraction [39] and/or differences in the opportunities for mosquitoes to encounter and bite different people. Host blood detected multiple times in mosquitoes may represent individuals with the potential to contribute more to transmission than others. The relative impact of these individuals as potential “super spreaders” on the dynamics of dengue outbreaks merits additional study [40]. Based on laboratory optimization experiments, our ability to detect more than one host in a single blood meal was limited if the time interval between the two host blood meals was greater than 6 hrs. Additional experiments revealed that the likelihood of identifying a multiple blood meal decreases when mosquito feeding was interrupted. Together, these results suggest that we may significantly underestimate multiple blood feeding within a gonotrophic cycle (typically 3–6 days), and when meals are interrupted due to host defenses and other factors. Given these limits of human DNA detection in a mosquito blood meal our field estimates for multiple feeding likely represent minimum frequencies for Ae. aegypti. Although we completely profiled human DNA from a large number of mosquitoes (n = 1,186), 64% did not match anyone in the study community. Even in our most isolated village, Pai Lom, and taking into account error rates (approximately 0. 20–0. 26), we were not able to match up to 28% of human hosts in blood meals collected. This result likely was not due to Ae. aegypti feeding on non-human blood meals. Numerous studies have shown low non-human blood feeding rates for Ae. aegypti in Thailand and elsewhere [6], [31], [41], [42], including the villages studied here. In addition, the majority of the samples analyzed with our slot blot procedure contained human DNA. Although our house resident vs. non-resident analysis revealed that most mosquitoes fed on house residents, our comparison was only for those non-residents living in the same village that could be matched to blood meals. Our laboratory studies generated incomplete profiles as a result of increasing blood meal digestion time and with multiple meals. Direct sequencing of mosquito blood meals in future studies will likely improve the quality and accuracy of profiled blood meal data. Although we did not monitor human movement in our study we suspect that a considerable number of people were moving transiently through the study communities, either to conduct daily business, work during peak harvest times, or visit friends and relatives. Given our knowledge of the limited short-range movement of Ae. aegypti [43], and recent work on human movement patterns and DENV transmission by Stoddard et al. [24], [25], it is reasonable to speculate that a significant proportion of mosquitoes fed on visitors entering village houses in our study area. Aedes aegypti biting human visitors is a mechanism by which virus could be introduced into and/or carried away from the communities we studied. If feeding rates are high in an introduction zone, dengue transmission “hot spots” could occur, contributing to the focal nature of dengue cases [36], [44]. In our study a local market in one village represented a hot spot during two high dengue transmission seasons. Feeding on visitors may explain how DENV is introduced into communities, while localized feeding hotspots and network biting patterns may explain the focal nature of DENV outbreaks [36], [44]. In addition, more work is needed to understand the factors that influence of mosquito biting clusters. For example, those factors may include the number of people living in a house, the type, quantity, and microclimate of optimal mosquito resting sites and other refugia, as well as landscape barriers to mosquito movement. Due to Ae. aegypti' s propensity to bite older people, in regions where DENV transmission is low, unstable or a novel virus serotype or genotype is introduced, a significant portion of susceptible older individuals may be at higher risk of infection than younger people. In areas of hyperendemic transmission, such as our study area in Thailand, children and young adults are at greater risk of infection because older people will have already been infected with all four virus serotypes [26]. We found a significantly lower feeding frequency on people under 25 yrs of age (approximately 10% of blood meals) than expected if bites were random. Our field cage studies reinforced this result, demonstrating a strong feeding preference of Ae. aegypti for larger hosts when given a choice, consistent with other studies [37], [45]. A range of choices may not be available, however, at all times for host seeking Ae. aegypti. For example, transmission may occur at school where the majority of hosts are small. Or, if a child is the first to return home after a day at school, he/she may be the only person available to bite. We did not collect mosquitoes from schools, and our methods may not have been sensitive enough to detect fine scale timing of mosquito feeding that would convey opportunistic feeding on naïve hosts such as the scenario of a child coming home from school before adults. House scale differences are likely important in DENV transmission, especially when considering naturally low mosquito infection rates (approximately, 1% in or around the home of an infected person and 0. 1% across communities [36], [46]). These results may help explain why DENV outbreaks can rapidly expand in areas of virus introduction and low transmission stability if mosquitoes feed preferentially on susceptible adults. One way to overcome relatively small probabilities of mosquito infection and transmission is with high multiple feeding rates [30], especially by infectious mosquitoes. We estimate from our detection of multiple blood feeding on different human hosts, that nearly half of the engorged mosquitoes we collected (43–46%) fed more than once in an egg-laying cycle, which can last from 3–5 days in the study area depending on the season (Harrington, unpublished data). This frequent biting rate coupled with high survival may overcome low DENV mosquito infection rates and relatively low feeding frequency on epidemiologically naïve hosts. Although we did not detect an age related trend in frequency of feeding, our sample size of feeding rates in older mosquitoes was too small to be conclusive. Future studies with the sole focus of understanding age-related host biting patterns would help clarify this enigma. Our results do not support some explanations for fluctuations in DENV transmission, such as higher feeding rates during the high dengue transmission season [29], [47]. However, given the variation we observed, our data may not have had the precision to detect these differences. Similarly, we did not detect higher rates of feeding on naïve (and, therefore, potentially susceptible or infectious) hosts [47]–[49]. Interestingly, we did observe higher rates of in-house resident biting by mosquitoes during the rainy season, which correlated with the high dengue transmission season in this region of Thailand. We did not investigate mosquito biting by time of day or the amount of time people spent in their homes where they would be at risk of being bitten by resident mosquitoes. Future studies can focus on feeding patterns in the various places outside of home that people visited (including schools, homes of friends and relatives, and gathering places) and how the risk of these aspects of dengue vector biology may reveal epidemiologically important trends. Our results highlight the importance of identifying local hotspots for mosquito biting. If hotspots can be identified, focal insecticide spraying could be more effective and cost less for reducing DENV transmission than treating entire villages. More emphasis should be placed on strategies that identify and test the epidemiologic significance blood feeding hot spots. Our results help explain why vector control alone is difficult for dengue prevention, due to the very low mosquito population levels that may need to be achieved with heterogeneous biting mosquitoes. Aedes aegypti is a highly efficient virus vector because of its frequent and non-random interactions with human hosts. Consequently, relatively few Ae. aegypti females can lead to unacceptable levels of DENV transmission. Additional studies of mosquito feeding behavior, which when integrated with a greater understanding of human behavior, are needed to refine estimates of dengue risk and to improve strategies for its control.

Dengue, a potentially lethal infection impacting hundreds of millions of human lives annually, is caused by viruses transmitted during mosquito blood feeding. With no vaccine or treatment commercially available, understanding the underlying factors linked to virus exposure is critical for developing more effective dengue interventions. We conducted a study in an endemic region of Thailand where transmission is high and children are expected to be the non-immune, amplifying portion of the host population. We examined Ae. aegypti feeding patterns and risk by matching human DNA profiles in blood-fed mosquitoes to study area residents. A small number of meals matched people from the study area, suggesting that mosquitoes feed on people moving transiently through communities. People under 25 years of age were bitten less frequently than older people. We constructed network models to explore the presence of mosquito feeding “hotspots” and detected a local market “hotspot” in one study village during the high dengue transmission season. Our results provide new details on dengue vector feeding patterns and highlight the need to conduct integrated studies of vector feeding and human behavior, and virus transmission patterns in order to better understand the dengue transmission efficiency and spread.

Abstract Introduction Methods Results Discussion

biology and life sciences medicine and health sciences

2014

Heterogeneous Feeding Patterns of the Dengue Vector, Aedes aegypti, on Individual Human Hosts in Rural Thailand

Exotic invasive species can influence the behavior and ecology of native and resident species, but these changes are often overlooked. Here we hypothesize that the ghost ant, Tapinoma melanocephalum, living in areas that have been invaded by the red imported fire ant, Solenopsis invicta, displays behavioral differences to interspecific competition that are reflected in both its trophic position and symbiotic microbiota. We demonstrate that T. melanocephalum workers from S. invicta invaded areas are less aggressive towards workers of S. invicta than those inhabiting non-invaded areas. Nitrogen isotope analyses reveal that colonies of T. melanocephalum have protein-rich diets in S. invicta invaded areas compared with the carbohydrate-rich diets of colonies living in non-invaded areas. Analysis of microbiota isolated from gut tissue shows that T. melanocephalum workers from S. invicta invaded areas also have different bacterial communities, including a higher abundance of Wolbachia that may play a role in vitamin B provisioning. In contrast, the microbiota of workers of T. melanocephalum from S. invicta-free areas are dominated by bacteria from the orders Bacillales, Lactobacillales and Enterobacteriales that may be involved in sugar metabolism. We further demonstrate experimentally that the composition and structure of the bacterial symbiont communities as well as the prevalence of vitamin B in T. melanocephalum workers from S. invicta invaded and non-invaded areas can be altered if T. melanocephalum workers are supplied with either protein-rich or carbohydrate-rich food. Our results support the hypothesis that bacterial symbiont communities can help hosts by buffering behavioral changes caused by interspecies competition as a consequence of biological invasions. Rapid development of global trade and travel have created conditions for long-distance migration and concomitantly increased the threat of biological invasion by exotic species [1]. Invasive species can affect the distribution, abundance and reproduction of native taxa [2,3] and disturb the structure and function of ecosystems [4]. Invasive species can also cause severe economic losses in agriculture, forestry and fishery, and potentially threaten the health of humans [5,6]. Invasive ants are among the greatest threats to ecosystems; dozens of species have invaded islands and continents around the world [7]. In particular, the red imported fire ant, Solenopsis invicta Buren, relies on behavioral and numerical dominance to displace endemic, native and other locally occurring taxa, including previously introduced species (hereafter referred to collectively as “resident” species) across its introduced range [8]. Once established, S. invicta acts as an omnivore and ecosystem engineer, with dramatic effects on ecologically similar resident ants. In the United States, the congeneric species S. xyloni McCook and S. geminata Fabricius appear to be particularly sensitive to displacement by S. invicta [9]. In central Texas, S. invicta has also been reported to destroy and eradicate colonies of the harvester ant, Pogonomyrmex barbatus Smith [10]. With their superior competitive abilities, invasive species impose strong selection on resident taxa, some of which have been shown to adapt to these pressures in different ways [11]. To persist in invaded areas, common adaptive behavioral responses by resident species include altered anti-predator defenses [12,13] and changes in the spectrum of used resources [14] and habitats [15]. Recognizing these interactions is critical to understanding the long-term impacts of biological invasions. However, responses by resident species to selective pressures imposed by invasive species, as well as specific underlying mechanisms that give rise to these responses are still poorly understood. Symbiotic bacteria can be essential for the growth and survival of their hosts [16–18]. They can play an integral role in the breakdown of food, recycling and provision of energy, production of vitamins, and even shape innate immunity [19–22]. Microbial symbionts have been shown to have broad effects on the health and behavior in humans and other mammals [23]. In the context of biological invasions, they have largely been studied to identify or assess their effects in enhancing the invasion process of introduced species [24–27]. A better understanding of the composition and function of bacterial symbionts, however, might also reveal potential mechanisms for behavioral change of resident host species to exotic invasives, since changes in the bacterial symbionts have been shown to correspond to changes in food resources in both vertebrates and invertebrates [28–33]. Furthermore, characterizing the bacterial symbionts of resident species could provide important clues about how to manage invasive species. Tapinoma melanocephalum (Fabricius) (Hymenoptera: Formicidae), the ghost ant, is a cosmopolitan ant species that is common in southern China. It was first recorded in China in 1921 and is likely to have originated in the Indo-Pacific region [34]. It has successfully invaded both human-disturbed and undisturbed natural habitats of tropical and subtropical regions of the world [34–36]. Tapinoma melanocephalum colonies are typically polygynous, unicolonial and resilient to disturbance [37], and therefore display key features of successful invasive ants [38–40]. Although interactions between invasive and resident ants are well documented [41,42], examples of interactions between two different invasive ant species are rare. In a previous field investigation, we found that T. melanocephalum often persists in areas invaded by the fire ant S. invicta, first recorded in mainland China in 2004 [43]. Due to their aggressiveness and capacity to reach high population densities, most available food resources can be used by workers of S. invicta [44,45]. In the field, workers of S. invicta can outcompete those of T. melanocephalum in the use of available honeydew and thus may coerce T. melanocephalum into utilizing a different ecological niche [44,45]. Here, we examined three main hypotheses: (1) Workers of T. melanocephalum may exhibit less aggression in response to invasions by S. invicta. (2) Colonies of T. melanocephalum living in S. invicta invaded areas have a different diet than those inhabiting non-invaded areas, presumably due to competition for resources. (3) The bacterial community found in workers of T. melanocephalum in S. invicta invaded areas differs from that found in workers in non-invaded areas, and these differences are associated with their difference in diet. To address these hypotheses, we investigated several possible mechanisms involved in the behavioral differences of T. melanocephalum following invasions by S. invicta. Stable isotope analysis can reveal differences in feeding activities [46], and previous research has indicated that δN is typically correlated with the trophic level and nutritional state of an organism [47]. We examined stable isotope composition and symbiotic bacterial communities of T. melanocephalum workers from S. invicta invaded and non-invaded sites, and measured responses of colonies to different diets in an attempt to simulate the field conditions. The laboratory simulation provided additional support for the field observations as we were not allowed to introduce S. invicta into non-invaded sites. Thus we did not directly assess the situation before and after invasion by S. invicta at individual sites, but assessed effects by comparing invaded and non-invaded sites in combination with laboratory experiments. The results provide insights into mechanisms of host responses to interspecific competition by two co-occurring ant species with invasive traits. By trapping and baiting, 3,947 workers belonging to 22 ant species were collected in the S. invicta non-invaded areas, while 8,005 S. invicta workers and 8,412 workers belonging to 14 other ant species were collected in the S. invicta invaded areas. We found that the numbers of workers of several ant species, especially T. melanocephalum in the S. invicta invaded areas, were significantly higher (626 T. melanocephalum in the non-invaded areas versus 4,034 T. melanocephalum in the invaded areas) (S1 Table). Compared with the non-invaded areas, the Simpson dominance index (C) was significantly higher in the invaded areas, while the Shannon-Wiener index (H’) and Pielou evenness index (E) were significantly higher in the non-invaded areas (Fig 1). These results indicate that species composition of resident ant communities are significantly different between S. invicta invaded and non-invaded areas, and S. invicta invasion may be one reason for the differences. A comparison of the aggressiveness index based on encounters between T. melanocephalum workers from invaded and non-invaded areas with S. invicta shows that most encounters involved lower levels of attack (level I and level II), accounting for 66% and 52% of the total scores in invaded and non-invaded areas, respectively (Fig 2A). Based on the aggressiveness index, T. melanocephalum workers inhabiting S. invicta invaded areas displayed lower levels of antagonism compared with those inhabiting non-invaded areas (Fig 2B). In the group aggression experiment, T. melanocephalum workers from the S. invicta non-invaded areas showed higher rates of mortality than those from the S. invicta invaded areas at 0. 5h, 1h, 2h and 4h (Fig 2C). Moreover, after first contact with workers of S. invicta, the T. melanocephalum workers from the non-invaded areas had significant lower attack indexes at 0. 5 h, 1 h and 4 h than those without contact experience (F4,70 = 4. 411, P = 0. 003) (Fig 2D). These results support the hypothesis that T. melanocephalum workers living in habitats invaded by S. invicta display submissive behaviors in order to avoid attack by fire ants. Analysis of stable isotopes of T. melanocephalum workers from invaded and non-invaded areas showed that workers from invaded areas had significantly higher δN than those from non-invaded areas (Fig 3). Usually, lower δN values are associated with ants primarily feeding on plant-derived diets such as nectar and insect honeydew, while higher δN are found in omnivorous ants [48]. Thus the higher δN value in T. melanocephalum from invaded areas may suggest that the feeding habits of these ants differ from those in non-invaded areas. To assess fine-scale geographic variation in stable isotope values across sites, we analyzed three species of annual Asteraceae (including a dominant species, Bidens pilosa) occurring as invasive weeds in both invaded and non-invaded areas, and found that these plants did not differ in δN values between these two areas. A fourth species, the invasive annual (or sometimes perennial) herb Mimosa pudica (Fabaceae) had significantly lower δN values in invaded sites than in non-invaded sites (S1 Fig). Overall these results, in particular those of the dominant plant, B. pilosa, suggest that the stable isotope composition of plant communities are similar across the two areas. Thus the different stable isotope signature of T. melanocephalum in S. invicta invaded areas seems likely to indicate that colonies have changed their feeding preferences as a consequence of S. invicta invasion. The assessment of bacterial titers by qPCR indicated no significant difference in the bacterial content in the gut and gut tissues of T. melanocephalum workers between colonies from invaded and non-invaded areas. Rarefaction analysis of sequence reads from amplicon sequencing of bacterial 16S rRNA gene showed sufficient depth for analysis (S2 Fig). Of the total 841 bacterial OTUs present in T. melanocephalum from both areas (only 40 OTUs in invaded areas with abundance above 0. 01, versus 32 OTUs in non-invaded areas with abundance above 0. 01), 418 OTUs (32 OTUs with abundance above 0. 01) including the common insect endosymbiont Wolbachia were shared between the bacterial communities of T. melanocephalum collected from fire ant invaded and non-invaded areas (S2 Fig). This indicates that many OTUs present in T. melanocephalum from non-invaded areas are also present in T. melanocephalum from invaded areas. A large number of OTUs in T. melanocephalum from the invaded areas were unique, while some unique OTUs were also identified in colonies from non-invaded areas (S2 Fig). We compared the alpha diversity (Shannon index) of bacterial communities in each T. melanocephalum sample, and those from the invaded areas had a lower Shannon index than those from the non-invaded areas (S2 Fig) with no overlap in bacterial communities as shown by distinct clustering patterns in NMDS analysis (stress value = 0. 01, S2 Fig). These results indicate that differences in feeding preferences are also correlated with differences in gut microbiota in T. melanocephalum. Functional assignments were predicted from microbial community composition and structure using PICRUSt. Although this analysis, which is based on sequence similarities of the short 16S rRNA gene amplicons needs to be interpreted with caution [49], it revealed potential differences in predicted microbial function across invaded and non-invaded areas. Microbes associated with T. melanocephalum from invaded areas were different in an array of hypothetical metabolic functions from their counterparts found in non-invaded areas. Pathways for metabolic function, energy metabolism, metabolism of cofactors and vitamins, amino acid metabolism and nucleotide metabolism appeared enriched in T. melanocephalum from invaded areas (Fig 4A). Amino acid metabolism (that contributes to N cycling) may be the reason for higher δN values detected in T. melanocephalum from S. invicta invaded sites. We also used the LEfSe method to identify bacterial OTUs that were likely to explain most of the differences between the invaded and non-invaded sites. The bacterial orders of OTUs differed between colonies of T. melanocephalum from the two areas. Alphaproteobacteria were more abundant in T. melanocephalum from invaded sites, whereas Gammaproteobacteria and Bacilli were more abundant in T. melanocephalum from non-invaded sites (LDA scores > 4) (Fig 4B). Differences in OTUs mainly spanned two phyla and three classes, with the orders Lactobacillales, Rickettsiales (primarily Wolbachia) and Enterobacteriales (primarily Enterobacteriaceae) accounting for the majority of the differences (Fig 4B and S3 Fig). Based on 16S rRNA gene analysis, the Wolbachia found in T. melanocephalum is similar to Wolbachia strains found in Drosophila fruit flies (Fig 4C). We hypothesized that changes to the food supply of T. melanocephalum colonies might affect the relative abundances of different bacterial groups. Quantitative PCR assays undertaken with whole ant specimens unveiled a striking reverse effect on the abundance of Wolbachia, Lactobacillales and Enterobacteriaceae. Wolbachia abundance significantly decreased in workers of T. melanocephalum from S. invicta invaded sites (where they feed on a protein-rich diet) with sugar water as a carbohydrate-rich food in the laboratory, while Lactobacillales and Enterobacteriaceae abundances significantly increased (Wolbachia: F 4,10 = 125. 796, P < 0. 001; Enterobacteriaceae: F 4,10 = 536. 462, P < 0. 001; Lactobacillales: F 4,10 = 174. 642, P < 0. 001; Fig 5A). Conversely, Wolbachia abundance significantly increased when supplying workers of T. melanocephalum from S. invicta non-invaded sites (where they feed on a carbohydrate-rich diet) with only locusts as protein-rich food in the laboratory, while Lactobacillales and Enterobacteriaceae abundances significantly decreased (Wolbachia: F 4,10 = 1168. 171, P < 0. 001; Enterobacteriaceae: F 4,10 = 1348. 921, P < 0. 001; Lactobacillales: F 4,10 = 1021. 421, P < 0. 001; Fig 5A). For the reciprocal controls, colonies of T. melanocephalum ants from invaded and non-invaded areas were fed locusts and sugar water, respectively, and abundances of Wolbachia, Lactobacillales and Enterobacteriaceae were largely unchanged (in non-invaded sites, Wolbachia: F 4,10 = 3. 47, P = 0. 0503; Enterobacteriaceae: F 4,10 = 1. 019, P = 0. 443; Lactobacillales: F 4,10 = 0. 613, P = 0. 663; in invaded sites, Wolbachia: F 4,10 = 0. 543, P = 0. 708; Enterobacteriaceae: F 4,10 = 0. 303, P = 0. 87; Lactobacillales: F 4,10 = 1. 403, P = 0. 301, Fig 5B). These results indicate that the change in the diet of T. melanocephalum as a consequence of the presence of S. invicta affected the relative abundances of Wolbachia, Lactobacillales and Enterobacteriaceae in T. melanocephalum. This change appears to be plastic because it was reversed by supplying ants with different foods. Wolbachia has been shown to be associated with nutritional roles in other insect species, for example supplementation of B vitamins in bedbugs [50]. We measured the B vitamin contents in workers T. melanocephalum, and found that the concentrations of vitamin B2 and vitamin B3 were significantly decreased by supplying T. melanocephalum from S. invicta invaded areas with sugar (vitamin B2: F 4,10 = 37. 942, P < 0. 001; vitamin B3: F 4,10 = 17. 609, P < 0. 001), while the concentration of vitamin B1 and vitamin B12 were not affected (vitamin B1: F 4,10 = 2. 124, P = 0. 152; vitamin B12: F 4,10 = 1. 118, P = 0. 401; Fig 5C). The concentrations of vitamin B2 and vitamin B3 were significantly increased by supplying workers of T. melanocephalum from S. invicta non-invaded areas with locusts (vitamin B2: F 4,10 = 27. 355, P < 0. 001; vitamin B3: F 4,10 = 32. 297, P < 0. 001, Fig 5C). Concentrations of vitamin B2 and vitamin B3 could be recovered after the ants were fed with complementary food (sugar for the ants from S. invicta invaded areas and peptone for the ants from S. invicta non-invaded areas) (ants from S. invicta invaded area: vitamin B2: F 4,10 = 22. 492, P < 0. 001; vitamin B3: F 4,10 = 14. 523, P < 0. 001; ants from S. invicta non-invaded area: vitamin B2: F 4,10 = 17. 993, P < 0. 001; vitamin B3: F 4,10 = 43. 626, P < 0. 001; Fig 5D). The concentrations of vitamin B2 and vitamin B3 were not significantly affected by supplying the workers of T. melanocephalum from invaded and non-invaded areas with locusts and sugar, respectively (P > 0. 05 for all, S4 Fig). The concentrations of vitamin B2 and vitamin B3 in workers increased or decreased depending upon the abundance of Wolbachia. Wolbachia abundance was positively correlated with vitamin concentrations identified in T. melanocephalum in S. invicta non-invaded areas: Wolbachia abundance versus vitamin B2: R = 0. 99, P = 0. 001; Wolbachia abundance versus vitamin B3: R = 0. 898, P = 0. 039; in S. invicta invaded areas: Wolbachia abundance versus vitamin B2: R = 0. 924, P = 0. 025) expect vitamin B3 in T. melanocephalum from S. invicta invaded areas (R = 0. 756, P = 0. 14). The functional relationships of T. melanocephalum in the presence or absence of fire ants was assessed for seven further field sites that are more distant from each other than the core study sites. These additional field data showed that δN values of T. melanocephalum from four additional invaded sites (13 colonies) were significantly higher than those from three additional non-invaded sites (16 colonies) (Fig 6A). qPCR assays also revealed that the abundance of Wolbachia was significantly higher in workers from invaded areas than non-invaded areas (Fig 6B) and the abundance of Lactobacillales and Enterobacteriaceae was significantly higher in workers from non-invaded areas than invaded areas (Fig 6B). To directly test the effects of S. invicta on T. melanocephalum' s diet and microbiota, we collected three T. melanocephalum colonies from the wild and reared them in the lab either with or without S. invicta for one month. We then measured δN as well as abundances of Wolbachia, Lactobacillales, and Enterobacteriaceae. The δN values in T. melanocephalum were significantly increased after being reared in competition with S. invicta (Fig 6C) and the abundances of Wolbachia, Lactobacillales and Enterobacteriaceae were significantly affected (Fig 6D). Our study provides evidence that invasion by S. invicta may change the composition of the resident ant community. We tested this indirectly by comparing invaded and non-invaded sites since we were not allowed to infest uninvaded areas with S. invicta to elucidate this problem directly, and, therefore, other factors may also contribute to the findings. Some ant species do not exist in S. invicta invaded areas, whereas others, particularly T. melanocephalum have significantly higher numbers. This effect may result in part from reduced competition between T. melanocephalum and other ant species following S. invicta invasion [51]. However, T. melanocephalum workers inhabiting fire ant invaded areas also displayed submissive and avoidance behaviors when attacked by S. invicta workers. Similar behavioral or morphological changes can be found in other ants [52], frogs [53], plants [54], marine animals [55] and other terrestrial animals [56]. We also found that the stable isotope composition of T. melanocephalum workers differed between colonies located in the invaded and non-invaded areas. For δC, no differences were observed, while significant differences were detected for δN. Thus the higher δN value in T. melanocephalum from invaded areas suggests that the feeding habits of these ants may have changed as a consequence of the S. invicta invasion. To persist, T. melanocephalum may settle for different diets to avoid interspecific competition with S. invicta for nectar and insect honeydew. More variable food resources are expected to result in greater variation of δC values [57], and no significant difference was detected in δC for T. melanocephalum from the two different areas. In addition to diet differences, we found systematic differences in the composition and structure of bacterial symbiont communities in T. melanocephalum workers collected from S. invicta invaded and non-invaded areas. Although Wolbachia was present in both invaded and non-invaded sites, we found a larger number of Wolbachia sequence reads in T. melanocephalum workers from S. invicta invaded sites. Wolbachia are intracellular bacteria that exist mainly in the reproductive tissues (testis and ovary) of arthropod hosts so that they can manipulate the reproduction of their hosts. In various insects, Wolbachia has also been detected in other tissues, including in gut epithelial tissues and in accessory digestive glands such as salivary glands. Wolbachia has also been found to be important to fitness in some host species, and Wolbachia titres can increase in some stressed insects [58]. Wolbachia has been shown to be associated with nutritional roles in other insect species, e. g. supplementation of B vitamins in bedbugs [50]. In Drosophila, Wolbachia was shown to improve host fitness via metabolic provisioning during periods of nutritional stress [59]. Based on 16S rRNA gene analysis, the Wolbachia found in T. melanocephalum is similar to Wolbachia strains found in Drosophila fruit flies. Difference in Wolbachia titers may be linked to changes in B vitamin metabolism in T. melanocephalum from S. invicta invaded areas. Usually ants are attracted to feed on nectar and insect honeydew, which have high sugar content, and studies have indicated that Lactobacillales and Enterobacteriales have strong abilities to decompose sugar [60–62]. The observation of a higher abundance of Lactobacillales and Enterobacteriaceae in T. melanocephalum from S. invicta non-invaded areas is consistent with easier access to nectar and insect honeydew. Bacteria such as Wolbachia, Lactobacillales and Enterobacteriaceae have been identified by 16S rRNA gene amplicon sequencing studies in a variety of ants [63–65]. Different hypotheses have been investigated regarding their possible functions in the host, particularly those related to nitrogen metabolism [66]. Our experiments further indicate that the difference in relative abundances of Wolbachia, Lactobacillales and Enterobacteriaceae between T. melanocephalum from S. invicta invaded areas and non-invaded areas can be affected by the type of food available to colonies of T. melanocephalum. This is consistent with results from other studies showing that Wolbachia titers can be affected by diet [67–69], and that microorganisms are important for B vitamin metabolism in insects [70]. For example, high larval mortality occurred in species of lice in the genus Pediculus when symbiotic bacteria were removed, but this effect was reduced when the diet was supplemented with vitamin B3 (nicotinic acid) [71]. In aphids, symbionts have been shown to possess the biosynthetic pathway for the synthesis of vitamin B2 (riboflavin) [72]. The role in riboflavin synthesis has been supported by dietary experiments and genomic data are consistent with the expectation that microorganisms play a role in B vitamin provisioning to aphids [73]. The ecological effects of biological invasions have been well studied [74]. However, few investigations have examined the likely ecological and behavioral change to invasions by native and other co-existing resident species. In our study, we demonstrated that host–symbiont interactions of resident species may change in response to biological invasions, and that symbiotic bacteria may play a role in the adaptation of resident host taxa to invasive species. We found that bacterial communities associated with workers of T. melanocephalum in S. invicta invaded areas were distinct from those of T. melanocephalum in S. invicta non-invaded areas. However, additional genetic and/or functional studies of associated microbiota, such as experimental manipulation of ant communities in the field (to overcome spatial issues), are required to fully understand the microbial-host interactions that occur in T. melanocephalum following S. invicta invasion. We also found that the titer of Wolbachia is dependent on the nutritional status of its host. This has previously been observed for Drosophila [68] and is an important finding in the context of pest management strategies that rely on Wolbachia. Our field experiments and sample collections were carried out in three S. invicta invaded and three non-invaded sites in Guangzhou, China (S5 Fig). At each site, one plot (approximate 1,000 m2) was randomly selected for further investigation and sampling. This region has a humid, subtropical, monsoon climate, with 1,696 mm of annual rainfall, a minimum monthly average temperature of 21. 9°C in January, and a maximum of 28. 4°C in July [75]. The invaded and non-invaded sites are more than 3 km apart, and our continuous observations over more than five years confirmed that they are in zones that have either been invaded by S. invicta, or not been invaded by S. invicta. Significant genotypic differentiation of workers between sites demonstrate that the invaded and non-invaded sites contain different T. melanocephalum colonies [76]. The study areas have not been used for farming for more than five years [77] but carry many weeds, and are dominated by the weed Bidens pilosa L. To evaluate the potential impact of S. invicta invasion on the diversity of resident ant communities, ants were sampled from each plot between September and October 2015. Pitfall trapping and baiting were used to sample the ants in three invaded and three non-invaded sites. For trapping, a 100 mL centrifuge tube containing 40 ml of 45% alcohol was buried so that the opening of the tube was flush with the ground surface. In each plot, three traps were randomly set and left for 24 h. For baiting, ham sausage and honey were placed in a 30 ml transparent plastic bottle that was placed horizontally on the ground for 30–60 min. In each plot, three baits were randomly set between 8: 00–18: 00 h. The Simpson index (C), Shannon-Wiener index (H’) and Pielou evenness index (E) were calculated and compared between fire ant invaded and non-invaded areas. To determine whether coexisting colonies of T. melanocephalum change their behavior in response to fire ant invasion, individuals were collected from T. melanocephalum colonies in the S. invicta non-invaded sites (2 colonies were collected from GZ1,3 colonies from GZ2, S5 Fig), and individuals of S. invicta and T. melanocephalum colonies from the S. invicta invaded sites (all colonies were collected from GZ4, S5 Fig). The collected colonies were maintained in plastic nest boxes whose walls had been painted with Fluon in a temperature controlled room at 26°C and 80% humidity, and provided with sugar water (20% w/v) and locusts, Locusta migratoria manilensis (Meyen) as food every day. After rearing for one month, the ants were used for experiments. We quantified interspecific aggression between the two ant species using the following behavioral assay adapted from a previous report [78]. To test for interspecific individual aggression, one medium-sized (length, 4–5 mm) S. invicta worker and one T. melanocephalum worker were placed in a Petri dish (diameter = 4. 0 cm, height = 1. 5 cm) using a brush. Interactions were scored on a scale from 1 to 4, following protocols from an earlier study [52] and adapted for fire ants in this study: ants exhibited no change in direction or posture upon encounter or turned and moved away (Level I), ants made antennal contact that lasted for more than one second (Level II), ants opened their mandibles or turned their gasters upwards or towards their heads (Level III), both ants attacked each other and were twisted together, or one ant fiercely attacked the other with upper jaws grappling or stinging (Level IV). After five minutes, the ants’ attack scores and times were recorded, and an aggressiveness index calculated using the following formula ∑i=1nδifiT for each trial [52]. In this formula, δi and fi are the interaction score and frequency of each act, respectively, and T is the total interaction frequency, which is defined as the sum of all contacts between ants. Five pairs of colonies were tested. Ten trials, each involving different workers, were conducted for each pair of colonies. For group aggression experiments, ten medium-sized (length, 4–5 mm) workers of S. invicta and ten T. melanocephalum workers from colonies of fire ant invaded and non-invaded sites were randomly selected and placed in a Petri dish (diameter = 9 cm, height = 1. 5 cm, sides coated with Fluon) using a brush. Mortality was recorded after 0. 5 h, 1 h, 2 h and 4 h. Five pairs of colonies were tested, and three trials, each involving different worker samples, were conducted for each pair of colonies. Ants whose bodies were so damaged that they could not stand after the encounter were considered dead. To determine whether contact experience between S. invicta and T. melanocephalum influenced levels of aggression between the two species, a single worker of T. melanocephalum collected from the non-invaded area (workers from 2 colonies were collected at GZ1, and from 3 colonies at GZ2, S5 Fig) was placed in the Petri dish, and 1 min later, a worker of S. invicta (collected from GZ4, S5 Fig) was introduced to the same Petri dish. We removed the S. invicta worker when it became apparent that the two workers were going to fight. At intervals of 0. 5 h, 1 h, 2 h or 4 h later, we introduced another S. invicta worker and again tested the level of interspecific individual aggression for 5 min. Fifteen pairs of colonies were tested. Three trials, each involving different workers, were conducted for each pair of colonies. In order to investigate whether S. invicta invasion could affect the feeding habits of T. melanocephalum, the stable isotope composition of T. melanocephalum workers from both invaded and non-invaded sites was assessed. Workers of T. melanocephalum were collected from three sites within the area invaded by S. invicta (sites GZ4,5, 6, S5 Fig), and three sites within the area not invaded by S. invicta (sites GZ1,2, 3, S5 Fig) between September and October 2015. For each site, 200 ant workers were collected, pooled together and stored at -80°C. We removed the gasters from each worker of T. melanocephalum to prevent recent stomach contents from influencing δ15 N values [79]. The same species of plants (i. e. Bidens pilosa, Mimosa pudica, Ageratum conyzoides and Erigeron canadensis) from invaded and non-invaded areas were collected between September and October 2015. Three young leaves for each plant and eight plants from each site were randomly collected. We dried all samples at 60°C for 24–48 h. To prepare samples for isotopic analysis, samples were ground with a mortar and pestle, and 1 mg of each sample was packed into a tin capsule. An Isotope Ratio Mass Spectrometer (Thermo Fisher Scientific, Inc. , USA) was used to measure stable isotopes according the manufacturer’s instructions. Stable isotope abundance (δ) was calculated as follows [79]: δ (‰) = (RSaRSt−1) *1000; RSa is the detected value of the collected samples; RSt is the detected value of the standard sample. To determine whether the fine-scale geographic distribution of stable isotopes was the same between the invaded and non-invaded sites, the stable isotope composition of plants was compared. Bacterial symbiont communities of T. melanocephalum in the S. invicta invaded sites (sites GZ4,5, 6, S5 Fig) were compared with those in the non-invaded sites (sites GZ1,2, 3, S5 Fig) and the potential function of the symbionts was investigated. For each site, 100 ant workers were randomly collected from at least 10 colonies. Colony boundary aggression tests suggested that all 10 colonies were separate colonies [76]. Guts of workers from each site were pooled together in pure alcohol and stored at -80C. For each site, 100 worker guts were transferred into centrifuge tubes containing DNA extraction buffer. DNA was extracted using a DNA extraction kit (Tiangen biotech CO. , LTD, Beijing, China) following the manufacturer’s instructions. The bacterial 16S rRNA gene was amplified from the extracted DNA by PCR using two primers targeting the V3+V4 variable region of the 16S rRNA gene (16S-F: 5’-CCTACGGGNGGCWGCAG-3’, 16S-R: 5’-GGACTACHVGGGTATCTAAT-3’) [80]. qPCR was used to estimate the absolute content of bacterial DNA in the next generation sequencing samples by using universal bacterial 16S rRNA gene primers (see below). A standard curve for qPCR was generated by amplifying a 16S rRNA gene fragment of E. coli. Each sample was analyzed in a total reaction volume of 25 μL containing 2. 5 μL of Takara 10× Ex Taq buffer, 1. 5 μL of Mg2+ (25 mM), 2 μL of dNTPs (2. 5 mM), 0. 25 μL of Takara Ex Taq (2. 5 U/μL), 0. 5 μL of each primer (10 μM), 16. 75 μL of ddH2O and 25ng of template. Three PCR amplifications for each sample were performed with a 2 min incubation at 95C followed by 30 cycles at 94C for 30 s, 57C for 30 s, and 72C for 30 s, with a final 5 min extension at 72C. Each set of experiments included negative controls with sterile distilled water instead of template DNA. No amplified products were found in the negative controls. The PCR products were purified using a QIAGEN MinElute PCR Purification Kit to remove unincorporated primers and nucleotides. A micro-spectrophotometer ND-1000 (NanoDrop Technologies, Wilmington, DE, USA) was used to measure the concentration of the purified DNA. Adapters were added to the purified DNA to build a library for sequencing using the Illumina sequencing kit (www. illumina. com/company/legal. html) and an Illumina MiSeq sequencer (Illumina, San Diego, CA, USA). Amplicons were then pooled in equimolar fashion and paired-end sequenced (2 × 250) on an Illumina platform according to the standard protocols. For each sample, more than 50,000 reads were obtained. After sequencing, the data were filtered to remove reads containing more than 10% of unresolved nucleotides (N) and reads containing less than 80% of bases with a Q-value > 20. Paired end clean reads were merged as raw tags using FLSAH (v 1. 2. 11) with a minimum overlap of 10bp and mismatch error rates of 2%. Noisy sequences of raw tags were filtered by QIIME (V1. 9. 1) pipeline under specific filtering conditions to obtain the high-quality clean tags. Clean tags were searched against the reference database (http: //drive5. com/uchime/uchime_download. html) to perform reference-based chimera checking using UCHIME algorithm (http: //www. drive5. com/usearch/manual/uchime_algo. html). All chimeric tags were removed and finally obtained effective tags for further analysis. To obtain unique tags and to determine the number of tags in the dataset, the dataset was subjected to redundancy treatment using Mothur software (v. 1. 27. 0) [81]. Moreover, rarefaction curves were calculated by Mothur for all samples to evaluate the sequencing saturation. The representative sequences were classified into organisms by naïve Bayesian model using RDP classifier [82] based on GreenGene Database (http: //greengenes. lbl. gov/cgi-bin/nph-index. cgi). The tags were clustered into operational taxonomic units (OTUs) at ≥ 97% similarity using the UPARSE [83] pipeline. The tag sequence with highest abundance was selected as representative sequence within each cluster. The species annotations and abundance information of the OTUs were used to generate OTU abundance profiles for all samples. To determine the bacterial taxa that most likely explained differences between sites, we used the linear discriminant analysis (LDA) effect size (LEfSe) method (http: //huttenhower. sph. harvard. edu/galaxy/) [84]. Metagenomic data would be the best option for functional evaluation [85], but unfortunately such data sets are limited. Albeit there is a limited predictive power of 16S rRNA gene diversity for function of insect associated symbionts, we have used PICRUSt [49,86] to explore putative functions and pathways of OTUs against the KEGG database. Briefly, OTUs were normalized by copy number, and the gene categories were predicted at level 2 and level 3 KEGG orthology groups (KOs). The metagenomic prediction can produce the KEGG IDs and Enzyme Commission IDs. Unweighted unifrac distance matrix was generated by QIIME. Non-metric multidimensional scaling (NMDS) of unweighted Unifrac distances was calculated for the OTUs at phylum level and plotted in R with Welch' s t-test. Ordination was done in two dimensions. Ten iterations were performed, and the iteration resulting in the lowest stress was plotted. The 16S rRNA gene sequencing data was deposited in TSA database of NCBI (Accession number: PRJNA496064) The 16s rRNA gene sequence of the most abundant Wolbachia OTU was submitted to BLAST in NCBI (https: //blast. ncbi. nlm. nih. gov/Blast. cgi). With the top hits in BLAST, a neighbor-joining phylogenetic analysis was performed with MEGA 5. 0 [87]. The neighbor-joining (NJ) method was used to construct a phylogenetic tree based on the sequence of 16S rRNA gene and the phylogenetic tree was evaluated by Bootstrap analysis. To investigate whether symbiotic bacteria of T. melanocephalum workers were different between S. invicta invaded and non-invaded areas, 12 individual colonies of T. melanocephalum from S. invicta invaded and non-invaded sites were collected (2 colonies were collected from each site of GZ1-6 and were kept in boxes in the laboratory, S5 Fig). Colony boundary aggression tests suggested that all 12 colonies were separate colonies [76]. The experimental design was as follows: for S. invicta non-invaded sites three colonies were fed with sugar (a carbohydrate-rich diet), and three were fed with locusts (a protein-rich diet); for S. invicta invaded sites three colonies were fed with sugar, and three colonies were fed with locusts. Workers were randomly sampled from these colonies every 3 days for 12 days. The DNA of 15 workers per colony was extracted as one pooled sample using a DNA extraction kit (Tiangen, Beijing, China) following the manufacturer’s instructions. The absolute abundances of Wolbachia, Enterobacteriaceae and Lactobacillales bacteria were measured by real-time fluorescent quantitative PCR with designed specific 16S rRNA gene primers for Wolbachia (F: GCTGCAGTGGGGAATATTGG; R: TAACGCTAGCCCTCTCCGTA), Enterobacteriaceae (F: TATTGCACAATGGGCGCAAG; R: GGAGTTAGCCGGTGCTTCTT) and Lactobacillales (F: TATTGCACAATGGGCGCAAG; R: GGAGTTAGCCGGTGCTTCTT). Quantitative PCR analyses were performed for individuals that were fed different diets. PCR analyses were conducted on an Agilent Technologies Stratagene M×3005P by real-time quantitative PCR. Each treatment was measured in three separate technical replicates with a total reaction volume of 25 μL containing 0. 5 μL of each primer (diluted to 10 mM), 12. 5 μL SYBR Premix Ex TaqTM, 9. 5 μL ddH2O and 2 μL template. Cycling conditions were as follows: 95°C for 10 min and 40 cycles of 95°C for 30 s, 60°C for 45 s and 72°C for 1 min. A standard curve for qPCR was generated by amplifying the 16S rRNA gene of Escherichia coli as a representative bacterium of the family Enterobacteriaceae. For this purpose, E. coli was inoculated in LB liquid medium and cultivated for 2 days. Then, Colony-Forming Units (CFUs) of E. coli were calculated with a blood counting chamber. The DNA of E. coli was extracted from 2 ml culture. The extracted DNA was diluted (1×, 0. 1×, 0. 01×, 0. 001×) and the DNA dilution series was submitted to 16S rRNA gene qPCR. For each concentration, 3 replicates were done. To test whether availability and type of food source might affect colonies of T. melanocephalum inhabiting S. invicta invaded and non-invaded areas, workers collected in the previous step (workers from 2 colonies were collected from each site of GZ1-6, S5 Fig) were also sent to test the contents of vitamin B1, vitamin B2, vitamin B3 and vitamin B12 using an enzyme-linked immunosorbent assay (ELISA) kit (Shanghai, China, http: //www. mlbio. cn/? bdmlbio-12) according to the manufacturer’s instructions. For each of these colonies, 15 workers were randomly collected to measure their content of B vitamins. Moreover, the colonies fed with sugar for 12 days were provided with peptone as an alternative protein food source (to demonstrate that changes in abundance of Wolbachia and other bacterial DNA is not due to locust-associated bacteria when T. melanocephalum is feeding on locust), and the colonies fed with locusts for 12 days were provided with sugar, for another 12 days. Then the ants were sampled every 3 days. After the samples were collected, the content of vitamin B1 and vitamin B2 was measured. The content of vitamin B1 and vitamin B2 was also measured after the ants collected from the invaded sites and non-invaded site were fed with locusts and sugar, respectively. To further confirm the responses of bacterial symbionts of T. melanocephalum to S. invicta invasion, T. melanocephalum from another three non-invaded sites (16 colonies) and four invaded sites (13 colonies) were collected (colony boundary aggression tests suggested that all colonies were separate colonies, S6 Fig). Then, the δN values and the abundance of Wolbachia, Enterobacteriaceae and Lactobacillales were compared. To further confirm the different trophic patterns of T. melanocephalum in invaded and non-invaded areas are caused by the competition of S. invicta, we performed the competition tests in laboratory in order to control for spatial factors. Three colonies of T. melanocephalum (300 workers for each colony) from non-invaded sites were collected. The experiment was run in boxes divided into three rooms, two smaller rooms and one larger room (S7 Fig). One colony of T. melanocephalum and one colony of S. invicta (100 workers) were placed into the two smaller rooms. Meanwhile, sugar (a carbohydrate-rich diet) and locusts (a protein-rich diet) were randomly placed into the larger room. Ants were given access to either rooms via holes in the room walls. In this way, we were able to evaluate the impact of competitive pressure from S. invicta on the trophic patterns of T. melanocephalum. As control, T. melanocephalum was reared in a two room box (T. melanocephalum was placed in one room, sugar and locusts were randomly placed into the other room). In this case, we were able to simulate the situation of T. melanocephalum without the competition from S. invicta. After the colonies were reared for one month, the δN values (15 workers per sample) and the abundance of Wolbachia, Enterobacteriaceae and Lactobacillales were compared between treatment and control. Ant species diversity indices (C, H’, E) were calculated for all samples. Independent sample t tests were used to compare alpha diversity values and δN values between T. melanocephalum in invaded areas and non-invaded areas. One-way analysis of variance (ANOVA) followed by Tukey’s test for multiple comparisons was used to compare the bacterial abundance and vitamin B contents in T. melanocephalum after reared with different diets. δ15N values and the abundance of Wolbachia, Enterobacteriaceae and Lactobacillales in T. melanocephalum after rearing with or without S. invicta for one month were compared with paired sample t tests. For 16S rRNA gene sequences, we used the linear discriminant analysis (LDA) effect size (LEfSe) method.

Insects display a wide range of dependence on symbiotic bacteria for basic functions. Responses by resident species to selective pressures imposed by invasive species, as well as specific underlying mechanisms that give rise to these responses are still poorly understood. Here we investigate the role of the symbiotic bacteria of the ghost ant, Tapinoma melanocephalum, to changes in host behavior associated with interspecies competition in areas invaded by fire ants, Solenopsis invicta. We show that Wolbachia is significantly enriched in workers of T. melanocephalum from S. invicta infested areas, and that these bacteria also increase in abundance in colonies that have been supplied with protein-rich food. Our results suggest that bacterial symbiont communities can play an important role in enabling ants to tolerate changes in behavior and diet as a result of biological invasions.

Abstract Introduction Results Discussion Methods

b vitamins invertebrates species colonization ecology and environmental sciences medicine and health sciences chemical compounds invasive species sociology social sciences diet animals wolbachia organic compounds social systems riboflavin animal behavior nutrition enterobacteriaceae zoology bacteria animal sociality hymenoptera ants behavior chemistry vitamins insects arthropoda psychology eukaryota organic chemistry biology and life sciences physical sciences organisms

2019

Symbiotic microbiota may reflect host adaptation by resident to invasive ant species

Cholera remains an important public health problem. Yet there are few reliable population-based estimates of laboratory-confirmed cholera incidence in endemic areas around the world. We established treatment facility–based cholera surveillance in three sites in Jakarta (Indonesia), Kolkata (India), and Beira (Mozambique). The annual incidence of cholera was estimated using the population census as the denominator and the age-specific number of cholera cases among the study cohort as the numerator. The lowest overall rate was found in Jakarta, where the estimated incidence was 0. 5/1000 population/year. The incidence was three times higher in Kolkata (1. 6/1000/year) and eight times higher in Beira (4. 0/1000/year). In all study sites, the greatest burden was in children under 5 years of age. There are considerable differences in cholera incidence across these endemic areas but in all sites, children are the most affected. The study site in Africa had the highest cholera incidence consistent with a growing impression of the large cholera burden in Africa. Burden estimates are useful when considering where and among whom interventions such as vaccination would be most needed. Cholera is an acute, diarrheal illness caused by infection of the intestine with O1 or O139 serogroups of Vibrio cholerae [1]. Profuse watery diarrhea and vomiting can lead to dehydration and shock. Without treatment, death can occur within hours. Oral and intravenous rehydration therapy has markedly decreased case fatality rates [2], [3] but cholera remains a dreaded illness because of its rapid onset, severity, and potential to cause outbreaks that easily overwhelm public health systems in impoverished settings. Seasonal disease occurs in many less developed countries that cannot afford to establish or to maintain essential infrastructure for safe water supply and sanitation. Outbreaks may arise during natural disasters and complex emergencies. In 2006,52 countries officially reported a total of 236,896 cholera cases including 6,311 deaths with a CFR of 2. 7%, to the World Health Organization (WHO) [4]. These numbers do not reflect the true burden of cholera due to limitations in the surveillance and notification systems of many countries where the disease is endemic, as well as widespread underreporting because of fear of unjustified travel and trade-related sanctions [4], [5]. Hospital-based studies and outbreak reports provide important information, but they usually do not have a clear population denominator to allow estimation of the age-specific incidence of cholera in a community. A further challenge in the estimation of disease burden is the dearth of microbiology laboratories capable of detecting V. cholerae O1 and O139. Reported cases are often based solely on clinical diagnosis of the illness, adding further uncertainty. In the recent past, population-based information on the burden of culture-confirmed cholera cases has come principally from a single research institution in Bangladesh [6]. Policymakers from several developing countries have indicated that information on the age-specific burden of cholera is essential to decide the urgency of control strategies, including vaccination [7]. We established population-based surveillance for cholera in three study areas in Indonesia, India, and Mozambique in preparation for potential vaccine trials [8]–[10]. The areas were selected based on known endemicity of cholera and the presence of pre-existing research infrastructure or the potential to create such infrastructure. Using similar methodology in the three study areas allowed us to compare the overall and age-specific incidence of cholera across sites. Various findings from the three sites have been published elsewhere [8]–[10]. In this article, we compare and contrast the incidence of cholera across the sites. In each site, a catchment area was selected, census data were obtained, and surveillance for diarrhea was established in treatment centers serving the catchment population. A diarrhea episode was defined as passage of three or more loose or liquid stools in the 24 hour period prior to presentation for care. Repeat visits for the same episode of diarrhea, defined as three or fewer days apart between the end of the first episode and the onset of the second, were excluded. For every patient agreeing to participate, a case report form was completed and a rectal swab was obtained and inoculated into Cary-Blair transport media. The definition of diarrhea and laboratory methods was standardized across the study sites. Rectal swabs in Cary-Blair media were brought on the same day to the study laboratory and plated directly onto thiosulfate citrate bile salt sucrose (TCBS) agar (Eiken Chemical Company, Tokyo, Japan). The specimens were also incubated in alkaline peptone water (pH 8. 6) for 6 to 8 hours at 37°C then plated onto TCBS. After overnight incubation at 37°C, suspected colonies on the TCBS plates were tested biochemically and confirmed by agglutination with polyvalent O1 and monovalent Ogawa and Inaba antisera (Difco Laboratories, Detroit, Michigan). Non-agglutinating strains were tested with antiserum to V. cholerae O139 strain. Laboratory methods for isolation of V. cholerae were similar in each site. Isolation of V. cholerae was conducted in reference laboratories in two sites (the U. S. Naval Medical Research Unit No. 2 in Jakarta and the National Institute of Cholera and Enteric Diseases in Kolkata). In Beira, a consultant from the ICDDR, B: Centre for Health and Population Research, Dhaka, Bangladesh provided training and supervision to ensure standard procedures were followed. V. cholerae isolated in Beira were confirmed at the ICDDR, B. In Indonesia, the study area consisted of two adjacent districts (kecamatans), Tanjung Priok and Koja, in North Jakarta [8]. Residents live in homes that are temporary structures without running water and more than a third of households have no access to tap water. In 2001, the population in the catchment area was 160,257 (Table 1). Surveillance was conducted from August 2001 to July 2003 and included residents in the study site of all age groups who presented with diarrhea to participating health care providers: primary health centers (puskesmas) in Tanjung Priok and Koja, as well as the Infectious Disease Hospital and Koja Hospital. Rectal swabs were brought to the study laboratory for isolation of V. cholerae O1 and O139. The site in India consisted of legally registered urban slum areas (bustees) within administrative wards 29 and 30 in the city of Kolkata [9]. The area has a high population density and residents do not have sufficient water supply or sanitary facilities. A baseline census of the study population was done in early 2003 and was updated yearly. Surveillance was conducted from May 2003 to April 2005. The mid-year population of the study area was 58,063, based on two censuses, 12 months apart (Table 1). Surveillance included residents in the study site of all age groups who presented with diarrhea to any of the five project health outposts set-up in the field and two at the city' s infectious diseases and children' s hospitals, the main referral centers for diarrhea. Rectal swabs were brought to the study laboratory for isolation of V. cholerae O1 or O139. In Mozambique, the study area was Esturro, an impoverished urban neighborhood (bairro) in the city of Beira. From December 2003 to January 2004, healthy, non-pregnant residents of Esturro who were two years of age or older were invited to participate in a mass-vaccination campaign using a 2-dose recombinant cholera toxin B subunit, killed whole-cell oral cholera vaccine [10]. A baseline census in 2003 enumerated a total population of 21,818 persons in Esturro of whom 1,177 were less than two years of age and an estimated 5% (or 1,091 residents) were excluded because of potential pregnancy, leaving a target population of 19,547 persons (Table 1). About 57% of the study population received 2 doses and 72% received the first dose of vaccine [10]. As part of a case-control study, surveillance was conducted from January to December 2004 and included Esturrro residents who presented with diarrhea to the Beira Cholera Treatment Center. Pregnant women and children under two years of age were excluded from the surveillance. Rectal swabs were brought to the study laboratory for isolation of V. cholerae O1 and O139. The case-control analysis showed adjusted ORs for vaccine protection of 0. 16 from 2 doses and 0. 22 from at least one dose of vaccine [10]. Thus, the incidence of cholera was corrected (not accounting for herd immunity) according to the following formula: Where: Pv is the proportion of the target population that is vaccinated (received at least 1 dose) Pc is the proportion of the target population that is not vaccinated (received no vaccine) nv is the number of cholera cases detected among vaccinees during the first year nc is the number of cholera cases detected among non-vaccinees during the first year Nv is the number at baseline who were vaccinated (received at least 1 dose) Nc is the number at baseline who were not vaccinated (received no vaccine) Pout is the proportion who outmigrated between baseline and one year (data not available) OR is the odds ratio for vaccine protection (adjusted OR = 0. 22) In all study sites, case report forms were double-entered into data entry programs using FoxPro software (Microsoft, Redmond, WA). The data management programs included checks for error and consistency. We estimated the annual incidence of cholera using population as the denominator and the age-specific number of cholera cases among the residents of the study area as the numerator. The cohort under surveillance was dynamic and included all cases with culture-confirmed cholera residing in the catchment area. The surveillance was conducted following the principles governing biomedical research involving human subjects. In the Jakarta and Kolkata surveillance, verbal informed consent was obtained. This was considered as appropriate and sufficient since obtaining a history, physical examination, and stool specimen for culture of V. cholerae in cholera-endemic sites are part of good management of diarrhea patients [8], [9]. During the mass vaccination in Beira, written informed consent was obtained. During the case-control study in Beira that followed the mass vaccination, written informed consent was also obtained. Aside from a history, physical examination, and stool specimen for culture of V. cholerae, we asked each case for permission to visit his/her home (to recruit neighborhood controls) and we collected socio-behavioral data from the cases and controls [10]. The local ethics committees of each participating site, the WHO Secretariat Committee on Research Involving Human Subjects, and the International Vaccine Institute Institutional Review Board approved the study procedures and protocols. We compared the annualized incidence (per 1,000 population) of cholera across the study sites (Figure 1 and Table 1). Overall rates ranged from 0. 5 to 4. 0 cases/1,000 population/year. The lowest overall rate was found in Jakarta, where the estimated incidence was 0. 5/1,000/year. The incidence was three times higher in Kolkata (1. 6/1000/year) and eight times higher in Beira (4. 0/1000/year). The rates were highest in children under 5 years of age, with 8. 8/1,000,6. 2/1,000, and 1. 2/1000 among the 24 to 59 months old in Beira, Kolkata, and Jakarta, respectively. In the two sites where children under two years were also under observation (Jakarta and Kolkata), they were found to have even higher rates of cholera: 8. 6/1,000 in Kolkata and 3. 2/1,000 in Jakarta. Only V. cholerae O1 was isolated at all sites. We found that young children bear the greatest burden of cholera. Cholera has traditionally been considered to occur infrequently in young children, and consequently, the WHO recommends that cholera should be suspected among those over two years of age who have acute watery diarrhea and severe dehydration if cholera is endemic in the local area [11]. Aside from our data, two other studies have shown that cholera is a significant problem in young children [6], [12], but neither provide population-based incidence. Our findings have implications for the enhanced benefit of cholera vaccination targeting specific age-groups in cholera-endemic areas. Protecting children against cholera may not only decrease the burden in this age group but decrease transmission of the disease to their family members and the community [13]. Immunization of adult women with killed oral cholera vaccines has been shown to confer herd immunity against cholera to children too young to be vaccinated [14]. Our comparison shows that the overall rates of cholera cases presenting for treatment varied widely across the study sites in three different countries, with the highest incidence in the African site. These findings add to the growing impression of the large cholera burden in Africa. In 2006, Africa reported 234,349 cholera cases to the WHO, accounting for 99% of the officially-notified global cholera [4]. Between 1995 and 2005,66% of cholera outbreak reports to ProMed came from sub-Saharan Africa [15]. It has been suggested that the number of individuals at risk for cholera may be higher in Asia than in subSaharan Africa because of the higher population density in the former. But since cholera-endemic areas are likely to be more widespread in subSaharan Africa as evidenced by the officially reported cases [4], then the number of individuals at risk could potentially be higher in this continent. Culture-confirmation showed that all isolates were V. cholerae O1. Previously, only V. cholerae serogroup O1 caused epidemic cholera. In late 1992, large outbreaks of cholera began in India and Bangladesh that were caused by a previously unrecognized serogroup of V. cholerae, designated O139, synonym Bengal. V. cholerae O1 had since been isolated in 11 countries in South-East Asia but less frequently in recent years. In 2006,120 laboratory-confirmed V. cholerae O139 cases were reported from mainland China and 3 from Thailand but from no where else [4]. The expectation that V. cholerae O139 would become more widespread has apparently not occurred so far. There are limitations to our data. First and foremost the surveillance sites were arbitrarily selected based on known cholera endemicity and may not be representative of the region. Although sentinel surveillance is the most reliable approach to collect comparable surveillance data between continents, data have to be interpreted with this limitation in mind. Second, there may have been variations in the intensity of surveillance despite similar case capture procedures across sites. Third, the size of the catchment population at each site differed considerably, ranging from 20,000 in Beira to 60,000 in Kolkata and 160,000 in Jakarta. Furthermore, the estimate in Beira had to be corrected for the direct protection from cholera vaccination. The true disease incidence may be even higher considering the herd (indirect) protection conferred by the vaccine [13]. Fourth, the data is based on study periods of one year in Beira and two years in Kolkata and Jakarta. Cholera incidence may vary over several years. It is possible that Beira was having an unusually high number of cases that year. Longer-term surveillance could provide additional important information but surveillance for long periods would be complicated by population mobility and consequent variations in the denominator. Fifth, we are reporting cholera incidence in areas with seasonal cholera, where the population is likely to have immunity from previous exposures. Attack rates during outbreaks at the time of natural disasters and complex emergencies occurring among populations without previous exposure to cholera are likely to be higher. Finally, passive surveillance in all three sites could only detect those cases which were perceived to be severe enough to require medical care and although we implemented methodology in each study site to optimize surveillance, some cases may have escaped our detection system. For example, health care utilization patterns may have influenced the relative differences in detected cholera between age groups. Young children with diarrhea may be more frequently taken for health care treatment compared to adults. It is possible that patients, particularly adults, escaped our detection system and that active surveillance could have identified more cases. However, it was not our intention to find all diarrhea cases in the community but only those perceived to be severe enough to require medical care and thus burdening the public health service. This is to our knowledge the first comparison of the incidence of culture-confirmed treated cholera cases in endemic areas using standardized methods. In all our study sites, rates were highest in young children indicating the need to revisit the standard guidelines for clinically suspecting cholera. V. cholerae O139 was not detected in the study sites. We found considerable differences in the burden of disease in the three sites, with the African site having a multi-fold higher incidence compared to the Asian sites. Incidence of cholera is important when considering where and among whom interventions such as vaccination would be most beneficial.

Cholera is an often forgotten disease affecting the world' s forgotten people. When a large cholera outbreak occurs, the disease appears briefly on the radar of public attention. Some unfortunate populations around the world suffer recurrent episodes of cholera but their plight goes unnoticed. We established cholera surveillance in impoverished areas in Jakarta (Indonesia), Kolkata (India), and Beira (Mozambique) where the disease is known to occur regularly. The cholera burden was calculated using the site population as the denominator and the number of cholera cases as the numerator. The lowest overall rate was in Jakarta with 0. 5 cases per 1000 population per year. The incidence was three times higher in Kolkata (1. 6/1000/year) and eight times higher in Beira (4. 0/1000/year), adding to the growing impression of the large cholera problem in Africa. In all sites, children are the most affected. Estimates such as these are useful when considering where and among whom interventions against the disease are most needed. Improvement of water supply and sanitation is the best strategy against cholera and other diarrheal diseases but may not be achievable in these impoverished areas in the near future. Other immediate, short- to medium-term strategies such as vaccination against cholera may be useful.

Abstract Introduction Methods Results Discussion

public health and epidemiology pediatrics and child health

2008

The High Burden of Cholera in Children: Comparison of Incidence from Endemic Areas in Asia and Africa

Loop-mediated isothermal amplification (LAMP) of DNA is a novel technique that rapidly amplifies target DNA under isothermal conditions. In the present study, a LAMP test was designed from the serum resistance-associated (SRA) gene of Trypanosoma brucei rhodesiense, the cause of the acute form of African sleeping sickness, and used to detect parasite DNA from processed and heat-treated infected blood samples. The SRA gene is specific to T. b. rhodesiense and has been shown to confer resistance to lysis by normal human serum. The assay was performed at 62°C for 1 h, using six primers that recognised eight targets. The template was varying concentrations of trypanosome DNA and supernatant from heat-treated infected blood samples. The resulting amplicons were detected using SYTO-9 fluorescence dye in a real-time thermocycler, visual observation after the addition of SYBR Green I, and gel electrophoresis. DNA amplification was detected within 35 min. The SRA LAMP test had an unequivocal detection limit of one pg of purified DNA (equivalent to 10 trypanosomes/ml) and 0. 1 pg (1 trypanosome/ml) using heat-treated buffy coat, while the detection limit for conventional SRA PCR was ∼1,000 trypanosomes/ml. The expected LAMP amplicon was confirmed through restriction enzyme RsaI digestion, identical melt curves, and sequence analysis. The reproducibility of the SRA LAMP assay using water bath and heat-processed template, and the ease in results readout show great potential for the diagnosis of T. b. rhodesiense in endemic regions. Human African trypanosomiasis is endemic in tropical Africa. In eastern and southern Africa the disease is caused by Trypanosoma brucei rhodesiense, while T. b. gambiense infections are common in central and West Africa. T. b. rhodesiense causes an acute form of disease, whereas T. b. gambiense causes a more chronic form. Moreover, the treatment regimen for the two infections is different, expressing the need for a specific diagnostic test for each trypanosome. The geographical demarcation of T. b. rhodesiense and T. b. gambiense to a large extent forms the basis of trypanosome identification and treatment. In East Africa the introduction of T. b. rhodesiense into the T. b. gambiense region is certain to occur due to the closeness of the two disease foci and continuous movement of the livestock-reservoir host for T. b. rhodesiense. This prospect further obligates the development of test kits that can differentiate the two parasites. The serum resistance-associated (SRA) gene [1], [2] is conserved and specific to T. b. rhodesiense [3]–[5] and therefore provides unequivocal identification of this parasite. It is a low-copy gene, therefore the polymerase chain reaction (PCR) test is inadequate to amplify this target reliably in clinical samples without recourse to parasite multiplication in mice. Besides, available molecular methods of parasite detection require elaborate precision instruments [3]–[7], which make their use under field conditions unfeasible. There is therefore a need for a simplified method of amplification and product detection that would compliment the available tests and make feasible molecular diagnosis for case detection and confirmation of cure in the regions that are endemic for sleeping sickness. Recently, a technique called loop-mediated isothermal amplification (LAMP) of DNA has been developed [8]. The technique uses four to six primers that recognise six to eight regions of the target DNA, respectively, in conjunction with the enzyme Bst polymerase, which has strand displacement activity. The simultaneous initiation of DNA synthesis by multiple primers makes the technique highly specific. The LAMP test is carried out under isothermal conditions (60–65°C) and produces large amount of DNA [8]. The reaction shows high tolerance to biological products [9], meaning that DNA extraction may not be necessary [10], and the product can be inspected visually by the addition of SYBR Green I [11], [12]. Briefly, LAMP proceeds when the forward inner primer (FIP) anneals to the complementary region (F2c) in the target DNA and initiates the first strand synthesis, and then the outer forward primer (F3) hybridises and displaces the first strand, forming a loop structure at one end [8]. This single-stranded DNA serves as template for backward inner primer (BIP) -initiated DNA synthesis and subsequent outer backward (B3) -primed strand displacement DNA synthesis, leading to the formation of dumbbell-shaped DNA structures [8]. The stem-loop thus formed acts as a template, and subsequently one inner primer hybridises to the loop on the product and initiates the displacement DNA synthesis, forming the original stem loop and a new stem loop that is twice as long [13]. The final products are stem-loop DNAs with several inverted repeats of the target DNA, and cauliflower-like structures bearing multiple loops [8]. A number of LAMP tests to detect parasitic protozoa have been designed and used successfully [14]–[16]. The rapidity, specificity, and simplicity of the technique make it appealing for use in trypanosomiasis-endemic regions. The purpose of the present study was to develop a LAMP test for detection of T. b. rhodesiense based on the SRA gene and compare it with PCR test that is specific for T. b. rhodesiense. Our results indicate that the SRA LAMP is sensitive and specific and has the potential to be developed into a field-friendly diagnostic test. Institutional Ethical Clearance for the collection of human samples had been obtained from the Livestock Health Research Institute (LIRI), Tororo, Uganda, and the Uganda National Council of Science and Technology (UNCST), Kampala, Uganda, which records and regulates all research activities in the country. At Murdoch University, Perth, Australia, the use of mice was approved by Murdoch University Animal Ethics Committee (AEC). The trypanosome DNA samples used in this study are shown in Table 1. The samples which most had been passaged in mice were chosen to ensure a wide geographical representation, different times of isolation, and hosts (Table 1). Six samples designated as JE (three each from blood and cerebrospinal fluid [CSF]) were direct isolates from human hosts. The DNA had been prepared using several methods (see footnotes in Table 1). The samples for studying analytical sensitivity and tolerance of LAMP were obtained from the blood of mice infected with T. b. rhodesiense and divided into two portions. The first portion was centrifuged at 3,000 rpm for 10 min and the buffy coat was collected, and the second portion was divided into aliquots of 10 µl. Then each of the two portions was mixed with 40 µl of ultrapure water, boiled for 3 min, and centrifuged at 14,000 g for 5 min. Samples of 10–15 µl of supernatant were recovered and stored at −20°C for later use. Trypanosomes belonging to the subgenus Trypanozoon were analysed using TBR1 and 2 primers [7]. Furthermore T. b. rhodesiense was detected by a PCR specific for the SRA gene [3]), whereas T. b. gambiense was detected using a PCR for the T. b. gambiense-specific glycoprotein (TgsGP) gene [17]. LAMP reactions of 25 µl were standardised for optimal reagent concentrations, temperature, and time conditions using T. b. rhodesiense isolate LVH 56 and following the Taguchi design [18]. Briefly, the FIP and BIP were varied from 0. 8 µM to 2. 4 µM, dNTPs from 100 µM to 400 µM, betaine from 0. 2 M to 0. 8 M, and MgSO4 from 0 to 4 mM. The FIP, BIP, F3, and B3 primers were designed using the PrimerExplorer v3 software (http: /primerexplorer. jp/lamp) based on the SRA gene sequence (GenBank accession number Z37159) (Table 2). Loop primers [loop forward (LF) and loop backward (LB) ] were designed manually. The reactions were optimised at 2. 0 µM for FIP and BIP primers, 0. 8 µM for loop primer (LF and LB), 0. 2 µM for F3 and B3 outer primers, 200 µM for each dNTP, 0. 8 M betaine (Sigma), 20 mM Tris-HCl (pH 8. 8), 10 mM KCl, 10 mM (NH4) 2SO4,2 mM MgSO4,0. 1% Triton X-100, and 8 U of Bst DNA polymerase large fragment (New England Biolabs). For real-time reactions 3. 34 µM SYTO-9 fluorescence dye (Molecular Probes) was added. The template was ∼100 pg for trypanosome lysate DNA samples and 2 µl of buffy coat and supernatant prepared from boiled blood. To find the optimum temperature for the LAMP test, the reactions were carried out for 1 h at 58,60,62, and 64°C using the Rotor-Gene 3000 thermocycler (Corbett Research) or in a water bath at the same temperature settings. The reaction was terminated by increasing the temperature to 80°C for 4 min. Three methods were used to analyse DNA amplification, and included electrophoresis in 1. 5% agarose gels stained with ethidium bromide, direct visual inspection of the LAMP product after addition of 1 µl of 1/10 dilution of SYBR Green I (Invitrogen), and by monitoring fluorescence of the double-stranded DNA (dsDNA) -specific dye SYTO-9 [19] in a Rotor-Gene 3000 thermocycler. Real-time fluorescence data was obtained on the FAM channel (excitation at 470 nm and detection at 510 nm) [19]. Three approaches were used to confirm that the SRA LAMP test amplified the correct target: (1) the product was digested with restriction enzyme RsaI (New England Biolabs) at 37°C for 3 h, followed by electrophoresis in 3% agarose gel; (2) following amplification, the DNA melting curves were acquired on the FAM channel using 1°C steps, with a hold of 30 s, from 62 to 96°C [19]; and (3) some of the LAMP amplicon bands were excised from an agarose gel and cloned into a TOPO-TA vector (Invitrogen), transformed in E. coli and inserts sequenced using an automated DNA 3730 analyser (Applied Biosystems). The resulting sequences were aligned with the target sequence using the DNAman computer software version 5. 0 (Lynnon Biosoft). 10-fold dilutions were made from infected mouse blood containing 1. 0×106 trypanosomes/ml and from 100 ng of purified T. b. rhodesiense DNA, and used to determine the analytical sensitivity of SRA LAMP and PCR tests. The reactions were done in triplicates and repeated after 2 wks. The LAMP test was carried out using both cold and heated templates. The specificity of the tests were assessed with DNA from human, tsetse fly, bovine, Plasmodium falciparum, and trypanosomes belonging to other species (Trypanosoma brucei brucei, T. b. gambiense, T. b. evansi, Trypanosoma congolense savannah, T. c. kilifi, T. c. forest, Trypanosoma simiae, T. s. tsavo, Trypanosoma godfreyi, Trypanosoma vivax, and Trypanosoma lewisi). The results of the SRA LAMP assay are shown in Figures 1–4 and Table 1. When the test was carried out without loop primers a product was detected after 50 min. The inclusion of loop primers reduced the reaction time from an average of 50 min down to between 20 and 25 min and increased the sensitivity 100-fold. The best results were obtained when the reaction temperature was maintained at 62°C. All the positive LAMP reactions produced a characteristic ladder of multiple bands on an agarose gel (Figure 1A and 1B), indicating that stem-loop DNA with inverted repeats of the target sequence was produced. Positive reactions turned green on addition of SYBR Green I, while the negative ones remained orange (Figure 3). RsaI restriction enzyme digestion and electrophoresis gave the predicted sizes of 90 bp and 114 bp (Figure 1B). The SRA LAMP amplicons showed reproducible melt curves with a Tm of ∼87. 5°C, suggesting amplicons of the same sequence (Figure 4). The cloned sequence showed 100% identity with the target sequence, and revealed that the length varied with sequence repeats of primers and there complementary sequences. The analytical sensitivity of SRA LAMP assay improved from a dilution of 10−4 to 10−6 when a template (DNA or supernatant) was preheated before being added to a reaction (Figure 2), with the best detection limit of dilution 10−7 recorded with supernatant prepared from the buffy coat. The classical PCR based on the same gene [3] showed a detection limit of dilution 10−4. The SRA LAMP detected all the 49 (100%) T. b. rhodesiense (including the six samples isolated directly from HAT patients), while TBR1 and 2 primers detected 39 out of 46 (84. 8%) and SRA PCR detected 31 out of 46 (67. 4%) samples (Table 1). The SRA LAMP test was specific and no cross-reaction was recorded with nontarget DNA. In the present study we were able to demonstrate the successful amplification of T. b. rhodesiense DNA within 20–25 min at 62°C using the SRA LAMP assay. However, we set the optimal time at 35 min to amplify DNA at low concentrations. The results of the SRA LAMP assay were identical when either a water bath or a thermocycler was used to maintain the temperature at 62°C, demonstrating its robustness. Preheating of the template increased the efficiency of the assay by shortening the duration (Figure 2) and increasing sensitivity of the test. DNA amplification is preceded by strand separation under isothermal conditions using betaine, which destabilises the DNA helix [8]. It would appear that preheating of the sample produced a faster and/or a greater amount of strand separation, which translated into a far more rapid assay. All positive samples detected by gel electrophoresis or in real-time using SYTO-9 fluorescence dye could also be detected visually by addition of SYBR Green I to the product. This ability highlights another advantage of LAMP technique: the results of amplification can visually be observed through addition of a DNA intercalating dye (Figure 3), eliminating the need for gel electrophoresis and greatly reducing the time taken for result analysis. When pure trypanosome DNA was used, the detection limit of the SRA LAMP test without loop primers was an equivalent of 1,000 trypanosomes/ml. This limit was improved to an equivalent of one trypanosome/ml with the inclusion of loop primers. Increased sensitivity and reduction in LAMP reaction time with the addition of loop primers is well documented [20] and has been demonstrated in detection of Mycobacterium [11], periodontal pathogens [12], and Plasmodium falciparum malaria [10]. Loop primers accelerate the LAMP reaction by hybridising to the stem-loop region, initiating further DNA amplification [20]. When different templates were used, heat-treated buffy coat from mice blood performed better than the supernatant obtained after boiling blood samples. The higher sensitivity recorded could be the effect of concentrating the parasites in the buffy coat through centrifugation; therefore, buffy coat seems a superior template for SRA LAMP test. The robustness of the LAMP test is demonstrated by the ability to amplify target DNA from various templates without the expensive and time-consuming process of DNA purification. We observed no inhibitory effects in using 2–5 µl of supernatant in a 25 µl reaction or an increase in sensitivity beyond 2 µl, indicating that this volume was the optimal for our samples. The possibility of using heat-processed samples without compromising sensitivity eliminates the need for DNA extraction and further shortens the LAMP reaction. Other studies have shown superior tolerance of LAMP tests for biological substances [9], [13] and heat processed blood has been used successfully in detection Malaria [10]. The method of template preparation for use in LAMP tests, however, needs to be further developed. The potential usefulness of SRA LAMP is confirmed by its ability to detect T. b. rhodesiense directly from parasitaemic and apparently aparasitaemic clinical samples (human blood and CSF). The human blood (JE2 and JE3) and CSF samples JE8–JE10 used in the present study were negative by microscopy at the time of sampling. Parasites were demonstrated only following inoculation of the samples in mice. When the samples were tested, they were positive by SRA LAMP assay while only JE4, JE9, and JE10 were positive using TBR PCR (Table 1) [7]. Detection of aparasitaemic samples demonstrates one of the practical values of SRA LAMP in sleeping sickness diagnosis-time-consuming parasite multiplication assays in mice are unnecessary, and early diagnosis increases the chances of cure after treatment. In the present study, amplification of the target sequence was confirmed by restriction enzyme digestion using RsaI, melting curve analysis, and sequence analysis. It is important to distinguish T b. rhodesiense and T. b. gambiense since the two parasitic infections have different treatments. In recent years the T. b. rhodesiense region in Southern Uganda has been expanding towards the T. b. gambiense focus as a result of livestock movement [21], [22]. There is therefore a need to continue development of rapid and sensitive techniques to differentiate the two parasites and to compliment the available PCR tests, and to this end the SRA LAMP assay has shown great potential for this application. The LAMP test should theoretically not amplify nontarget sequences, since the specificity is enhanced by using a set of six primers. However there is a high risk of amplicon contamination since the tubes have to be opened to add the dye. Analysis of any false positive reactions through sequencing and restriction enzyme analysis would easily distinguish between false positive and contamination. To reduce the chances of contamination, similar protocols to those followed for PCR are required. However, the great potential for LAMP is that reactions can be performed and results read without opening tubes [23]. On this end, more work is needed to develop such a closed reaction system for diagnosing sleeping sickness. This study has shown that the SRA LAMP assay could be developed into an assay for T. b. rhodesiense that is simple to use at point of care. The detection of the equivalent of one trypanosome/ml in the buffy coat (with the possibility of reducing this further to 0. 1 trypanosomes/ml) compares well with the normal parasitaemia in humans. Since DNA amplification and reading of results require minimum equipment, the technique has great potential for use in the HAT-endemic countries as back-up test for other HAT tests currently in use.

Control of human African trypanosomiasis (HAT) or sleeping sickness relies on diagnosis and treatment of infected patients. However, the diagnostic tests in routine use have limited sensitivity, due to a characteristically low parasitaemia in infected individuals. Differentiation of infections by Trypanosoma brucei rhodesiense (causes acute disease) and T. b. gambiense (causes chronic disease) is essential, as the two forms of disease have different treatment regimens. In the present work, loop-mediated isothermal amplification (LAMP) of DNA was successfully used to detect T. b. rhodesiense, with a sensitivity of up to one trypanosome/ml of blood. The LAMP test was efficient and robust, and results were obtained within 35 min. Amplification was possible when a water bath was used to maintain the temperature at isothermal conditions (60–65°C), and results could be read by visual observation of colour change. These findings have increased the prospects for developing a simple molecular test for HAT that can be used with limited equipment at point of care in endemic rural areas.

Abstract Introduction Materials and Methods Results Discussion

molecular biology

2008

Loop-Mediated Isothermal Amplification (LAMP) Method for Rapid Detection of Trypanosoma brucei rhodesiense

Anthrax lethal toxin (LT) is a bipartite protease-containing toxin and a key virulence determinant of Bacillus anthracis. In mice, LT causes the rapid lysis of macrophages isolated from certain inbred strains, but the correlation between murine macrophage sensitivity and mouse strain susceptibility to toxin challenge is poor. In rats, LT induces a rapid death in as little as 37 minutes through unknown mechanisms. We used a recombinant inbred (RI) rat panel of 19 strains generated from LT-sensitive and LT-resistant progenitors to map LT sensitivity in rats to a locus on chromosome 10 that includes the inflammasome NOD-like receptor (NLR) sensor, Nlrp1. This gene is the closest rat homolog of mouse Nlrp1b, which was previously shown to control murine macrophage sensitivity to LT. An absolute correlation between in vitro macrophage sensitivity to LT-induced lysis and animal susceptibility to the toxin was found for the 19 RI strains and 12 additional rat strains. Sequencing Nlrp1 from these strains identified five polymorphic alleles. Polymorphisms within the N-terminal 100 amino acids of the Nlrp1 protein were perfectly correlated with LT sensitivity. These data suggest that toxin-mediated lethality in rats as well as macrophage sensitivity in this animal model are controlled by a single locus on chromosome 10 that is likely to be the inflammasome NLR sensor, Nlrp1. Anthrax lethal toxin (LT), a major virulence factor of Bacillus anthracis, is composed of two proteins, lethal factor (LF) and protective antigen (PA). PA binds to cellular receptors and facilitates LF entry into the cytosol (for review see [1]). LF is a protease which cleaves and inactivates members of the mitogen-activated protein kinase kinase (MAPKK or MEK) family, resulting in proliferation arrest in most cell types and a unique, rapid (<90 min), caspase-1 dependent lysis of mouse macrophages from certain inbred strains through poorly characterized mechanisms (for review see [2]). In mouse macrophages, sensitivity to LT-mediated lysis is a dominant trait that maps to the highly polymorphic Nlrp1b (Nalp1b) gene on chromosome 11 [3]. Mouse Nlrp1b (mNlrp1b) has five alleles that correlate with LT sensitivity or resistance in macrophages, and it is one of three tandem Nlrp1 paralogs on chromosome 11 [3]. mNlrp1b, the paralog controlling LT sensitivity, is a NOD-like receptor (NLR) which, when activated, leads to assembly of the inflammasome, a multiprotein complex responsible for the activation of caspase-1 [4]. The mNlrp1b inflammasome-mediated activation of caspase-1 is necessary for murine macrophage cell death in response to LT [3], [5]–[7]. Furthermore, expression of mNlrp1b from LT-sensitive macrophages together with caspase-1 is sufficient to render other cell types sensitive to the effects of LT [8]. Unlike human Nlrp1 (hNlrp1), mNlrp1b (despite the acronym representing NLR family, pyrin domain containing 1b) lacks an N-terminal pyrin domain. The pyrin domain is required for hNlrp1 binding to the inflammasome adaptor protein ASC, and mNlrp1b is not believed to interact with this adaptor [9]. However, mNlrp1b does have the NACHT (nucleotide oligomerization), LRR (leucine-rich repeat), and CARD (caspase recruitment) domains commonly found in NLR proteins (for recent reviews see: [10], [11]). It is unclear how polymorphisms in the mNlrp1b protein result in such striking variation in the ability of LT to activate caspase-1 (and subsequently induce cell death) in murine macrophages. While the mNlrp1b inflammasome requirement for murine macrophage death in response to LT is well established, it is unclear if this inflammasome is involved in LT-mediated death of animals. LT injection into rodents induces an atypical vascular collapse, replicating the shock state associated with anthrax disease [12]–[14]. Susceptibility in mice, however, is controlled by multiple loci [15], and macrophage sensitivity does not control animal susceptibility to LT [16]. Furthermore, factors such as adrenal function can also modulate LT toxicity in mice [17]. Thus, the molecular basis for the death induced by LT in mice is currently unknown. The rapid LT-mediated death of the Fischer rat [18] can occur in as little as 37 minutes through a unique vascular shock [19], [20]. In rats, left ventricular failure accompanied by a rapid accumulation of pleural fluid (a hallmark of anthrax disease) is typically associated with LT-mediated death [12], [21], [22]. In contrast, LT-induced murine death occurs by vascular collapse over a longer period of days [13], [15]. Early targeting of cardiac function by LT has also recently been demonstrated in mice [23]. The role of MEK cleavage and/or inflammasome components in vascular collapse induced by LT in rodents is also currently unknown. Thus, studies of determinant molecular pathways would be greatly assisted if genomic targets controlling susceptibility were identified. Strain-specific variations in macrophage and animal sensitivity to LT were previously noted for four rat strains [24]. Toxicity testing of first filial (F1) progeny from crosses of LT-sensitive (Brown Norway and F344) and resistant (Lewis and Wistar Kyoto) strains led to the conclusion that toxin sensitivity exhibited a dominant mode of inheritance [24]. Further, the authors concluded that the inter-cross results were consistent with LT sensitivity in rats being determined by a single, dominant gene [24]. In the current report, we used the HXB/BXH recombinant inbred (RI) rat collection, developed by two gender-reciprocal matings of the Wistar Kyoto-related strain, the SHR/Ola rat (an LT-resistant rat) with a Brown Norway congenic (BN-Lx, an LT-sensitive rat) [25], [26] as an ideal genetically-derived animal model to map LT sensitivity of rats. We report that susceptibility of rats to anthrax LT maps to a single locus on chromosome 10 that contains the Nlrp1 (rNlrp1) gene. Furthermore, LT sensitivity of a large number of rat strains was found to perfectly correlate with their macrophage sensitivity to toxin. Sequence analysis of rNlrp1 from twelve rat strains identified specific variations within a limited 100-aa N-terminal region of rNlrp1 that correlate perfectly with LT sensitivity. Taken together, these data suggest that a single locus on chromosome 10, likely rNlrp1, controls both rat macrophage sensitivity to anthrax LT as well as rat death in response to this toxin. Twelve inbred rat strains and their bone marrow-derived macrophages (BMDMs) were tested for sensitivity to anthrax LT (Figure 1A). Both the rats and their corresponding BMDMs exhibited a qualitative dichotomous phenotype. Thus, macrophages were either sensitive or resistant to toxin, and rats showed an identical pattern, either succumbing within 60 minutes or remaining completely resistant to systemic toxin treatment (Figure 1A). For all the inbred rat strains, sensitivity of BMDMs was predictive of animal susceptibility to toxin. This result is notably different from what was previously observed in a comparison of mouse strains, where correlation between mouse strain susceptibilities to LT and their macrophage sensitivities was not found [15]. We and others have proposed that multiple genetic loci control LT susceptibility in mice [15], [27]. The absence of intermediate sensitivities in the rats supported the possibility that LT sensitivity in rats is controlled by a single gene [24]. This fact suggested that a recombinant inbred (RI) rat strain panel derived from LT-sensitive and LT-resistant progenitors could be used to map the LT susceptibility locus. RI strains allow linking of allelic variation at specific chromosomal loci to particular phenotypes. The widely used HXB/BXH RI rat collection was developed by two gender-reciprocal matings of the hypertensive SHR/Ola rat (an LT-resistant rat; “H” alleles) with a Brown Norway (BN) congenic expressing polydactylyl luxate syndrome (BN-Lx, an LT-sensitive rat; “B” alleles) [25], [26]. This RI panel has been successfully used for identification of quantitative trait loci that control a range of phenotypes, including cardiovascular function, insulin resistance and multiple behavioral traits (for review see [26]). We tested nineteen HXB/BXH RI rat strains and their macrophages for sensitivity to toxin. Ten of nineteen strains were sensitive, and once again the LT sensitivity of their isolated macrophages correlated perfectly with animal susceptibility (Figure 1B). Differences in PA receptor function were ruled out as progenitor strains had similar sensitivity to an LF-Pseudomonas exotoxin A fusion protein (FP59), which requires PA for cell entry but induces lethality by inhibition of protein synthesis (data not shown). Analyses of LT sensitivity phenotypes in the context of published marker data for all chromosomes of the RI rat panel [28] pointed to the existence of a single sensitivity locus on chromosome 10, with the strongest linkage being to marker D10Rat102 (52. 5 M), where marker genotypes matched sensitivity and resistance to LT in all but two rat strains (P = 0. 001) (Figure 2A and Figure S1). An abridged set of markers (O. Seda and L. Sedova, unpublished) that were mapped to the initial marker set for agreement (P. Flodman et al. , unpublished) indicated that the genotype at another marker (D10Rat77,56. 9 M) was fully consistent with the LT phenotype for all genotyped RI strains (P = 0. 00005; Figure 2A). We noted that rNlrp1 (58. 0 M), the ortholog for the murine LT macrophage sensitivity locus, lies within 1. 1 M of D10Rat77 (Figure 2B). Analysis of SNP data in the region between marker D10Rat77 and marker D10Rat80 (also from the unpublished abridged data set) showed a perfect correlation between genotype and LT phenotype in all rat strains for the region surrounding rNlrp1 (P = 0. 00001; Figure 2C). The boundaries of the LT susceptibility locus were determined by SNP analyses of all RI strains to lie between SNP Cpn_10055303964 at 55. 3 M and WKYc98d01_s1_778 at 58. 2 M (http: //gscan. well. ox. ac. uk/gsBleadingEdge/rat. snp. selector. cgi). Analyses of SNPs within this locus comparing RI progenitor strains to several of the previously characterized inbred strains (COP, LEW, WKY, Dahl/SS, and F344) found only 7 individual SNPs that perfectly correlated with sensitivity among these LT sensitive and LT resistant strains. Three of these SNPs lie very close to rNlrp1 (rat101_030_o22. q1ca_511 at 57. 8 M, rat102_003_p19. q1ca_444 at 57. 9 M, and J577324 at 58. 1 M), further supporting rNlrp1 as the leading candidate sensitivity locus among a small number of candidate genes. In view of the prior demonstration that mNlrp1b controls mouse macrophage sensitivity to LT, it was not surprising that the mapping data identified a locus containing rNlrp1 as determining LT sensitivity in rat macrophages. However, a single locus control of animal death was not anticipated. BLAST searches (http: //blast. ncbi. nlm. nih. gov/Blast. cgi) using the predicted BN rNlrp1 (allele 1, see later sections) sequence also identified a potential rNlrp1 paralog (GenBank accession: XM_001080760 “similar to NACHT, leucine rich repeat and PYD containing 1” at LOC691998 in the Rat Genome Sequencing Consortium (RGSC) v3. 4, or alternatively, XM_001080056 at LOC687768 in the Celera assembly) located immediately adjacent to rNlrp1 with a predicted protein sequence that has 76% aa identity with the BN rNlrp1 sequence. RT-PCR analyses utilizing intraexonic and intron-spanning primers specific to this paralog showed that it exists in all strains except Copenhagen (COP) but is not expressed in macrophages and thus it is unlikely that this paralog is involved in macrophage toxicity (Figure S2). The perfect correlation of macrophage sensitivity and animal susceptibility to toxin, along with the established role of mNlrp1b in controlling murine macrophage sensitivity suggested that rNlrp1 was the best candidate for control of sensitivity. We next sequenced rNlrp1 cDNA from twelve rat strains in order to identify sensitivity-correlated variations. BMDMs isolated from ten strains analyzed previously (Figure 1) as well as the progenitor strains for the RI panel (Figure 2) were used as sources of mRNA for sequencing. The sequencing identified a 3657-bp coding region corresponding to a 1219-aa rNlrp1 protein. By aligning the cDNA sequences to the BN genomic sequence data we determined that the rNlrp1 mRNA is formed through splicing of 14 exons, arranged like those of mNlrp1b [3]. Conserved Domain Database (CDD) searches indicated that rNlrp1 contains the same functional NACHT (nucleotide oligomerization; pfam05729), LRR (leucine-rich repeat; cd00116), and CARD (caspase recruitment; pfam00619) domains as mNlrp1b and hNlrp1 (Figure 3). rNlrp1 is similar to mNlrp1b in lacking a pyrin domain. Sequence alignments showed there to be two sensitivity-associated (“sensitive”) alleles (1 and 2) and three resistance-associated (“resistant”) alleles (3,4, 5) among the studied strains (Figure 3 and Figure S3). The differences within these two groups turn out to be minor. The protein encoded by the second sensitive allele, allele 2, differs from allele 1 at a single amino acid (Asn61 to Lys61), but this substitution is also found in resistant alleles 3–5. More interestingly, alleles 3 and 4 both contain sequences corresponding to the N-terminal region of resistant allele 5 and several polymorphisms associated with the C-terminus of sensitive allele 1. Thus, the predicted proteins in these two rats combine a few elements of both a resistant and a sensitive rNlrp1, making it unlikely that the NACHT, LRR and CARD domains determine LT susceptibility. A single difference between resistant alleles 3 and 4 was found, where a substitution results in a Gln to Arg change (Arg561). Allele 5, however, is similar to allele 4 and contains this Arg561 residue, indicating that this residue is also unlikely to be associated with resistance to LT. Thus, we conclude that all the polymorphisms that correlate perfectly with LT sensitivity lie in the 100-aa N-terminus of rNlrp1. Unfortunately, no information is available about the function of this region in rodent Nlrp1 proteins. There is no homologous region in hNlrp1, which instead harbors an N-terminal pyrin domain (absent from rodent Nlrp1 proteins) [10]. In the work reported here, a RI rat panel was used to identify the LT susceptibility locus for both rats and their macrophages. Analyses of LT sensitivity in several rat strains as well as the RI panel identified a complete correlation between macrophage and rat sensitivity to the toxin. This locus, on chromosome 10, contains rNlrp1, which is the homolog for the mouse mNlrp1b gene, previously proven to be critical for determining murine macrophage susceptibility to LT. Sequence analyses of rNlrp1 in primary macrophages from twelve rat strains identified five polymorphic alleles. Surprisingly, the few polymorphisms that correlated with LT macrophage and animal sensitivity in rats were located within the first 100 aa of rNlrp1, in an area of undefined function, and not within the previously described Nlrp1 functional domains (NACHT, LRR and CARD). The mapping data strongly suggests (with a P = 0. 000001) that this rNlrp1-containing region of chromosome 10 is the LT sensitivity locus for both rats and their macrophages. Among the approximately 250 microsatellite markers previously characterized for this RI set [28] and new previously unpublished markers sets, we found that marker D10Rat77 on chromosome 10 had an absolute genotypic correlation with the LT sensitivity phenotype. SNP analysis in this region also confirms the marker data and shows perfect correlation for LT susceptibility within the locus containing rNlrp1 (genome-wide empiric p-value = 0. 001). Our mapping data does not rule out the possibility that another gene very closely linked to rNlrp1 could be mediating LT' s effects in the rat. However, two additional lines of evidence argue against this possibility. First, rNlrp1 aa sequence variations between several inbred rat strains unrelated to the RI panel progenitors correspond perfectly with the sensitivity phenotypes. Second, an absolute correlation was found between macrophage sensitivity and animal susceptibility for 34 rat strains. Considering the established role of mNlrp1b in control of murine macrophage sensitivity, it is unlikely that a different gene controls macrophage sensitivity in rats. However we cannot exclude the possibility that whole animal susceptibility is controlled by multiple closely-linked genes within the single chromosome 10 locus identified in this study, and that these genes are inherited in a fashion such that polymorphisms associated with sensitivity in rNlrp1 are also always found in another candidate gene. Historically, a similar issue plagued the identification of mNlrp1b as the mouse macrophage sensitivity locus for LT. The sensitivity locus in mice was first identified as the closely linked Kif1c gene, which presented almost perfect polymorphism correlations with sensitivity [29]. In the absence of a transgenic rat model providing definitive proof linking the rNlrp1 gene to rat death, we believe the mapping and sequence data presented here strongly support rNlrp1 as the most likely determinant of LT sensitivity. Gene predictions and BLAST searches identified a potential paralog immediately adjacent to rNlrp1, but this paralog is not expressed at the mRNA level, at least in macrophages. Similarly, of the three tandem mNlrp1 paralogs found in mice, only mNlrp1b was shown to be expressed in the LT-sensitive 129S1/SvImJ macrophage, and expression of this paralog was sufficient to confer LT sensitivity to resistant mouse macrophages and fibroblasts [3], [8]. However, the other two mouse paralogs are expressed in a number of inbred strains, further complicating analyses of mouse susceptibility. Curiously, phylogenetic analyses indicate that the predicted rNlrp1 paralog sequence is distant from the other rat Nlrp1 sequences and is more similar to the mouse and human Nlrp1 sequences (Figure S4). The highly polymorphic nature of the mNlrp1b alleles has made it difficult to associate specific polymorphisms with the macrophage sensitivity phenotype in mice. Fortunately, there are far fewer differences in rNlrp1 between sensitive and resistant rat strains. Sequence differences that correlated with phenotypic differences were found only within the extreme N-terminal region of rNlrp1, and not in the domains (NACHT, LRR, and CARD) which have recognized roles in Nlrp1 function. This surprising finding draws attention to the N-terminal domains of rodent Nlrp1 proteins, absent in hNlrp1, which instead contain a pyrin domain at the N-terminus [11]. The pyrin domain in hNlrp1 is required for association with the inflammasome adaptor protein ASC, which is not part of the LT-induced mNlrp1b inflammasome complex [9]. Interestingly, all human macrophages tested to date have been LT-resistant (unpublished observations), a behavior that might relate to the absence of the N-terminal pyrin domain in rodent Nlrp1. However, hNlrp1 polymorphisms are now being identified and associated with a number of human diseases [30]–[32], so it may be necessary to test a larger number of donors to identify any LT-sensitive hNlrp1 alleles. As the N-terminus in hNlrp1 plays an important role in protein-protein interactions, it is tempting to postulate that the N-terminal 100 aa of rodent Nlrp1 proteins may also interact with other cellular components to modulate function. The perfect correlation of rat macrophage LT sensitivity to that of the animals might at first suggest that the lysis of macrophages in vivo causes the rapid death of LT-injected rats. However, this is unlikely to be the case, for several reasons. Rat macrophages begin to die only 2 h after treatment in vitro with saturating toxin doses, whereas the rats may die in as little as 37 min [20]. Initial studies in LT-treated mice were interpreted as showing that death (which occurs only after 2–3 days) resulted from cytokines released following macrophage lysis [33]. However, more extensive later studies showed that mice harboring resistant macrophages also succumb to LT through a vascular collapse that is similar to that in mice with sensitive macrophages [13], and the correlation within mouse strains between the LT sensitivities of isolated macrophages and the animals is poor [13], [15]. Studies with mNlrp1b transgenic mice confirm that macrophage and animal susceptibility to LT are not correlated [16]. Preliminary studies in our laboratory suggest that cell types other than macrophages control the lethal response to LT (data not shown). Consistent with this view, Nlrp1 has recently been demonstrated to play a functional role in a number of cell types, including neuronal cells [34]–[37]. Furthermore, it should be noted that LT-induced death in both rats and mice has recently been associated with early changes in cardiac function [21], [23]. Thus, it is possible that LT targeting of rNlrp1 function in the heart plays a role in the rapid lethality phenotype. A better understanding of the distribution and function of different Nlrp1 isoforms in various cell types is needed to fully understand the mechanisms by which LT may influence Nlrp1 activity, and whether this gene alone is sufficient for control of animal susceptibility to toxin. In summary, we present data mapping the control of rapid LT-induced rat death to a single chromosome 10 locus. This locus contains the polymorphic rNlrp1 gene, which is the best candidate for conferring sensitivity to macrophages, and possibly to animals. As such, this is the first suggestion that an inflammasome NLR protein may directly control animal lethality. While both the mechanistic basis for the rapid LT-induced lethality in the rat and direct proof of rNlrp1-mediated rat death require further experimentation, identification of the limited polymorphisms within rNlrp1 that correlate perfectly with LT sensitivity suggest a starting point for analysis of the possible role this protein may play in controlling rapid rat death in response to LT. All animal experiments were performed in strict accordance with guidelines from the NIH and the Animal Welfare Act, under protocols approved by the Animal Care and Use Committee of the National Institute of Allergy and Infectious Diseases, National Institutes of Health. PA, LF, and FP59 were purified from B. anthracis [38]–[40]. The LF used here is a recombinant protein having an N-terminal sequence beginning HMAGG. Doses and concentrations of LT given for each experiment correspond to that of each toxin component (i. e. , 1 µg/ml LT is 1 µg/ml PA +1 µg/ml LF and 100 µg LT is 100 µg PA +100 µg LF). Rats purchased from Charles River Laboratories (Wilmington, MA) were maintained there as either inbred or long-term outbred colonies. Rats strains used included (with strain designations, abbreviations and inbred/outbred status): Brown Norway (BN/Crl; BN; inbred), Fischer CDF (F344/DuCrl; CDF; inbred), SASCO Fischer (F344/NCrl; F344; inbred), Dahl Salt Sensitive (SS/JrHsdMcwiCrl; Dahl/SS; inbred), Lewis (LEW/Crl; LEW; inbred), Wistar (Crl: WI; WIS; outbred), Wistar Kyoto (WKY/NCrl; WKY; inbred), Sprague Dawley (CRL: SD; SD; outbred), Spontaneously Hypertensive Rat (SHR/NCrl; SHR; inbred), Copenhagen (COP/CrCrl; COP; inbred), Zucker-Lean (Crl: ZUC-Leprfa; ZUC; outbred) and Fawn Hooded Hypersensitive (FHH; inbred). The recombinant inbred (RI) rat strain panel used in this study was derived from the progenitor strains BN-Lx and SHR/Ola (indicated to be genetically equivalent to SHR/Lj used in this study) [25], [26]. The microsatellite marker genotypes and linkage maps for the RI panel were most recently characterized by one of our laboratories [28]. Additional microsatellite markers were identified in progenitors and mapped across the RI strains (P. Flodman et al. , unpublished; O. Seda and L. Sedova, unpublished) by PCR. Marker data were correlated with SNP genotypes available through the Wellcome Trust Centre for Human Genetics STAR Rat SNP Selector (http: //gscan. well. ox. ac. uk/gsBleadingEdge/rat. snp. selector. cgi). Adult female RI rats (9–12 weeks old) of the 19 strains of the HXB/BXH set were rederived, bred, and maintained at the University of California, San Diego, and shipped to Bethesda, MD, for toxin testing and bone marrow collection. The progenitor strains and a congenic strain, SHR-Lx, were included in the analysis. Rats were acclimated for four-five days prior to experiments. For all rat LT challenge studies, female rats (130–160 g) were injected with LT (100 µg, IV) and monitored continuously for 5 h followed by a 24-h check of surviving animals. This dose of toxin represents 10× LD100 for the sensitive F344 and CDF rats when using the well-characterized toxin prepared in our laboratory. L929 mouse fibroblast cells were grown in Dulbecco' s modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum, 10 mM HEPES, and 50 µg/ml gentamicin (all obtained from Invitrogen, Carlsbad, CA) at 37°C in 5% CO2. Bone marrow-derived macrophages (BMDMs) were cultured in complete DMEM (as described above) with 30% L929 cell culture supernatant. BMDMs were grown for 7–9 days to allow time for differentiation before use in assays. BMDMs were plated in 96-well plates 24 h prior to assays at 90% confluence. For basic macrophage LT sensitivity testing, cells were exposed to LT at the indicated concentrations and times. Viability was assessed by addition of MTT [3- (4,5-dimethyl-2-thiazolyl) -2,5-diphenyl tetrazolium bromide] (USB Corporation, Cleveland, OH) to a final concentration of 0. 6 mg/ml in DMEM. Following 30–45 min of incubation with MTT dye, cell culture medium was removed and cells were dissolved with 0. 5% SDS, 25 mM HCl in 90% isopropanol and A570 was measured. Percent viabilities were calculated relative to medium-treated controls. Statistical analyses were performed using SAS (ver. 9. 1. 3, SAS Institute Inc. , Cary, NC). Association between genotype and phenotype across the RI strains was assessed for each marker using Fisher' s exact test. Nominal p-values for the tests of association are reported without correction for multiple comparisons. The genome-wide significance of the linkage findings was assessed using MapManager QTXb20 to calculate an empiric p-value based on 10,000 permutations [41]. RNA was isolated from BMDMs by a TRIZOL extraction method according to manufacturer' s protocol (Invitrogen). RNA was reverse transcribed using the SuperScript III First Strand Synthesis System (Invitrogen). Sequencing primers were designed to cover the full coding sequence of rNlrp1 using the predicted rNlrp1 mRNA sequence (GenBank accession: XM_340835) from the Rat Genome Sequencing Consortium BN rat genomic sequence data (RGSC v3. 4, GenBank accession: NW_047334). The primary sequencing reactions consisted of amplifying five overlapping cDNA regions (locations indicated in Figure 3 and primer sequences are provided in Table S1). All PCR was performed with the TaKaRa Ex Taq (TAKARA Bio Inc. , Otsu, Japan). PCR products were purified using PureLink PCR purification kits (Invitrogen) and sequenced on Applied Biosystems 3730xl DNA analyzers at MACROGEN USA (Rockville, MD). Additional primers and reactions were used for confirmation of specific regions and clarification of the overlaps for the primary reactions (Table S1). Sequences were assembled and analyzed using the Lasergene program suite (DNASTAR, Inc. , Madison, WI). Alignments were created using ClustalX (http: //www. clustal. org/) and phylogenetic trees were visualized with TreeViewX (v0. 5; http: //darwin. zoology. gla. ac. uk/~rpage/treeviewx/index. html). Exon structure was determined by aligning the cDNA sequences to the BN genomic data by a BLAT search (http: //genome. ucsc. edu/cgi-bin/hgBlat). Amino acid sequences used in alignments and phylograms included a potential rNlrp1 paralog predicted sequence (XP_001080760. 1), hNlrp1 isoform 1 (NP_127497), C57BL/6J mNlrp1a (AAZ40527), C57BL/6J mNlrp1c (AAZ40528), BALB/cJ mNlrp1b Allele 1 (AAZ40509), C57BL/6J mNlrp1b Allele 2 (AAZ40517), NOD/LtJ mNlrp1b Allele 3 (AAZ40521), DBA/2J mNlrp1b Allele 4 (AAZ40523), and CAST/EiJ mNlrp1b Allele 5 (AAZ40526). cDNA GenBank sequences for new rNlrp1 sequences determined in this work are as follows: HM060628 (BN), HM060629 (BN-Lx), HM060630 (COP), HM060631 (Dahl/SS), HM060632 (CDF), HM060633 (LEW), HM060634 (SD), HM060635 (SHR), HM060636 (SHR/Lj), HM060637 (WIS), HM060638 (WKY), HM060639 (ZUC).

Inflammasomes are multiprotein cytoplasmic complexes that respond to a variety of danger signals by activating the host innate immune response. The sensor components of these complexes are NLR (NOD-like receptor) proteins. In this report, a recombinant inbred rat strain collection was used to genetically map anthrax lethal toxin (LT) susceptibility to a limited region of chromosome 10 containing one such sensor, Nlrp1. Similar to its mouse ortholog, Nlrp1b, which controls murine macrophage sensitivity to this toxin, the locus containing rat Nlrp1 was shown to control macrophage sensitivity to anthrax LT. However, unlike the situation in mice, where multiple genetic loci influence animal susceptibility to LT, the single chromosome 10 locus alone appears to control the rapid anthrax LT-induced death, which can occur in as little as 37 minutes. Sequencing of Nlrp1 from 12 rat strains identified polymorphisms which correlated perfectly with animal sensitivity to toxin. These polymorphisms were within the N-terminal 100-amino acid portion of Nlrp1, in an area of unknown function, which suggests that the N-terminus of rodent Nlrp1 could be an important functional domain.

Abstract Introduction Results Discussion Materials and Methods

genetics and genomics/animal genetics microbiology/innate immunity immunology/innate immunity genetics and genomics/genetics of disease infectious diseases/bacterial infections microbiology/cellular microbiology and pathogenesis

2010

Susceptibility to Anthrax Lethal Toxin-Induced Rat Death Is Controlled by a Single Chromosome 10 Locus That Includes rNlrp1

Embryonic stem cells exhibit pluripotency: they can differentiate into all types of somatic cells. Pluripotent genes such as Oct4 and Nanog are activated in the pluripotent state, and their expression decreases during cell differentiation. Inversely, expression of differentiation genes such as Gata6 and Gata4 is promoted during differentiation. The gene regulatory network controlling the expression of these genes has been described, and slower-scale epigenetic modifications have been uncovered. Although the differentiation of pluripotent stem cells is normally irreversible, reprogramming of cells can be experimentally manipulated to regain pluripotency via overexpression of certain genes. Despite these experimental advances, the dynamics and mechanisms of differentiation and reprogramming are not yet fully understood. Based on recent experimental findings, we constructed a simple gene regulatory network including pluripotent and differentiation genes, and we demonstrated the existence of pluripotent and differentiated states from the resultant dynamical-systems model. Two differentiation mechanisms, interaction-induced switching from an expression oscillatory state and noise-assisted transition between bistable stationary states, were tested in the model. The former was found to be relevant to the differentiation process. We also introduced variables representing epigenetic modifications, which controlled the threshold for gene expression. By assuming positive feedback between expression levels and the epigenetic variables, we observed differentiation in expression dynamics. Additionally, with numerical reprogramming experiments for differentiated cells, we showed that pluripotency was recovered in cells by imposing overexpression of two pluripotent genes and external factors to control expression of differentiation genes. Interestingly, these factors were consistent with the four Yamanaka factors, Oct4, Sox2, Klf4, and Myc, which were necessary for the establishment of induced pluripotent stem cells. These results, based on a gene regulatory network and expression dynamics, contribute to our wider understanding of pluripotency, differentiation, and reprogramming of cells, and they provide a fresh viewpoint on robustness and control during development. In multicellular organisms, cells that exhibit stemness during development can both self-renew and differentiate into other cell types. In contrast, differentiated cells lose the ability to further differentiate into other cell types and terminally differentiated cells can only self-renew. Currently, how stemness and the irreversible loss of differentiation potential are characterized by gene expression patterns and dynamics are key questions in developmental biology. Cells with stemness include embryonic stem cells (ESCs), which are derived from the inner cell mass of a mammalian blastocyst and are pluripotent, i. e. , they can differentiate into all the types of somatic cells [1,2]. To maintain pluripotency, pluripotent genes such as Pou5f1 (also known as Oct4) [3,4] and Nanog [5,6] are activated in ESCs. Expression of these genes gradually decreases during cell differentiation, whereas expression of differentiation marker genes increases. Understanding these changes in gene expression patterns over the course of cell differentiation is important for characterizing the loss of pluripotency. During normal development, the loss of pluripotency is irreversible. However, the recovery of pluripotency in differentiated cells was first achieved by experimental manipulation in plants, and then in Xenopus laevis via cloning by Gurdon [7]. More recently, the overexpression of four genes that are highly expressed in ECSs, Oct4, Sox2, Klf4, and Myc (now termed Yamanaka factors), has been used to reprogram differentiated cells. Overexpression of these genes leads to cellular-state transition and changes in gene expression patterns, and the transition generates cells known as induced pluripotent stem cells (iPSCs) [8]. Previous studies have also uncovered the gene regulatory network (GRN) related to the differentiation and reprogramming of cells [9,10]. To understand the differentiation process theoretically, Waddington proposed a landscape scenario in which each stable cell-type is represented as a valley and the differentiation process is represented as a ball rolling from the top of a hill down into the valley [11]. In this scenario, the reprogramming process works inversely to push the ball to the top of the hill [12–14]. As a theoretical representation of Waddington’s landscape, the dynamical-systems approach has been developed over several decades, pioneered by Kauffman [15] and Goodwin [16]. In this approach, the cellular state is represented by a set of protein expression levels with temporal changes that are given by GRNs. According to gene expression dynamics, the cellular state is attracted to one of the stable states, which is termed an attractor. Each attractor is assumed to correspond to each cell type. Indeed, this attractor view has become important for understanding the diversification of cellular states and their robustness. Both theoretical and experimental approaches have been developed to assign each cell-type to one of the multi-stable states [17–19]. In these approaches, a pluripotent state is regarded as a stationary attractor with relatively weak stability, and the loss of pluripotency is the transition by noise to attractors with stronger stability. An alternative approach investigated how the interplay between intra-cellular dynamics and interaction leads to differentiation and the loss of pluripotency [20–23]. Specifically, the pluripotent state is represented by oscillatory states following the expression dynamics of more genes, whereas the loss of pluripotency is represented by the decrease in the degree of expressed genes necessary for oscillatory dynamics. Here, differentiation is triggered by cell-cell interactions, which lead to robustness in developmental paths and the final distribution of cell types [20,24,25]. By using several GRNs, cells with oscillatory intracellular gene expression dynamics are found to differentiate into other cell types by cell-cell interactions [21,26–28]. Indeed, the recovery of pluripotency by gene overexpression is a process predicted to facilitate recovery of lost degrees of freedom and oscillation [20]. However, of the question of whether this theory applies to realistic GRNs has yet to be explored. Despite these earlier studies, pluripotency has not yet been confirmed in a realistic GRN observed in experiments, and the mechanism of reprogramming remains elusive. Epigenetic modifications such as DNA methylation and histone modification are now also recognized as important in cell differentiation. Epigenetic change solidifies differentiated-cellular states by altering chromatin structure to generate transcriptionally active and inactive regions [29,30]. With epigenetic change, the activity of gene expression is preserved in a process known as epigenetic memory [31]. Indeed, epigenetic modification is suggested as a barrier to reprogramming [32]. However, the theoretical inter-relationship between expression dynamics and epigenetic modification has yet to be fully explored. The aim of the present study was three-fold. First, by using GRNs obtained from a previous experimental study, we examined the validity of two differentiation scenarios: 1) oscillation + cell-cell interaction and 2) multistability + noise. Second, to demonstrate that differentiation by gene expression dynamics is solidified by epigenetic modification, we introduced a mathematical model for epigenetic feedback regulation. Third, we investigated how overexpression of some genes leads to reprogramming, i. e. , regaining pluripotency from differentiated states (scenario 1) by initializing epigenetic changes. Below, we have first introduced a simple model extracted from an experimentally observed GRN. This model consists of several genes, including pluripotent and differentiation genes, with mutual activation and inhibition. We then examined the oscillatory dynamics and multistable states scenarios to show that differentiation with the loss of pluripotency progresses from a stem cell state with oscillatory expression through cell-cell interactions. Additionally, the two scenarios were compared with regard to their robustness to noise and the regulation of the ratio of differentiated cells. We also investigated the epigenetic process by introducing variables that give the threshold for the expression of genes to demonstrate that the cellular state derived from gene expression dynamics is fixed by epigenetic feedback regulation. Differentiation by gene expression is fixed according to these threshold variables; thus, the pluripotent and differentiated states are fixed. Finally, we investigated reprogramming by temporally imposing overexpression of genes and examining whether the differentiated state is reversed to the pluripotent state. Via overexpression of several genes, epigenetic fixation was relaxed and the expression levels and dynamics of the pluripotent state were recovered. This reprogramming was shown to require the overexpression of several genes, including pluripotent genes, over a sufficient period beyond the time scale of epigenetic fixation. Indeed, by using a model with five relevant genes, we found that four genes corresponding to the Yamanaka factors must be overexpressed for reprogramming to occur. It was also demonstrated that insufficient overexpression of genes, i. e. , overexpression of pluripotent genes only, results in partially reprogrammed cells (which, experimentally, are known as pre-iPSCs). In the pluripotent state, cells can proliferate and retain their potentiality for differentiation. The expression of pluripotent genes is necessary for pluripotency, but it is not always sufficient. In the differentiation process, expression of pluripotent genes gradually decreases, while expression of differentiation marker genes increases. These temporal changes are a result of gene-gene regulation, which can be integrated as a GRN consisting of pluripotent and differentiation genes. Here, we adopted the GRN reported by Dunn et al. [33] (Fig 1) and produced simplified models by compressing some paths and genes while maintaining the structure of the GRN (see Models). Using the four-gene model (Fig 1C), we first present the behavior of single-cell dynamics. Depending on the parameter Kij, which gives the strength of activation or inhibition from gene j to gene i, there are three possible behaviors: (i) fixed-point attractor with high expression of pluripotent genes (FP), given by a fixed-point x1 ∼ 1; (ii) fixed-point attractor with high expression of differentiation genes (FD), given by a fixed-point x1 ∼ 0; and (iii) the oscillatory state (O), in which all expression levels show temporal cycles (Fig 2). These three states appear as attractors depending on the parameter values Kij. Because the expression level of pluripotent gene x1 is most important for determining the three states, the regulation of gene x1, which is controlled by the parameter K1j, is crucial for determining cellular behavior. In particular, threshold K11 and K13 were found to be critical parameters. Where the value of K11 was low, expression of gene x1 was promoted; where the value of K13 was low, gene x1 was suppressed. First of all, we set all parameters Kij randomly, and examined the dynamics. If the parameter value of K11 (K13) was set to a much lower (larger) value (say 0. 1 (1. 0), respectively), the expression of x1 is fixed to a high value, and the differentiation process would be more difficult. On the other hand, if this parameter value was high (low), x1 was not easily expressed or always expressed, respectively, so that the stem cell state is difficult to be obtained unless other parameter values are finely tuned. With the neighborhood of the above parameters values, the expression level of x1 changes flexibly to other parameters. We then observed that the expression dynamics changed between fixed-point and oscillation easily by changing other parameter values. Indeed, as will be shown, differentiation behavior was observed for a broader range of other parameters. As the parameter space is so huge, we here fixed K11 and K13 at these values and drew the phase diagram against other parameters. For the parameters K11 and K13, for example K11 = 0. 35, K13 = 0. 78, gene expression levels showed oscillation. To study FP, FD, and O, i. e. , the three states described above, we fixed the parameters K11 and K1j (for specific values, see Models), and assessed dependence on the other three parameters K34, K42, and K43 (Fig 3). In most parameter regions, two attractors (stable states) existed, either FP+FD or FD+O depending on the initial conditions. Where the initial condition involved high expression of pluripotent genes, FP or O was reached depending on the parameters; where the initial condition involved high expression of differentiation genes, FD was reached. For higher values of K34 and K43, gene-expression oscillation, i. e. , the oscillatory state, did not appear, and FP and FD coexisted. Conversely, for lower values of K34 and K43, the oscillatory state appeared for 0. 1 < K42 < 0. 5 if pluripotent genes were initially highly expressed. However, where differentiation genes were initially highly expressed, cells fell into FD; thus, FD and O coexisted. As an example of the oscillatory pluripotent state, we fixed the parameters at K34 = 0. 45, K42 = 0. 30, and K43 = 0. 45 for most of the simulations described below. For oscillatory gene expression, negative feedback is generally required. In our model, negative feedback of gene x1 exists through genes x2, x3, and x4. For the parameter values that generated oscillatory expression, O, auto-promotion and negative feedback of gene x1 (as the key factor in pluripotency) were balanced. Where either of the two regulations was dominant, oscillation ceased and the cellular state fell into either of FP or FD. To understand developmental processes, we must investigate the switch from pluripotent to differentiated states. This differentiation event can be mediated either by cell-cell interactions (i. e. , by chemicals from other cells) or by noise. Here we explored these two possibilities. Cellular differentiation in multi-cellular organisms also involves epigenetic changes, such as histone modification and DNA methylation, which stabilize differentiated states: once differentiated, cells do not regain pluripotency even if the expression level is perturbed. Hence, we introduced epigenetic modification into our model to strengthen the stability of the differentiated state. Currently, there is no definitive method for introducing the epigenetic process because the precise molecular process of histone modification is difficult to implement in a model with gene expression dynamics. However, it is possible to model the influence of the epigenetic process on expression dynamics phenomenologically [34–38]. The epigenetic process tends to fix the expression state; for example, when a given gene is expressed for a given period, its expression tends to become fixed, and when it is not expressed for a given period, it remains silenced. In other words, the threshold for expression decreases or increases when the gene is expressed or suppressed over a given time span, respectively. Thus, we introduced epigenetic feedback regulation as a change in the threshold for expression, which was previously given by the expression threshold parameter Kij in our GRN model. Here, we replaced the parameter Kij with an epigenetic variable θij (t), which changes over time depending on expression levels. Consequently, the expression level of the regulator xj affects that of the regulatee xi through this epigenetic variable. This is given as dynamics as θ ˙ i j (t) = 1 τ e p i (Θ i j - θ i j (t) - α x j (t) ). (1) The threshold θij (t) is elevated to Θij, when the gene xj is not expressed (i. e. , xj (t) ∼ 0), whereas the threshold decreases to Θij − αxj (t) when the gene is fully expressed, i. e. , xj (t) ∼ 1. Hence, the term −αxj (t) represents epigenetic feedback, i. e. , if gene xj is expressed, it is more likely to be expressed; if it is not expressed, it is less likely to be expressed. The term Θij thus represents the epigenetic barrier for genes that are not expressed. The strength of epigenetic fixation given by Θij generally depends on each regulation. Since the expression of pluripotent genes in our model is highly variable, a higher value of Θij is required to fix their expression. Hence, we chose higher Θij values for regulations associated with pluripotent genes. Specifically, epigenetic fixation threshold values were set to 1. 0 for the pluripotent genes Θ31, Θ21, and Θ42, while they were lowered to 0. 78 for the differentiation regulators Θ13, Θ34, Θ43. For auto-regulation Θ11, the threshold value was set lower, e. g. , at 0. 50, since self-activation tightly fixes the expression with small Θij. This is because the genes to regulate and to be regulated are identical. This was due to the simplification, which included the self-activation (we examine the five-gene model without self-activation below, in which all Θij for pluripotent genes are set to 1. 0). Given these parameters, we simulated our model with epigenetic feedback regulation and studied dependence on the epigenetic variables τepi and α. Initially, we focused on the epigenetic variable θij (0) = Kij, which was set with values for cases with (A) fixed-point states and (B) the oscillatory state. Mature cells can be dedifferentiated into iPSCs by inducing Yamanaka factors [8]. Indeed, in dynamical-systems theory, such recovery of pluripotency was predicted as cellular-state transition from a differentiated fixed-point to the pluripotent oscillatory attractor induced by forced-expression of several genes [20]. Here we discuss the conditions for reprogramming, i. e. , switching cellular states by experimental manipulation to regain pluripotency, in our model, also by taking the reversal of epigenetic fixation into account. First, we investigated reprogramming in the model without the epigenetic process. In this case, differentiated cells were reprogrammed by externally increasing the expression of the pluripotent genes instantaneously, i. e. , increasing the value of x1. Instantaneous increase in the expression was sufficient here, since the cellular state is represented only by the expression levels of xi. In order to stabilize the differentiated states against perturbations and sustain irreversibility of cell differentiation, the classic model including only gene expression dynamics is insufficient (see also [39]). By introducing epigenetic feedback regulation with a different time scale, we succeeded in obtaining the result consistent with reprogramming experiments. In the model with the epigenetic process, however, differentiated cells were not reprogrammed by an instantaneous increase in xi. Following overexpression, cells quickly returned to the differentiated fixed-point. This is because the epigenetic variable, which cannot be altered over a short period, was already increased so that expression of pluripotent genes could not be recovered by instantaneous, or short-term, overexpression. Indeed, we examined the instantaneous overexpression of each gene, as well as a combination of several genes, but reprogramming never occurred. We then introduced the overexpression of pluripotent genes into a differentiated cell over a sufficiently long time span Te. For example, pluripotent genes were overexpressed from t = 1 to Te = 100 to the level of xi = 15. Additionally, we added external activation of gene x4 to inhibit the expression of gene x3 (Fig 8). In this case, cells were reprogrammed, and gene expression levels regained oscillation and recovered pluripotency. The expression threshold was also reduced (Fig 9), so that epigenetic fixation was relaxed. By starting with this reprogrammed cell, some of the divided cells differentiated given a sufficient level of cell-cell interaction. After overexpression of xi to the value xe for time span Te, the epigenetic variable θij (t) was estimated to decrease to α x e × T e τ e p i. Hence, epigenetic fixation is relaxed if this value reaches θij (0), where θij (0) is the value after epigenetic fixation. Where τepi = 5. 0 × 10−4 and α = 0. 1, for example, xeTe must be larger than 3. 0 × 105 for θ11 (t) to return to the initial value 0. 35. For example, if the overexpression value is changed from 15 to 3, overexpression time required about 5 times. The product of overexpression value and time determines the reprogramming. The reprogramming ratio increases (in a threshold-like manner) as the product increases beyond a critical value 103 (Fig 10). Indeed, this is natural, as the relaxation process of epigenetic fixation is determined by the product. The epigenetic fixation is introduced so that the genes that are not expressed are harder to be expressed, following observations in cell-biology. Accordingly, the strength of epigenetic fixation Θij has to be larger than the value of θij chosen initially. Therefore, epigenetic fixation threshold values Θij for pluripotent genes were set to 1. 0 because the maximum value of initial threshold values θij (0) was 0. 94. If it is set to a lower value, the gene is not remained silenced due to the epigenetic change, even when it is not expressed. On the other hand, if the epigenetic fixation threshold values Θij for differentiation regulators were also set to 1. 0, the reprogramming by overexpressing the corresponding genes as well as external factors was not possible. In fact, we carried out both differentiation and reprogramming simulations by choosing a variety of values of Θij, and confirmed that epigenetic fixation threshold values for pluripotent genes have to be larger than that for differentiation regulators, to be consistent with experimental observations. In addition to overexpression levels and time span, the number of overexpressed genes is important. Reprogramming did not occur by overexpression of a single gene, even though its level and time span were sufficient to decrease the epigenetic variable: two or more appropriate genes had to be overexpressed. If a single gene x1 was overexpressed over a sufficient period, the transition to a different fixed-point state occurred, but gene expression did not regain oscillation. By starting from this cell with this new fixed-point state, differentiation did not occur again even when the number of cells was increased. These cells showed increased expression of pluripotent genes up to the level of the original pluripotent cell, but they did not regain the capacity for differentiation. Thus, decreasing the epigenetic threshold variable of pluripotent genes was not sufficient for reprogramming. We then conducted a reprogramming simulation by changing the initial values for the epigenetic variable θij (0), that is, Θij. In general, as Θij became smaller, epigenetic fixation became weaker, and the number of genes that had to be overexpressed decreased. For example, if Θ34 = 0. 5 and Θ43 = 0. 3, the overexpression of just two factors, x1 and x2, without the external overexpression of any other genes could lead to reprogramming (S4 Fig). According to our results, pluripotent stem cells had an oscillatory gene expression component; thus, the recovery of oscillation was necessary for recovery of pluripotency. However, oscillation alone was not always sufficient for pluripotency. If the decrease in the epigenetic threshold value was insufficient, the oscillation was weak and the bifurcation to a differentiation fixed point could not occur by cell-cell interactions. In this case, pluripotent genes were expressed. A cellular state such as this, with expression of pluripotent genes but without differentiation potential, corresponds to the pre-iPS state previously reported in reprogramming experiments [32,40]. To promote expression of pluripotent genes, there is an auto-expression loop. This auto-expression is mediated via positive feedback by mutual regulation of genes. In the four-gene model, which has been described and studied thus far, this positive feedback loop was introduced as the self-expression of x1. Auto-expression such as this may be over-simplified, especially considering epigenetic modification as already mentioned. In reality, the auto-expression feedback loop consists of a number of genes. Hence, we replaced auto-regulation of x1 in the four-gene model with a loop structure via a new gene x5 (as shown in Fig 1B), and attempted to validate our previous results and examine the conditions necessary for reprogramming in comparison with experiments. First, we confirmed that the two fixed-points, FP and FD, and the oscillatory state, O, existed in the five-gene model (see SI, S1 Text). Once confirmed, we also included epigenetic threshold variables, as in the four-gene model. For example, we used two epigenetic fixation parameters depending on the regulator type, i. e. , the epigenetic fixation value for the pluripotent regulators (Θ15, Θ31, Θ21, Θ51, Θ42) was 1. 0, and for the differentiation regulators (Θ13, Θ34, Θ43) it was 0. 65. Additionally, we confirmed that the switching from oscillatory state to FD progressed via cell-cell interactions (S6 Fig). To regain pluripotency from the differentiated state, in our reprogramming experiment with the five-gene model, overexpression of the genes x1, x2, and x5, as well as one external factor (to inhibit gene x4), was necessary. These four genes correspond to the Yamanaka factors (Oct4, Sox2, Klf4, and Myc) used for reprogramming (Fig 11, S7 Fig). As long as we started the reprogramming simulation after the threshold value θij (t) for differentiated cells reached the pre-set level Θij, these four genes were necessary for reprogramming. The number of genes that had to be overexpressed depended on the level of epigenetic fixation. In general, overexpression of the four aforementioned genes over a sufficient period was required for reprogramming to reset the value of epigenetic variables for the differentiated cells (epigenetic fixation was complete to have θij (t) ∼ Θij). In contrast, reprogramming was easier if epigenetic fixation was insufficient, and fewer genes, including x5, were sufficient for reprogramming. In this study, we assessed a simplified model that was a part of an inferred GRN previously reported by Dunn et al. [33]. Some regulations were simplified by deleting mediator genes, but the core network that is believed to be important for pluripotency, in particular the network motif for a toggle switch, was included. In accordance with the reported GRN, the genes in the model corresponded to Nanog, Oct4, Gata6, and Gata4, while the additional gene in our five-gene model corresponded to Klf4. We showed that oscillation and switching between high and low levels of gene expression causes some cells to fall into differentiated states via cell-cell interactions. This interaction-induced differentiation from the oscillatory state was robust to noise. Indeed, expression levels of the pluripotency-related gene Hes1 are reported to oscillate in stem cells, but oscillation is apparently lost after differentiation [41]. This observation is consistent with our oscillation-based mechanism. Alternative proposals for the differentiation mechanism are based solely on multistability and stochasticity. According to these views, both the differentiated and pluripotent states are given by one of the multi-stable fixed-points, and cellular-state transition is caused by noise. For example, a GRN with auto-promotion and mutual inhibition between two genes [17] can produce such bistability. The noise level is critical to this differentiation process. Unless noise level is optimally tuned, the transition between the pluripotent and differentiated states continued to occur via noise, and irreversible differentiation did not occur. Additionally, because switching is stochastic, this model could not control the ratio of pluripotent to differentiated cells, and once a cell was in one of the bistable states, the epigenetic process fixed this state. In contrast, differentiation from oscillatory dynamics and cell-cell interactions is robust to noise. This provides an explanation for pluripotency as oscillatory dynamics, and characterizes the irreversible differentiation as a transition from oscillatory to fixed-point dynamics, which, later, is consolidated by epigenetic feedback. In contrast to our findings, however, a recent study suggested that gene expression in stem cells shows stochastic switching between high and low levels, rather than oscillation dynamics [42]. We note that our mechanism works even with strong stochasticity. Even though the strength of noise is set at a large value (say, σ = 1. 0), the differentiation by cell-cell interaction in our model works well. Besides the noise during the expression dynamics, we have also studied the noise in the division process. Indeed, even though the strength of noise in cell division is large (say σd = 1. 0), the differentiation mechanism in our model still works well. Where this is the case, the oscillatory component underlies gene expression that shows noisy dynamics. Hence, the experimental observation did not contradict our oscillation scenario. Under such noise level, the differentiation ratio from sibling is not necessarily correlated as in the experimental results. Under these high noise levels for σ and σd, and by setting the parameter values say at τdiv = 12. 5 and D = 1. 5, about 4 switching occurred per 100 cell division, as is consistent with the experimental data, while preserving the stochastic oscillatory dynamics. To check the possibility of stochastic oscillation experimentally, one would need to examine whether an oscillatory component exists among noisy dynamics. This would be possible by measuring the transition probability among three states (A, B, C) and examining if the probability P (A → B) has a circulation component, as characterized by the deviation between P (A → B) P (B → C) P (C → A) and P (B → A) P (A → C) P (C → B). We also suggest that by measuring expression of pluripotent genes for a number of iPS cells by single-cell-PCR, one could uncover the loci of oscillatory attractor, as the phase of oscillation is expected to be scattered by cells. Second, in the experiment of [42] switching between Nanog-high to Nanog-low is less frequent than the result presented here. However, this switching frequency can be easily changed in our model by changing the parameter values τdiv, the strength of cell-cell interaction D and noise σ. Here, we also introduced epigenetic threshold variables to fix differentiated cellular states via epigenetic changes. The epigenetic variables in our study promoted gene expression if the regulator gene was highly expressed. Conversely, they inhibited gene expression if the expression of the regulator was low. Indeed, epigenetic modification represented by histone modification is known to reinforce gene function by reconstruction of chromatin [43]. For example, the maintenance of pluripotency is promoted and suppressed by open and closed chromatin states in cell differentiation, respectively. The epigenetic feedback process in our model was a mathematical representation of such reinforcement. In our model, the time scale of epigenetic change τepi was much slower than the time scale of gene expression dynamics, by a factor of 102 − 103. Therefore, because the time scale for transcription is seconds to minutes, epigenetic modification appears to occur over days. If the time scale for cell division is hours, the time scales for gene regulation τgene, cell division τdiv, and epigenetic variable τepi satisfy τgene < τdiv < τepi. Indeed, in our model, epigenetic fixation of cell differentiation works effectively given these conditions. If differentiation occurs, and the differentiation ratio depends on the time scale of epigenetic modification, the rate of epigenetic change can control the distribution of cell types. Hence, epigenetic fixation controls cell distribution and is, therefore, essential to the stabilization of cellular states. However, epigenetic fixation also provides a barrier in reprogramming. In contrast to the scenario without epigenetic fixation, simply resetting gene expression patterns is not sufficient to reprogram differentiated cells. Even if the gene expression pattern of a differentiated cell is reset to the pluripotent state, the cellular state quickly returns to a differentiated state because of the change in the epigenetic threshold variables. Reprogramming also requires overexpression of pluripotent genes over a time span of τepi. Even with overexpression of the correct genes, an insufficient amount of time cannot relax the epigenetic threshold, and cells quickly return to the differentiated state. Indeed, in reprogramming experiments, Yamanaka factors are overexpressed for days by using retroviruses, during which time, it is suggested that chromatin is reconstructed. In our model, the overexpression of multiple transcription factors, including pluripotent genes, was generally necessary for reprogramming to occur. Indeed, in the five-gene model, the four factors required for reprogramming were the Yamanaka factors, Oct4, Sox2, Klf4, and Myc, which are adopted in iPS construction. Even though the GRN in our model contained only five genes, reprogramming required these four factors. In particular, Klf4 was a prerequisite for reprogramming. In iPS cell construction, Klf4 also plays an important role in promotion of reprogramming by interacting with Oct4 and Sox2 [44]. Note that the reprogramming efficiency in experiments is rather low. This might be related with a limited range in the overexpression level in Fig 10. However, at the moment, it is uncertain if this low efficiency is due to difficulty in adjusting such range of overexpression levels, or due to underlying noisy dynamics, or due to some other experimental constraint. Experimentally, reprogramming is reportedly easier if the epigenetic fixation of some genes is weaker. Indeed, epigenetic fixation levels depend on the derived cell type or chromatin structure [29,30]. Furthermore, highly efficient reprogramming, such as deterministic (or non-stochastic) reprogramming from the privileged somatic cell state [45,46], includes a chromatin remodeling factor or specific types of derived cell. This scheme is expected to relax the level of epigenetic fixation for some genes. Thus, it is consistent with the ease of reprogramming caused by reducing epigenetic fixation parameters Θij for some genes (j) in our model. Our study also demonstrates that cells fall into a fixed state with the expression of pluripotent genes when there is insufficient overexpression to suppress differentiation genes. Gene expression levels in such cells do not show oscillation, nor do cells show differentiation again. Even though pluripotent genes are expressed, the potential for differentiation is not regained. These cells are regarded as being in a pre-iPS state, which was previously reported in reprogramming experiments [32,40]. In these experiments, following overexpression of the Yamanaka factors, the cell did not regain pluripotency even though ES cells-markers (SSEA-1 and Oct4) were expressed. The GRNs we studied here are based on several experimental reports. In reality, the GRNs responsible for pluripotency and differentiation involve many more components, and other candidate GRNs have also been proposed for pluripotency [9]. Our conclusions with regard to oscillation-based differentiation, epigenetic fixation, and reprogramming, however, remain valid as long as the present core network is preserved. Additionally, in a differentiation process including the core network structure consisting of Nanog, Oct4, Gata6, and Gata4, as discussed here, the four factors are required for reprogramming, independently of the parameter values, as is consistent with experimental observations. In summary, in our study, oscillatory gene expression produced the pluripotency of cells, and differentiation occurred via a state transition to a fixed-point with the suppression of pluripotent genes. These expression patterns were then fixed epigenetically. In our model, differentiation and reprogramming were interpreted as creation (deletion) of gene expression oscillation and the enhancement or relaxation of epigenetic fixation, respectively. Pluripotent states involved the oscillation of expression of several (here, four to five) genes, while differentiated states suppressed the expression of these genes to reduce oscillation. Thus, our results showed that reprogramming to recover pluripotency involves recovery of gene expression, achieved by overexpression of several genes, and relaxation of epigenetic fixation. The simplified models consisted of either four or five genes with seven or eight regulations, respectively. In simplification of GRN, we decreased nodes and edges as long as the differentiation is possible. As a result, we extracted a four-factor model, as a minimal structure showing differentiation. Furthermore, the reprogramming simulation from this network, as presented in the present paper, is also consisted with experiments. The existence of these regulations in the constructed GRN is supported by earlier studies [9,47–50]. In the four-gene model, the self-activating gene (promotion-loop structure) and its cofactor were regarded as pluripotent genes, and the genes inhibited by these pluripotent genes were regarded as differentiation genes. For example, genes x1, x2, x3, and x4 corresponded with Nanog, Oct4, Gata6, and Gata4, respectively. In the five-gene model, the additional gene was Klf4. Among these genes, only Gata4 functions in cell-cell signaling (interaction) according to Gene Ontology. Hence, we considered cell-cell interactions through diffusive coupling by the gene product of x4. We introduced epigenetic feedback regulation into the model as a change in the threshold for gene expression. This depends on the expression levels of a regulator gene. If the regulator gene is highly expressed, expression of the regulated gene is promoted; however, where expression of the regulator gene is low, expression of the regulated gene is inhibited. Epigenetic change occurs via the change in threshold for expression dynamics and, with feedback, the cellular state is fixed. Here, we describe our gene expression dynamics model. Cellular states are represented by the gene expression pattern of four genes, x1, x2, x3, and x4. These genes regulate the expression levels of themselves and other genes. Additionally, we consider the expression dynamics of gene i of the kth cell at time t, denoted as x i k (t). Only gene x4 is involved in a cell-cell interaction, which is the diffusion of the gene expression level of x4. Hence, our differential equation model is as follows: {x ˙ 1 k (t) = (x 1 k (t) K 11) n 1 + (x 1 k (t) K 11) n 1 1 + (x 3 k (t) K 13) n - x 1 k (t) (+ η 1 k x 1 k (t) ) x ˙ 2 k (t) = (x 1 k (t) K 21) n 1 + (x 1 k (t) K 21) n - x 2 k (t) (+ η 2 k x 2 k (t) ) x ˙ 3 k (t) = 1 1 + (x 1 k (t) K 31) n 1 1 + (x 4 k (t) K 34) n - x 3 k (t) (+ η 3 k x 3 k (t) ) x ˙ 4 k (t) = 1 1 + (x 2 k (t) K 42) n (x 3 k (t) K 43) n 1 + (x 3 k (t) K 43) n - x 4 k (t) + D N (t) ∑ j (x 4 j (t) - x 4 k (t) ) (+ η 4 k x 4 k (t) ) where D is the diffusion coefficient, η i k is an uncorrelated Gaussian white noise term with the mean square deviation σ when the stochastic experiment is considered, and N (t) is the total number of cells at time t. Depending on the parameter Kij, which gives the strength of activation or inhibition from gene j to i, the behavior of our model changes. In cell division, two new cells are produced that have the same gene expression pattern as the original cell. Additionally, gene expression is slightly perturbed by adding a Gaussian white noise (σd = 1. 0 × 10−3, unless otherwise mentioned) after cell division, as ηi xi or (1 − ηi) xi (with η as a random number in [0, σd] after each cell division. In the four-gene model, the positive feedback loop of the pluripotent gene x1 is introduced for self-activation. Auto-regulation such as this may be over-simplified; in reality, this should be replaced by a feedback regulation loop including a number of genes. Therefore, we change the auto-expression of gene x1 in the four-gene model to a loop structure via the gene x5, and the five-gene model is described as follows: { x ˙ 1 k (t) = (x 5 k (t) K 15) n 1 + (x 5 k (t) K 15) n 1 1 + (x 3 k (t) K 13) n - x 1 k (t) (+ η 1 k x 1 k (t) ) x ˙ 2 k (t) = (x 1 k (t) K 21) n 1 + (x 1 k (t) K 21) n - x 2 k (t) (+ η 2 k x 2 k (t) ) x ˙ 3 k (t) = 1 1 + (x 1 k (t) K 31) n 1 1 + (x 4 k (t) K 34) n - x 3 k (t) (+ η 3 k x 3 k (t) ) x ˙ 4 k (t) = 1 1 + (x 2 k (t) K 42) n (x 3 k (t) K 43) n 1 + (x 3 k (t) K 43) n - x 4 k (t) + D N (t) ∑ j (x 4 j (t) - x 4 k (t) ) (+ η 4 k x 4 k (t) ) x ˙ 5 k (t) = (x 1 k (t) K 51) n 1 + (x 1 k (t) K 51) n - x 5 k (t) (+ η 5 k x 5 k (t) ) In the equation for x i k (t), each parameter Kij is replaced by the epigenetic variable θij (t), which changes over time depending on gene expression levels by introducing epigenetic feedback regulation as a change in the threshold for gene expression as follows: θ ˙ i j (t) = 1 τ e p i (Θ i j - θ i j (t) - α x j (t) ), where Θij is the threshold value after epigenetic fixation and τepi is the time scale of the epigenetic variable. The value of the epigenetic variable θij (t) changes depending on the expression levels of the regulator gene xj; if the regulator gene xj is highly expressed, expression of the regulatee gene xi is promoted, but if expression of the regulator xj is low, the regulatee xi is inhibited. To numerically investigate our model, we set the Hill coefficient as n = 6 and n = 4 in the in four-gene and five-gene models, respectively. The time of cell division tdiv was chosen to be 25. The results of the model do not depend on these parameters, as long as the former is sufficiently large (e. g. , n ≥ 6) and latter not too large (e. g. , tdiv < 1000). The maximum number of cell divisions is 5; hence, the maximum number of cells is 32. For most simulations, we used the parameters Kij as follows: K13 = 0. 78, K34 = 0. 45, K31 = 0. 94, K11 = 0. 35, K21 = 0. 80, K42 = 0. 30, and K43 = 0. 45. In the five-gene model, the additional regulations were K15 = 0. 14 and K51 = 0. 80. The parameters for epigenetic feedback regulation τ and α were set to 2. 0 × 103 and 0. 1, respectively. By using the code written by C/Python simulations were carried out by using standard Runge-Kutta algorithm.

Characterization of pluripotent states, in which cells can both self-renew and differentiate, and the irreversible loss of pluripotency are important research areas in developmental biology. In particular, an understanding of these processes is essential to the reprogramming of cells for biomedical applications, i. e. , the experimental recovery of pluripotency in differentiated cells. Based on recent advances in dynamical-systems theory for gene expression, we propose a gene-regulatory-network model consisting of several pluripotent and differentiation genes. Our results show that cellular-state transition to differentiated cell types occurs as the number of cells increases, beginning with the pluripotent state and oscillatory expression of pluripotent genes. Cell-cell signaling mediates the differentiation process with robustness to noise, while epigenetic modifications affecting gene expression dynamics fix the cellular state. These modifications ensure the cellular state to be protected against external perturbation, but they also work as an epigenetic barrier to recovery of pluripotency. We show that overexpression of several genes leads to the reprogramming of cells, consistent with the methods for establishing induced pluripotent stem cells. Our model, which involves the inter-relationship between gene expression dynamics and epigenetic modifications, improves our basic understanding of cell differentiation and reprogramming.

Abstract Introduction Construction of GRN model Results Discussion Models

2015

Pluripotency, Differentiation, and Reprogramming: A Gene Expression Dynamics Model with Epigenetic Feedback Regulation

Plants have a profound capacity to regenerate organs from differentiated somatic tissues, based on which propagating plants in vitro was made possible. Beside its use in biotechnology, in vitro shoot regeneration is also an important system to study de novo organogenesis. Phytohormones and transcription factor WUSCHEL (WUS) play critical roles in this process but whether and how epigenetic modifications are involved is unknown. Here, we report that epigenetic marks of DNA methylation and histone modifications regulate de novo shoot regeneration of Arabidopsis through modulating WUS expression and auxin signaling. First, functional loss of key epigenetic genes—including METHYLTRANSFERASE1 (MET1) encoding for DNA methyltransferase, KRYPTONITE (KYP) for the histone 3 lysine 9 (H3K9) methyltransferase, JMJ14 for the histone 3 lysine 4 (H3K4) demethylase, and HAC1 for the histone acetyltransferase—resulted in altered WUS expression and developmental rates of regenerated shoots in vitro. Second, we showed that regulatory regions of WUS were developmentally regulated by both DNA methylation and histone modifications through bisulfite sequencing and chromatin immunoprecipitation. Third, DNA methylation in the regulatory regions of WUS was lost in the met1 mutant, thus leading to increased WUS expression and its localization. Fourth, we did a genome-wide transcriptional analysis and found out that some of differentially expressed genes between wild type and met1 were involved in signal transduction of the phytohormone auxin. We verified that the increased expression of AUXIN RESPONSE FACTOR3 (ARF3) in met1 indeed was due to DNA demethylation, suggesting DNA methylation regulates de novo shoot regeneration by modulating auxin signaling. We propose that DNA methylation and histone modifications regulate de novo shoot regeneration by modulating WUS expression and auxin signaling. The study demonstrates that, although molecular components involved in organogenesis are divergently evolved in plants and animals, epigenetic modifications play an evolutionarily convergent role in this process. Differentiated somatic tissues of plants can be reprogrammed to generate various organs, a process called de novo organogenesis. This feature is not only critical for in vitro plant propagation and application of biotechnology, but also provides a good experimental system for understanding regulatory mechanisms underlying organogenesis. Recent studies have revealed some molecular mechanisms underlying de novo shoot regeneration in Arabidopsis [1]–[4], in which WUS, a transcription factor, plays a key role [5], [6]. WUS is a master regulator of stem cell fate determination in shoot apical meristem (SAM), on which many signaling pathways converge [7]. It turned out to be also critical for de novo shoot regeneration. During de novo shoot regeneration in Arabidopsis, expression of WUS is sufficient to specify the organizing center, which is required for stem cell induction and subsequent shoot regeneration [5], [6], [8]. WUS induction is also essential for shoot formation during de novo somatic embryogenesis [9]. Induction of the WUS expression during de novo shoot regeneration was regulated by the master phytohormone auxin [2], [5]. Recently, WUS expression in the organizing center of the Arabidopsis plant SAM was shown to be regulated by epigenetic modifications [10]. Epigenetic modifications, including DNA methylation and histone modifications, occur extensively during cellular differentiation and development in mammals [11]–[13]. In mammals, the patterns of DNA methylation are established by de novo DNA methyltransferase 3 (DNMT3) family and maintained by methyltransferase DNMT1 [14]. DNMT1 plays a vital role in controlling the self-renewal and differentiation of stem cells during hematopoiesis and leukemogenesis and is critical for progenitor maintenance and self-renewal in mammalian somatic tissues [15], [16]. DNA methylation and histone modifications regulate gene expression through changing chromatin structure and transcriptional activities [17]–[19]. For instance, transcriptional repression is associated with hypermethylation of DNA, histone deacetylation and histone H3K9 methylation, whereas active chromatin is linked with hypomethylation of DNA, histone acetylation and histone H3K4 methylation [17], [20]. In plants, pattern changes of DNA methylation and histone modifications leading to changes in chromatin state occur in plant cells undergoing dedifferentiation [21]–[24]. Furthermore, DNA methylation at some promoters is critical for establishing or maintaining the undifferentiated cell state in plants [25]. However, whether and how epigenetic modifications are involved in cell differentiation during de novo shoot regeneration is unknown. Here we showed that mutations of key epigenetic genes altered WUS expression and developmental rates of regenerated shoots in vitro. In addition, epigenetic marks of DNA methylation and histone modifications in the regions of WUS underwent dynamic changes during de novo shoot regeneration, correlating with dynamic WUS expression levels. Genome-wide transcriptional analysis indicated that some genes involved in auxin signaling and meristem development were methylated within the callus, but were demethylated following an induction treatment. Based on these results, we propose that dynamic DNA methylation and histone modifications mediate de novo shoot regeneration in Arabidopsis through WUS and auxin signaling. To find out whether DNA methylation and histone modifications played roles in de novo shoot regeneration, we first compared the capacity and rates of shoot regeneration between wild type and various epigenetic mutants after calli being transferred onto a shoot induction medium (SIM) from a callus induction medium (CIM) [26]. Arabidopsis METHYLTRANSFERASE1 (MET1), KRYPTONITE (KYP), JMJ14 and HISTONE ACETYLTRANSFERASE1 (HAC1), among diverse genes involved in epigenetic modifications, have been well characterized [27]–[31]. MET1 is an ortholog of DNMT1, which maintains DNA methylation directly at CpG motif and indirectly at non-CG motif [27], [32], [33]. Functional loss of MET1 resulted in delayed transition from vegetative phase to reproductive phase [32]. KYP encodes histone H3K9 methyltransferase, and mutation of which resulted in abnormal number of floral organs [28]. JMJ14 encodes histone H3K4 demethylase that inhibited flowering under long-day condition [29], [34]. HAC1 encodes histone acetyltransferase, regulating flowering time through histone acetylation [31], [35]. We used the final percentage of shoot primordia on SIM to reflect the capacity of de novo shoot regeneration, whereas the timely appearance of shoot primordia to reflect their developmental rates. Comparable maximal percentages of shoot primordia were reached after 18 days of incubation on SIM for both wild type and all tested mutants, including met1, kyp, jmj14 and hac1 (Figure 1A–1C), indicating that there was no significant difference in the capacities of de novo shoot regeneration. However, it took different induction time for the wild-type calli and the mutant calli to reach half of the maxima (Figure 1A–1C). Specifically, the mutants whose epigenetic changes were associated with more active transcription, such as met1, kyp, jmj14 [27]–[29], took significantly less time to reach half of the maxima as compared to the wild type (Figure 1A–1C). In contrast, the mutant associated with more repressed transcription such as hac1 took significantly more time to reach half of the maxima (Figure 1C). We obtained similar results indicating precocious or delayed initiation of shoots in these mutants using either pistils or roots as explants (Figure 1A–1C, Figure S1). Interestingly, calli of met1 cultured on SIM develop differently from those of the wild type (Figure 1D). At 4 days on SIM, around 70% met1 calli contained green regions from which the shoots would differentiate, but these green regions could not be identified in the wild-type calli. At 6 to 14 days on SIM, more shoots emerged from the met1 calli than those from the wild-type calli (Figure 1D). At 18 days on SIM, the shoots from the met1 calli were much precocious compared with those from the wild-type calli although the percentages of shoots from both the wild-type and the met1 calli were similar (Figure 1A). We also obtained similar results with roots as explants (Figure S2). Thus, these results indicated that epigenetic modifications, including DNA methylation and histone modifications, played roles in mediating developmental rate of de novo shoot regeneration. It was well established that WUS expression is critical for stem cell formation during de novo shoot regeneration [5], [6]. Here, we showed that induction of wild-type calli on SIM for 4 days (S4) and 6 days (S6) was accompanied by a significant increase of WUS level through qRT-PCR analysis (Figure 2A). In contrast, WUS transcripts were in a low level in wild-type calli on CIM for 16 days (C16) and 20 days (S0, non-induced calli), and similar results were obtained in the prolonged time, such as calli on CIM for 24 days (C24) and 26 days (C26). We further determined the expression patterns of WUS by pWUS: : GUS reporter and in situ hybridization, and the results demonstrated that local distribution of WUS transcripts occurred in wild-type calli on SIM (Figure 3, Figure S3). Because it was shown previously that WUS expression was mediated by epigenetic factors [10], we were tempted to hypothesize that the regulation of WUS expression during de novo shoot regeneration might have resulted from reduced DNA methylation. To test this possibility, we first compared DNA methylation of the ∼10 kb WUS genomic sequences between the calli of wild type on CIM (C16 and S0) and those on SIM (S6) by bisulfite genomic sequencing. Three regions within the WUS genomic sequences were hyper-methylated in S0 calli but substantially decreased in S6 calli (Figure 4A and 4B). Among the three regions, region I was previously proposed to regulate WUS expression [36]. Both CpG dinucleotide motifs and non-CG motifs in the three regions of the WUS genomic sequences showed induced demethylation upon induction on SIM (Figure 4B). These results showed that de novo shoot regeneration was accompanied with demethylation on methylated WUS genomic sequences. That could partially contribute to the regulation of WUS expression during de novo shoot regeneration. Because DNA methylation was significantly reduced in met1 mutant [27], we wondered whether DNA methylation in the WUS genomic sequences would be affected in met1 mutant. To find out, we used two approaches. First, we compared the expression patterns of WUS in wild-type calli and met1 calli at different induction points. Indeed, the met1 mutant showed much higher WUS level than that in the wild type at each time point by qRT-PCR (Figure 2A). Then, in situ hybridization analysis demonstrated that localization of WUS in the met1 calli on SIM was earlier than that in the wild-type calli on SIM (Figure S3A–S3F, Table S1). GUS staining confirmed that the pattern of WUS expression is similar to that in situ hybridization (Figure 3), and the number of GUS signal distribution in both the met1 calli and the wild-type calli on SIM is consistent to percentages of shoot primordia on SIM at different induction points (Figure 3, Figure S3, Table S2). Thus, the results indicated that WUS expression and corresponding developmental rate of de novo shoot regeneration were mediated by reduced DNA methylation. Next, we tested whether MET1 loss of function affected the methylation status of WUS genomic region by bisulfite genomic sequencing. We found that the calli of met1 mutant on CIM (C16 and S0) and on SIM (S6) showed much lower level of DNA methylation in the WUS genomic region than those of wild type under the same condition (Figure 4B). WUS expression was detected in met1 calli earlier than in wild type based on in situ hybridization and GUS reporter analysis (Figure 3 and Figure S3). In addition, met1 contained more WUS-expressing regions than wild type, indicating that increased WUS expression level contributed to elevated the number of organizing centers (Figure 3 and Figure S3). These results suggested that the regulation of WUS expression in met1 mutant during de novo shoot regeneration could at least partially be contributed by DNA demethylation on methylated WUS genomic sequences. Higher WUS level in the met1 mutant suggested the involvement of MET1-mediated DNA methylation in the regulation of WUS expression. However, the expression of WUS still responded to the induction by incubation on SIM in met1 mutant (Figure 2A), indicating additional pathways that regulated the dynamic expression of WUS. Because we showed that histone modifications were also important for de novo shoot regeneration (Figure 2B and 2C), we next tested whether histone modifications played a role in mediating WUS expression during de novo shoot regeneration. We analyzed several histone modifications for the WUS genomic sequences using chromatin immunoprecipitation at two developmental stages: S0 and S6. Methylation at histone H3 at lysine 4 (H3K4me3) was shown to occur in euchromatin undergoing active transcription [37]. Whereas methylation at histone H3 at lysine 9 (H3K9me2) was shown to inhibit transcription [38]. Additionally, acetylation at histone H3 at lysine 9 (H3K9ac) is one of the most characterized epigenetic marks invariably associated with active transcription in all species investigated so far [18]. It also plays a crucial role in plant development [39]. Our results showed that these three histone modifications were dynamically regulated at the WUS genomic sequences during de novo shoot regeneration. Compared with S0, S6 showed an increase in the levels of H3K4me3 at region a and d, but not at b and c (Figure 5A and 5B). H3K4me3 occurred in euchromatin undergoing active transcription [37], therefore increased H3K4me3 levels were consistent with WUS induction during de novo shoot regeneration (Figure 1C, Figure 2C). A mark for chromatin acetylation, H3K9ac, also showed increased levels at all four regions during induction (Figure 5C). In contrast to these epigenetic marks associated with active transcription, H3K9me2, which is associated with transcription suppression [37] were reduced during de novo shoot regeneration in all four regions (Figure 5B). The changes at these epigenetic marks around WUS genomic region explained the active state of WUS chromatin structure, and might well contribute to the regulation of WUS expression during de novo shoot regeneration. Dynamic histone modifications at the genomic regions of WUS during de novo shoot regeneration indicated that histone modifications contributed to regulation of WUS expression during de novo shoot regeneration. To provide further evidence that histone modifications regulated WUS expression in this process, we examined transcript level of WUS in mutants that were defective in histone modifications by qRT-PCR. As stated before, KYP, JMJ14 and HAC1 encoded enzymes for histone modification, mutations of which affected the developmental rate of de novo shoot regeneration (Figure 1B and 1C, Figure S1). Comparing with the wild-type calli, levels of WUS expression in the calli of the mutant kyp-2 were significantly enhanced compared to those of wild type for 6 days on SIM (Figure 2B). Similar results were obtained for the mutants jmj14-1 and jmj14-2 (Figure 2C). Contrast to the mutants kyp and jmj14, the levels of WUS transcripts in two different allelic hac1 mutants were reduced compared to that of wild type (Figure 2C). Then, we used kyp-2 calli on SIM (S0, S4, and S6) to do in situ hybridization analysis. The results showed that localization of WUS signals in kyp-2 calli on SIM occurred early comparing to that in wild-type calli on SIM (Figure S3G–S3L). Also, the number of localized WUS signals in kyp-2 calli on SIM (S4 and S6) was more than that in wild-type calli at the same time points (Table S1). Similar to the case of met1, expression of WUS appeared earlier in kyp-2 calli than in wild type (Figure S3). Thus, changes of WUS expression in these mutants correlated with their different developmental rates of de novo shoot regeneration, suggesting that WUS expression was regulated by histone modifications. Our results showed that DNA methylation and histone modifications regulated WUS expression during de novo shoot regeneration. To get a whole picture of epigenetic modifications during this process, we decided to do a genome-wide expression profiling using the Affymetrix ATH1 full genome array. We analyzed the transcriptomes of wild-type calli being transferred to CIM for 20 days (S0) and to SIM for 6 days (S6). Because met1 calli showed significantly different developmental rate from wild-type calli, we also analyzed transcriptomes of met1 calli being transferred to CIM for 20 days (M0) for comparison. Significance Analysis of Microarrays software package analysis was conducted for three biological samples replicates between the Ws and met1. The q value≤0. 05 and fold change ≥2 were used as the threshold for candidate gene selection (Figure 6A). This criterion gave 1334 upregulated genes, and 501 downregulated genes by induction on SIM (S6 versus S0) (Table S3). 768 candidate genes showed over 2 fold difference between M0 and S0, suggesting that they might be regulated by MET1-dependent DNA methylation (Table S4). 308 candidate genes showed over 2 fold difference both between S6 versus S0 and between M0 versus S0, suggesting that they might be induced on SIM and be regulated by MET1-dependent DNA methylation (Table S5). By qRT-PCR analysis, we confirmed the microarray data (Figure S4). Because auxin and cytokinin are essential for de novo shoot regeneration [2], [5], we selected genes involved in cytokinin and auxin signaling for bisulfite sequencing analysis. Indeed, some displayed differential methylation patterns during de novo shoot regeneration, such as AUXIN RESPONSE FACTOR3 (ARF3), AUXIN RESPONSE FACTOR4 (ARF4), INDOLE-3-ACETIC ACID INDUCIBLE18 (IAA18) and BELL1-LIKE HOMEODOMAIN7 (BLH7) (Figure 6B–6E). A loss of DNA methylation occurred in these genes, along with increased levels of their transcription in induced wild-type calli (Figure S4). Their expression levels were also higher in met1 than those in the wild type, suggesting that the expression of these genes might be regulated by a MET1-dependent dynamic DNA methylation during shoot regeneration. On the other hand, some candidate genes selected from SIM-induced and MET1-dependent pathways displayed no methylation, such as ASMMETRIC LEAVES1 (AS1), ARABIDOPSIS RESPONSE REGULATOR15 (ARR15), CYTOKININ OXIDASE/DEHYDROGENASE1 (CKX1), INDOLE-3-ACETIC ACID27 (IAA27) and PINOID2 (PID2), but they displayed great changes in their transcriptional levels upon SIM-induction, implying that those genes might not be directly regulated by MET1 (Table S5). DNA methylation and histone modifications are critical epigenetic processes that control chromatin structure and gene expression during development and differentiation [17], [18], and there are likely complicated interactions between these processes [20], [40]. In human, a crosstalk between DNA methylation and histone modifications has been proposed to regulate gene transcription in tumors [20]. Similarly, DNA methylation controls histone H3K9 methylation and further affect heterochromatin assembly in Arabidopsis [41]. Recent study has indicated that chromatin status facilitates the accessibility of transcription factor to FLOWERING LOCUS T (FT) in Arabidopsis, and distant regulatory regions are required for FT transcription [42]. WUS transcription is regulated through a fairly complicated chromatin remodeling mechanism in the SAM of the Arabidopsis plant [43]. It was shown that WUS expression was positively correlated with FASCIATA1 (FAS1) /FAS2, subunits of ASSEMBLY FACTOR-1 (CAF-1), and BRUSHY1 (BRU1), both of which regulate post-replicative stabilization of chromatin structure [44], [45]. Another study showed that the chromatin remodeling factor SPLAYED (SYD) directly regulated WUS to maintain proper WUS transcript levels in its spatial expression domain [46]. It has been demonstrated that at least 3. 5 kb fragment upstream of WUS is required for its spatiotemporal expression during plant development [36]. Here, we showed that the 5′ and 3′ regions of WUS were regulated by SIM-induced changes of DNA methylation and histone modifications. Because the met1-3 kyp-7 double mutant displayed more severe phenotypes than each single mutant [19], we propose that regulation of WUS by DNA methylation and histone modifications may function in a partially redundant manner during de novo shoot regeneration. To understand mechanism of the in vitro organogeneis mediated by the factors involved in both DNA methylation and histone modifications, knocking out both DNA methylation and histone modifications remains to be investigated in the future. It has long been thought that animal cells, once committed to a specific lineage, can no longer change their fate. However, recent studies suggested that differentiated animal cells do maintain plasticity and can be induced to undergo reprogramming [47], [48]. Further studies have shown that differentiated cells in mouse can be reprogrammed to pluripotent stem cells by introducing four transcription factors [49]. Plant cells can easily regenerate organs from the differentiated tissues under proper cultured conditions [1]. Previously, we used Arabidopsis ptstils as explants on CIM to obtain the callus, a mass of pluripotent cells [26], and by transferring calli onto SIM, the expression of WUS was induced in a group of cells termed the organizing center as a self-renewing source of stem cells within calli. The induced organizing center and stem cells were responsible for subsequent shoot regeneration. Here, we showed that expression of many genes was induced by SIM-induction (Figure 6A). Those genes were divided into either MET1-dependent or MET1-independent. Among MET1-dependent genes, WUS is a key transcription factor to regulate shoot regeneration [1]. ARF3 was required for shoot induction (Cheng et al. , unpublished data). Previous study showed that ARF3 and ARF4 act redundantly to establish the abaxial cell fate of the Arabidopsis leaves [50]. Thus, ARF3 and ARF4 may function on de novo meristem formation mediated by epigenetic modifications. MET1-independent genes might also be involved in the process of shoot induction. Our results suggested that pluripotent cells of the callus can be reprogrammed to stem cells and subsequent, shoot formation through the regulation of both MET1-dependent genes, such as WUS and ARFs, and some MET1-independent genes. In conclusion, our results indicate that dynamic DNA methylation and histone modifications contribute to the control of stem-cell formation and subsequent shoot regeneration. These epigenetic modifications regulate WUS and probably hormone-related genes, whose spatiotemporal expression was critical for de novo shoot regeneration. In mammals, epigenetic modifications of transcription factors and of components in hormone signaling pathways also play crucial roles in cell differentiation and organogenesis [51], [52]. Our results thus provide an interesting scenario in which epigenetic modifications were adopted as recurring themes during evolution for de novo organogenesis. The met1 mutant in the Wassilewskija (Ws) background was a kind gift from Dr. J. Bender (The MCB Department of Brown University) [27]. The kyp-2 [28] mutant in the Landsberg (Ler) background, jmj14-1, jmj14-2 [29], hac1-3, and hac1-5 [31] mutants in the Columbia (Col) background were generously provided by Dr. Xiaofeng Cao (Institute of Genetics and Developmental Biology, Chinese Academy of Sciences). Plants were grown as previously described [9]. Arabidopsis seeds were surface sterilized and plated on germination medium [53]. After cold treatment for 2 days at 4°C in the dark, they were transferred to sterile conditions or the growth chamber at 22°C in a 16 h light/8 h dark cycle. Shoot regeneration procedures used in this study were based on the previously described protocols [26], [54]. Pistils were excised from sterile Arabidopsis plants and transferred onto callus induction medium (CIM, MS medium [53] with 0. 5 mg/L 2,4-dichlorophenoxyacetic acid (2,4-D) and 1. 0 mg/L 6-benzylaminopurine (6-BA) ). The explants were incubated for 20 days on CIM to induce callus production, and calli were then transferred onto shoot induction medium (SIM, MS medium with 0. 01 mg/L indole-3-acetic acid (IAA) and 2 mg/L zeatin (ZT) ). Root explants of 5–10 mm length were excised from 7-day-sterile seedlings, then transferred onto callus induction medium (CIM, Gamborg' s B5 medium [55] with 0. 5 g/L MES, 2% glucose, 0. 2 µmol/L kinetin, and 2. 2 µmol/L 2,4-dichlorophenoxyacetic acid (2,4-D), 0. 8% agar), and incubated for 6 days in continuous light. Finally, explants were transferred onto shoot-inducing medium (SIM, Gamborg' s B5 medium with 0. 5 g/L MES, 2% glucose, 0. 9 µmol/L 3-indoleacetic acid, 0. 5 µmol/L 2-isopentenyladenine) and incubated in continuous light. The morphology of calli was examined and photographed with an Olympus microscope. We defined the number of regenerated shoots as the number of at least 2 mm long shoots on each callus. Probes were labeled using digoxigenin RNA labeling kit (Boehringer Mannheim). An antisense probe from a full-length WUS cDNA clone was generated using T7 RNA polymerase, and a sense probe was synthesized using SP6 RNA polymerase. The detailed protocol was carried out as described previously [56]. Primer sequences used for probes amplification are summarized in Table S6. Plant tissues were incubated in GUS assay solution (50 mmol/L Na2HPO4,50 mmol/L KH2PO4, pH 7. 2,10 mmol/L Na2EDTA, 0. 5 mmol/L K3Fe (CN) 6,0. 5 mmol/L K4Fe (CN) 6,1% Triton X-100 and 2 mmol/L X-Gluc (Bio. Basic Inc. , Canada) ) at 37°C for 12 h. To further investigate WUS expression pattern, some GUS-stained tissues were embedded in paraffin (Sigma) and sectioned. To display the outline of cells clearly, ruthenium red (200 mg/L) was used to stain cell walls. DNA methylation assays were performed by bisulfite sequencing as previously described [57]. PCR products were cloned into the pMD19-T Simple Vector (Takara), and 12 clones were sequenced to determine the methylation status of a locus in each genotype. Primer sequences are shown in Table S6. Bisulfite sequencing data were analyzed by the CyMATE software [58]. The results returned by CyMATE were input into SigmaPlot 10. 0 to illustrate DNA methylation frequencies at CG, CHG and CHH (where H = A, C or T) at the various cultured stages of each genotype. The Arabidopsis calli grown on CIM for 20 days (S0) and on SIM for 6 days (S6) were vacuum-infiltrated with formaldehyde crosslinking solution. Chromatin immunoprecipitation was performed according to manufactures' instructions (Epigentek Group Inc. USA, Catalogno. P-2014). Chromatin samples were immunoprecipitated with antibodies against a negative control normal mouse IgG and H3 dimethyl Lys 9 (both included in EpiQuik™ Plant ChIP Kit), or with antibodies against H3 trimethyl Lys 4 (Abcam USA, Catalogno. ab1012) and H3 acetyl Lys 9 (Abcam USA, Catalogno. ab10812). PCR amplification was performed in 25 µL volumes for 32 to 37 cycles to determine the appropriate conditions for the PCR products of each region. Primer sequences are shown in Table S6. The PCR products were electrophoresed in a 2% agarose gel. Three biological replicates were analyzed and each was tested by three technical replicates. Total RNAs were isolated from callus tissues 2 to 3 mm deep from the surface. Quantitative real-time PCRs (qRT-PCRs) were performed as described previously [9]. To check the specificity of amplification, the melting curve of the PCR products was detected. The expression levels of specific genes were standardized to the housekeeping gene TUBULIN2. Each reaction was carried out in three biological replicates. The relative expression level of each gene, corresponding to the expression level of TUBULIN2, was calculated using the comparative CT method [59]. Primer sequences used for qRT-PCR are summarized in Table S6. RNA of three plant samples was prepared from each of the following tissue types: the wild-type calli cultured on CIM for 20 days (S0), and on SIM for 6 days (S6); the met1 mutant calli cultured on CIM for 20 days (M0). RNA purification, probe labeling, chip hybridization, probe array scanning and data pre-processing normalization were performed by the Affymetrix custom service (CapitalBio, Beijing, China). Significance Analysis of Microarrays software package analysis was conducted for three biological samples replicates between the Ws and met1. When all replicates clustered together, further analysis was performed based on mean values. A two-fold change in the gene expression levels between one versus another samples with a q value≤0. 05 was set as the threshold for altered gene expression. Microarray data are available in the ArrayExpress database (www. ebi. ac. uk/arrayexpress) under accession number E-MEXP-3120.

Plants have a strong ability to generate organs from differentiated somatic tissues. Due to this feature, shoot regeneration in vitro has been used as an important way for producing whole plants in agriculture and biotechnology. Phytohormones and the transcription factor WUSCHEL (WUS) are essential for reprogramming during de novo shoot regeneration. Epigenetic modifications are also critical for mammalian cell differentiation and organogenesis. Here, we show that epigenetic modifications mediate the de novo shoot regeneration in Arabidopsis. Mutations of key epigenetic genes resulted in altered WUS expression and developmental rates of regenerated shoots in vitro. Bisulfite sequencing and chromatin immunoprecipitation revealed that the regulatory regions of WUS were developmentally regulated by both DNA methylation and histone modifications. By transcriptome analysis, we identified that some differentially expressed genes between wild type and met1 are involved in signal transduction of the phytohormone auxin. Our results suggest that DNA methylation and histone modifications regulate de novo shoot regeneration by modulating WUS expression and auxin signaling. The study demonstrates that, although molecular components involved in organogenesis are divergently evolved in plants and animals, epigenetic modifications play an evolutionarily convergent role during de novo organogenesis.

Abstract Introduction Results/Discussion Materials and Methods

genome-wide association studies developmental biology genetic mutation gene expression genetics epigenetics biology dna modification genetics and genomics cell differentiation dna transcription histone modification

2011

DNA Methylation and Histone Modifications Regulate De Novo Shoot Regeneration in Arabidopsis by Modulating WUSCHEL Expression and Auxin Signaling

Biological systems evolved to be functionally robust in uncertain environments, but also highly adaptable. Such robustness is partly achieved by genetic redundancy, where the failure of a specific component through mutation or environmental challenge can be compensated by duplicate components capable of performing, to a limited extent, the same function. Highly variable environments require very robust systems. Conversely, predictable environments should not place a high selective value on robustness. Here we test this hypothesis by investigating the evolutionary dynamics of genetic redundancy in extremely reduced genomes, found mostly in intracellular parasites and endosymbionts. By combining data analysis with simulations of genome evolution we show that in the extensive gene loss suffered by reduced genomes there is a selective drive to keep the diversity of protein families while sacrificing paralogy. We show that this is not a by-product of the known drivers of genome reduction and that there is very limited convergence to a common core of families, indicating that the repertoire of protein families in reduced genomes is the result of historical contingency and niche-specific adaptations. We propose that our observations reflect a loss of genetic redundancy due to a decreased selection for robustness in a predictable environment. Living organisms evolved to be functional in frequently harsh and variable environments, buffering internal molecular noise, genetic variation and unpredictable environmental fluctuations. Such ability is termed robustness [1]. One common source of robustness is genetic redundancy, in which one or more genes can perform the same function [2]. The exact contribution of genetic redundancy to the robustness of biological systems has, however, been a subject of considerable debate. On the one hand, it is hard to understand how full redundancy can be evolutionarily stable. After duplication the two copies will have identical functions and the loss of one by the accumulation of mutations is buffered by the other, having no fitness cost [3]. On the other hand, there is strong evidence for functional redundancy by duplicates. The deletion of singleton genes, i. e. , those without copies, is frequently lethal [4]. In contrast, deletion of genes with paralogues has frequently little fitness cost [4], even though the deletion of pairs of paralogues has frequently high fitness costs [5], suggesting that their compound function is essential, and arguing for functional redundancy of the paralogues. The capacity for functional compensation correlates with sequence divergence, with closer paralogues more likely to provide it [4], which argues for gene duplication providing functional redundancy. This redundancy can in fact be maintained over large periods of time, as two independent studies of functionally redundant duplicates showed recently [6], [7]. A theoretical analysis of the metabolic network of S. cerevisiae estimated that the dispensability of up to 28% of metabolic enzymes can be attributed to the existence of a compensating paralogue [8]. Recent work has suggested that the cost of maintenance of complete redundancy can be, to some extent, offset by partial functional redundancy [9]. Furthermore, incomplete and presumed functionally distinct duplicates may also provide additional backup [10]. The discussion on the role of genetic redundancy in robustness is also centered on the conditions for the emergence of robustness. A series of theoretical studies have resulted in the prediction that high robustness can only evolve in the presence of frequent perturbations (reviewed in [11]). Little attention has been given to the conditions necessary for the loss of robustness. Based on the above, we would anticipate that predictable environments should not place a high selective value on robustness. The intracellular environment is relatively invariant over time. Organisms that occupy this ecological niche are not subjected to repeated nor frequent perturbations, and represent a good system to study adaptation to such predictable environments. The rapid increase in the number of sequenced intracellular endosymbionts and parasites provides an ideal system to study the evolution of genetic redundancy, and for an empirical study on the importance of external perturbation in the emergence of robustness. Intracellular lifestyles have been frequently and independently adopted by bacteria and eukaryotes, in the context of endosymbiosis or parasitic relationships [12]–[16]. Obligate intracellular parasites and endosymbionts have committed to an intracellular lifestyle, only capable to replicate inside a host eukaryotic cell [14]. They include organisms like Buchnera aphidicola, a bacterial endosymbiont of aphids and also parasitic, pathogenic bacteria, such as Mycobacterium leprae and Rickettsia prowazekii, the causative agents of leprosy and typhus, respectively. The adaptation to the intracellular niche is invariably accompanied by extensive gene loss [12]–[14]. Reduction in gene repertoires is believed to be associated with adaptation to a new lifestyle where many molecules can be obtained from the host [17]. Since the host provides metabolites, many loci in the endosymbiont/parasite would become redundant and previously deleterious mutations would become de facto neutral, due to relaxed selection. Examples are the loss of biosynthetic pathways in many endosymbionts (e. g. [15]). A second driver of gene loss is the drastic reduction in effective population sizes [18]–[20], associated with high mutation rates [21]. Furthermore, inheritance modes of intracellular bacteria imply that only few individuals are transmitted across generations and/or hosts, generating repeated population bottlenecks [15]. Even “important” genes involved in DNA repair, transcriptional regulation and replication have been lost in Buchnera, suggesting that drift plays an important role in genome reduction [22]–[25]. Extreme reductive genome evolution is also observed in obligate parasitic bacteria like the Mycoplasmas, which are often described as the simplest self-replicating organisms [26]. These organisms are obligate parasites of vertebrates, living under an invariant environment within the hosts. We consider these organisms, together with obligate intracellular parasites and endosymbionts, as “Reduced genomes” living under predictable environments. Here we study the dynamics of gene loss in Reduced genomes, investigating which genes can be lost, and find a previously undescribed driver for gene loss. By combining data analyses with evolutionary simulations we find empirical evidence for a selective drive to maintain diversity of protein families at the expense of family size, with the emergence of many genes without any paralogues. We propose that the latter represents a loss of genetic redundancy due to a decreased selection for robustness in a predictable environment. Protein families represent groups of proteins that share a common evolutionary history [27]. Within protein families there is conservation of structure and biochemical function across large evolutionary distances [28]. The number of protein families can be construed as the degree of information coded in a genome – the more distinct families exist, the more information. Early analysis of a small number of completely sequenced genomes suggested that larger genomes have more protein families than smaller ones [29]–[31], and there is, in fact, a linear relationship between the number of genes and number of protein families [29]. Larger genomes also tend to have larger protein families [30], [31]. Furthermore, intracellular parasites and endosymbionts that have the smallest genomes known, also have the smallest gene families [31]. With the accumulation of completely sequenced genomes of bacterial parasites and endosymbionts we can now address whether these reduced genomes living under nearly constant environments display the same use of protein families. We chose to define protein families based on structural domain architectures [32], which provides a higher sensitivity than other sequence-based methods [33] and allows us to capture distant evolutionary relationships. Members of each family should be traceable to a common ancestor by duplication and speciation [27], [34]. Note that in bacteria, Lateral Gene Transfer is frequent and generates copies of genes (xenologues) that are indistinguishable from copies resulting from duplication (paralogues) [35]–[37]. For the purpose of this analysis, their specific origin is not relevant and we use the term paralogue loosely to include both. We studied 69 bacteria that have undergone extensive reductive genome evolution that we label “Reduced”, consisting of the obligate parasitic mycoplasmas and obligate intracellular parasites and endosymbionts, and 308 Free living bacteria, which we label “FL”. In our analysis these two classes are mutually exclusive and their genome size distribution significantly different (Figure 1A). Species name are provided in tables S3 and S4 in Text S1. As expected we observed a strong positive correlation between the number of genes and families (Spearman' s rank correlation ρ = 0. 97). We noted however that there were two statistically distinct trends in FL and Reduced organisms (Figure 1B). Reduced genomes have more families than would be expected if they were part of the FL. The same trend is observed when we consider individual protein domains instead of protein families (Figure S1 in Text S1). Because the number of genes and families in the two populations are very different and hence difficult to compare, we tested the potential difference between the two populations of organisms by estimating the elasticity of each population, a measure that captures the responsiveness of a function to parameters in a relative scale. The elasticity of Families in Reduced genomes is two times higher when compared to FL. In other words, adding one gene is 50% more likely to drive a number of families increase in Reduced than in FL. Technically, a 1% change in the number of genes will determine a variation of 0. 73% in the number of families, compared to a variation of 0. 48% in FL genomes. Thus FL genomes are more robust to gene number variation than are Reduced genomes. Smaller genomes, such as those found in intracellular parasites and endosymbionts, were previously shown to have smaller families [30], [31]. Our results reveal that Reduced genomes had smaller families than could be expected if they followed the same trend as the FL genomes, in particular, they had a significantly higher number of singletons, i. e. families of size one (Figure 1C - note that family size has been subjected to a high pass filter - see methods for details). These results hold when these comparisons are made only for organisms within the same order, which suggests that phylogenetic distance is not an important bias in this result (Figures S2, S3, S4 and S5 in Text S1). A simple averaging of the fraction of singletons in both populations illustrates this trend well - 22% of the families in FL are singletons, but this number rises to 48% in Reduced genomes (p<2. 2×10−16; Mann-Whitney U test, Figure 2C). Another way to look at the same problem is to compute the number of genes in paralogous families [31] (Figure S6 in Text S1). As before, phylogenetic distance does not bias this result (Figure S7 in Text S1). Note that although gene loss is the dominant force accounting for the difference in size between Reduced and Free Living families, there is also gene duplication in Free living organisms, and what we measure is the compound signal of both FL duplication and Reduced loss. Taken together these observations indicate that the reduced genomes are not a random sample of the FL genomes. They have relatively more distinct protein families than FL genomes but less elements per family, which suggests a selective drive to keep diversity of protein families at the expense of redundancy. This is the hypothesis we will test here. In order to claim that protein diversity or redundancy are selectively lost and/or retained we first need to determine whether this is not the outcome of a neutral process, or that it is not the byproduct of a selective drive on some other character. To address these points we modeled gene loss under a variety of scenarios. We considered two scenarios modeling neutral gene loss and two capturing functional selection. The details of the simulation are described in the methods section, and summarized in Figure 2A. In short, we randomly sample the FL genomes and then simulate gene loss up to a final genome size according to predefined scenarios, where the key variable between scenarios is the probability of losing each gene. We run the simulations 10,000 times for each scenario, creating populations of simulated reduced genomes that we then compare with the Reduced set. We first simulated two independent scenarios of neutral gene loss. In the first scenario (S1) genes to be lost are randomly sampled and have a constant probability of loss that corresponds to the average difference in number of genes in the genome between FL and Reduced genomes. A second, more sophisticated scenario accounts for the fact that longer genes may receive more mutations, which we simulate in scenario S2 by tying the probability of gene loss to its size. Neutral loss would result in significantly lower protein family diversity than observed in the Reduced genomes (Figure 2B). For example, a Reduced genome with 1000 genes would have 510 families, whereas simulated genomes under scenarios S1 and S2 would have 449 and 398 families respectively. Moreover, neutral loss would result in significantly fewer singletons than we observe in Reduced genomes, i. e. higher genetic redundancy (<39%, compared to 48% in Reduced - Figure 2C). These results hold even when we we perform the simulations within the same bacterial order, which indicates that our results are robust to the large phylogenetic distances considered (Figure S8 in Text S1). From this we conclude that neutral gene loss alone cannot account for the observed diversity of protein families, nor for the reduced genetic redundancy. Rejection of a neutral scenario is suggestive of selection but does not allow us to determine what is being selected. In other words, we cannot state that there is selection for protein family diversity or against redundancy as it is altogether plausible that there is selection on some completely unrelated character and what we observe is the byproduct of that selective drive. The genes preferentially conserved could be enriched in specific protein families, thus biasing our results. We now consider the major factors that can constrain gene conservation, and by extension its loss. We now investigate the possibility that there is preferential retention of a subset of genes on some functional grounds that incidentally result in retention of protein diversity. We first consider that Essential genes may define such set of genes that are preferentially retained, where essential genes are defined by having a lethal gene deletion phenotype. Essential genes in bacteria are preferentially retained in evolution [38], [39]. In eukaryotes essential genes have a lower probability of being lost in the context of lineage-specific gene loss [40]. We observed, as expected, that essential genes in E. coli are preferentially conserved in bacterial parasites and endosymbionts (Figure S9 in Text S1). Note that these genes can still be lost, as is well illustrated by experimental evolution studies of genome reduction in Salmonella enterica where essential genes were in fact lost [41]. In scenario S3 we thus preferentially keep protein families that have essential genes in E. coli, i. e. we consider essentiality a property of the family [42]. Although genes are lost in all categories, some functional classes are preferentially lost and others preferentially retained [13], [15], [43]. We calculated the functional class distributions in both populations, and observed several statistical significant differences, for example a preferential retention of genes annotated to the functional class Translation (Figure S10, Table S1 and S2 in Text S1). In scenario S4 we preferentially retain protein families annotated to the most abundant functional classes in the Reduced genomes. Proteins do not work in isolation but they establish interactions and form pathways, and this could constraint the probability of gene loss. We consider participation in metabolic pathways as these can be inferred from sequence alone with reasonable confidence, and physiological coupling in pathways was shown to be a constraint in reductive genome evolution, i. e. coupled reactions are more likely to be lost together [44]. We simulate gene loss in a scenario where once a member of a pathway is lost, the probability of losing other members of the pathway increase three-fold (S5). Protein-protein interactions may also play a role in gene retention, however we lack the data to address these interactions, and it is unclear at which evolutionary distances it is safe to transfer protein-protein interactions. Furthermore, there is conflicting evidence regarding the role they can play in gene loss. Ochmann and co-workers found that poorly connected proteins are more likely to be lost in the evolution of γ-proteobacteria [45], while Tamanes and co-workers found that in the reductive evolution of Buchnera aphidicola APS, gene loss did not correlate with the absolute number of links of a protein in the protein interaction network (some hubs were more likely to be preserved than others), nor did they observe any drive to keep functional modules intact [46]. We consider a final scenario where gene positioning can determine the likelihood of a gene being lost, as larger deletions could simultaneously delete more than one gene. This has been in fact proposed to be frequent for example in the evolution of B. aphidicola [47] and of Burkholderia mallei [48], even though other studies suggests that loss of individual genes may also be frequent [49]. Note that the organization of bacterial chromosomes in operons makes this scenario also pertinent to understand functional constraints to gene loss, as genes that are part of the same operon likely code to proteins that are functionally associated, as part of the same pathway, complex or directly interacting with each other, and gene order is frequently conserved [50]. We thus modeled a final scenario (S6) where once a gene is lost, adjacent genes become twice as likely to be lost. Comparison of these selective loss scenarios with the Reduced genomes indicates that selection based on predicted essentiality, functional classes, co-participation on predicted metabolic pathways or adjacency in the genome cannot account for the increased protein family diversity observed, which is substantially higher than observed in the simulations. Using the same example as above, simulated genomes with 1000 genes would have S3 = 464, S4 = 453, S5 = 386 and S6 = 431 families, compared to the 510 families in Reduced genomes. Furthermore, none of these simulations can produce singleton numbers as high as observed in reduced genomes (S3 = 43%, S4 = 39%, S5 = 34%, S6 = 37% compared to 48% in Reduced - Figure 2C. Thus, although all the factors we tested can constraint gene loss, our simulations indicate that they cannot account for the protein family diversity nor the reduction in genetic redundancy we observe in Reduced genomes. Genome reduction happened multiple independent times in the course of evolution, but it is plausible that there is convergence to a particular small set of genes necessary for parasitic or endosymbiotic life. Such convergence to a minimal gene set could represent a constraint to gene loss accounting for some of the protein family diversity we observe in the Reduced genomes. Previous attempts to define minimal gene sets compatible with cellular life using orthology, revealed a small number of genes [26], [51]. This lead to the proposal that non-orthologous gene displacement, where the same function is performed by unrelated or very distantly related, non-orthologous proteins [52], was far more important than previously anticipated [26]. In fact a comparison of the shared homologous protein coding genes between endosymbionts and the parasite Mycoplasma genitalium revealed a small set of 175 homologous groups that could represent the minimal core for cellular life [24]. Using a sensitive protein family detection method we now ask whether we can detect the convergence to a common set of protein families in the genomes we analyzed here. This could represent a minimal core of families necessary for parasitic and/or endosymbiotic life. We found that only a small proportion (8%) of the families observed in Reduced genomes is present in more than 90% of the organisms (118 protein families in 1433). Similarly, only 4% of the FL families are present in more than 90% of these organisms (293 out of 7405 - Figure 3). The 118 families common to Reduced organisms are a subset of the families common to FL organisms. This suggests that the common core of families necessary for parasitic or endosymbiotic life is a subset of those necessary for free life. Note that only 43% of the protein families retained in most Reduced genome are essential in E. coli (51/118), which further strengthens the idea that each ecological niche requires distinct sets of proteins families. Note that this small number is not due to the existence of two distinct life styles in the Reduced group, as when we break this group into parasites and endosymbionts, we observe a only marginal increase in the number of families that are present in more than 90% of the organisms (132 in parasites and 162 families in endosymbionts). In contrast, we find that most families are present in less than 10% of the organisms. In both populations the majority of the protein families falls into this group, but these “unique” families are more common in FL (84%) than in Reduced genomes (52%). From this we can extrapolate that although niche- and taxon-specific adaptations dominate Reduced genomes, they are comparably less important than in FL organisms. Thus, convergence to a common core of protein families does not appear to be a major force shaping the protein family diversity in reductive genome evolution. Our results so far are compatible with a scenario where there is a selective drive to retain a minimal set of families compatible with life in the specific niche occupied by the organism, and that this includes a small core of families common to all reduced genomes, as well as retention of specific functions. This results in the measured increase in protein family diversity in Reduced genomes. However, none of the neutral and selective scenarios we modeled or analyzed above can account for the marked reduction in protein family size, in particular the increase in the number of singletons in Reduced genomes. We hypothesize that this observation may be explained by loss of genetic redundancy, i. e. when more than one gene can perform the same function in a free living organism (larger families), those copies will be lost in the course of reductive genome evolution up to a point where only a single gene per function is retained (singletons). There are abundant anecdotal evidence that supports this hypothesis. For example, most Bacteria have two peptide chain release factor proteins with partial overlap in codon specificity (PrfA: UAG, UAA; PrfB: UGA, UAA). Legionella Pneumophila, a pathogenic γ-proteobacteria that is a facultative parasite, has even a third member of this family (lpg0167); in contrast the related intracellular parasite Coxiella burnetii, the causative agent of Q fever, retained only PrfB. This scenario requires then that larger families are more likely to lose genes than smaller ones. We tested this hypothesis and found that the probability of gene loss in families present in most organisms is positively correlated with family size (Spearman' s rank correlation ρ = 0. 74) (Figure 4). This relationship is best approximated by an inverse function (r2 = 0. 55), which suggests that the probability of gene loss is essentially random for larger families, but as families become smaller it decreases sharply, with small families having very small probabilities of gene loss. The probability of a gene being lost thus depends on the number of paralogues it has. In neutral scenarios such positive correlation is absent (ρS1 = 0. 04, ρS2 = −0. 32), and is also absent in the scenario where we retain specific functional classes (ρS4 = 0. 07), members of the same pathway (ρS5 = 0. 06) or adjacent genes (ρS6 = 0. 06). In scenario S3 we observed a correlation between family size and probability of gene loss (ρS3 = 0. 74), but inspection of the data in Figure 4 suggests that this is an artifact resulting from hardwiring two distinct levels of Probability of loss in the simulation. Are the genes being lost those that were functionally redundant with their paralogues? Anecdotal evidence suggest that this is the case. There are for example at least seven Cof-like phosphatases in E. coli (Cof, YidA, YbhA, YigL, YbiV, YbjI, YedP), with substantial overlap in their substrate specificities. In contrast, the endosymbiont Candidatus Blochmannia pennsylvanicus has a single gene assignable to this family (YigL), which is predicted to maintain 4 out of the 5 substrates that the different E. coli enzymes are known to process [53], [54]. Is this a general case? To answer this question we struggle with the absence of extensive functional information for most of the organisms studied here, the varying phylogenetic distances between these organisms and difficulty of large-scale mapping of orthologues in paralogous families. We first seek to address the issue of functional redundancy in a way that does not require such mappings nor functional information, by focusing on the most similar pairs of paralogues [4], [10]. The rational of our experiment is the following: if the most similar pairs of paralogues in a protein family are the ones that are more likely redundant, then one member of the pair will be preferentially lost in Reduced genomes, resulting in a decrease in the similarity between the pairs of paralogues in the family. Thus, we computed the sequence similarity between the pairs of closest paralogues for each family and within each genome (Figure 5A). We observed that the pairs of closest paralogues in the families in Reduced genomes are significantly less similar than those in the FL genomes (Figure 5B). We detected a reduction in the similarity of the closest paralogues in nearly 90% of the protein families (Figure 5F). Note that this is not an artifact of the increased sequence divergence in Reduced genomes, as we control for this - in fact, the overall sequence similarity within Reduced families is higher than in the same FL families (not shown). Furthermore, those families that did not reduce in size do not display this reduction in similarity (Figure S11 in Text S1). Additionally, the difference in family size could bias this analysis, but when we control for it we show that the reduction in similarity still holds (Figure S12 in Text S1). This analysis is also potentially biased by phylogenetic distance between organisms compared and different sizes of the universes being compared. However, when we consider specific pairs of phylogenetically close FL and Reduced organisms, i. e. one-to-one comparisons, we find the same trend (Figures 5C, D, E, G, H, I). Thus, reductive genome evolution results in the increasing of the distance between the closest paralogues, which we interpret as evidence that there is preferential loss of one of the pair of closest paralogues. One example of this scenario is the protein family that includes in E. coli the two redundant transketolases TktA and TktB (E. C. 2. 2. 1. 1) [55], [56], as well as the functionally distinct Dsx (1-deoxyxylulose-5-phosphate synthase, E. C. 2. 2. 1. 7). TktA and TktB are 99% identical, but only 29% identical to Dsx. In the closely related B. aphidicola, only one transketolase was retained (Tkt), together with the Dsx ortholog - they are ∼13% identical. Finally, we used predicted enzymatic functions to further investigate the loss of functional redundancy. We considered enzyme function predicted in KEGG [57], and described by E. C. numbers. This is a hierarchical classification of enzyme function, that describes enzyme function and substrate specificity. Two proteins that have the same E. C. number have the same function. In Figure 6A we show that when comparing phylogenetically close Reduced and FL genomes, the former have less enzymes that map to the same predicted E. C. number, which is consistent with the notion that in reductive genome evolution there is a drive to retain a single copy of each function. This is not simply a consequence of genome reduction, as when we simulate gene loss under a neutral scenario (S1 in Figure 2), using a closely related Free Living genome as a starting point of the simulation, we always obtain artificially reduced genomes with more proteins per E. C. number, i. e. more redundant, than observed in the Reduced genomes (Figure 6B). Our results show that organisms that suffered extensive genome reduction in response to adaptations to predictable environments maintain a higher than expected protein family and protein domain diversity, and concomitantly lost genetic redundancy. The excess diversity at the protein family and protein domain level that characterizes the reduced genomes cannot be accounted by a neutral scenario nor does it appear to be the by-product of selection on other characters. These families observed in Reduced genomes differ from organism to organism, and only 8% of these are present in more than 90% of the organisms, suggesting that the protein family repertoires of the Reduced genomes are the product of historical contingency as well as the specific adaptive value they represent in the ecological niche occupied by each organism. Historical contingency was also observed to play an important part in theoretical studies of reductive genome evolution of metabolic pathways [44]. Interestingly, less than half of the protein families defined by essential genes in E. coli are kept in Reduced genomes, which clearly illustrates how different environments demand different sets of solutions, in this case protein families. Our results suggest that while protein family diversity is preserved in genome reduction, genetic redundancy is lost. Bacterial genomes are widely reported to have smaller protein families than eukaryotes [29]–[31], relying less on genetic redundancy as a means of robustness. In fact, recently Freilich and co-workers showed that enzymes in prokaryotes are less functionally redundant than in eukaryotes [58]. Our results however suggest that free living bacteria still rely on genetic redundancy as a source of robustness. Reduced genomes have twice the number of singletons as FL, i. e. twice the number of genes that do not have copy backup. This is a lower bound for an estimate of the decrease in genetic redundancy in reductive genome evolution. We are not considering, for example, partial or domain redundancy [10], additional redundancy that may also be sacrificed in reductive genome evolution. One such example is that most members of the order Enterobacteriales, which includes E. coli, have the chaperone DnaJ as well as two proteins that share specific domains with it, CbpA and DjlA. These have been shown to be functionally redundant with DnaJ [59]. The intracellular endosymbionts Buchnera aphidicola APS and Candidatus Blochmannia floridanus, members of the same order, still have DnaJ but lost CbpA and DjlA. It is important to note however that genetic redundancy is but one source of robustness. There is anecdotal evidence suggesting that reductive genome evolution may sacrifice other types of robustness that do not involve copy redundancy, complete or incomplete. For example, loss of network redundancy, i. e. alternative pathways in the synthesis of acetylCoA (two pathways in E. coli), was reported in the reductive evolution of B. aphidicola (one pathway) [44]. In another example, Cyanobacteria have an oscillator coded by three unrelated genes (KaiA, KaiB, KaiC), capable of maintaining cell cycle rhythms independently of external light-dark cycles. Members of the marine genus Prochlorococcus, although free living, have undergone extensive genome reduction [60], and have lost KaiA. As a consequence, the oscillator became less robust to external light cycles [61]. One promising avenue of research is then to understand to what extent other sources of robustness are affected in the reductive genome evolution. An abundant body of theoretical work predicts that variable, unpredictable environments select for, or promote the emergence of robustness (reviewed in [11]). Abundant anecdotal examples support this prediction. For example, Sanchez-Perez and co-workers [62] proposed recently that after duplication, paralogues may retain the initial function but specialize to work under different environmental conditions. These ‘ecoparalogues’ which could still effectively compensate for each other, i. e. are functionally redundant, would support a link between environmental unpredictability and the emergence of robustness. They were able to find examples of proteins that are predicted to perform the same function but have different isoelectric points, and hence are predicted to operate at different ranges of salinity. Thus protein specialization under varying environments may provide the drive for the emergence of genetic redundancy. We now invert this reasoning and show that the transition to a predictable environment removes that drive, resulting in the selective loss of genetic redundancy and hence, robustness. To the best of our knowledge our results provide the first systematic description of the loss of robustness by genetic redundancy in the evolution of cellular organisms. Redundancy is common in higher organisms that experience low mutation rates and small population sizes, and low in organisms that have high mutation rates and large population sizes [63]. Since commitment to an intracellular lifestyle is typically associated to a radical reduction in the effective population size [18]–[20] and high mutation rates [21], it would be reasonable to expect that there would be a concomitant increase in redundancy [63], [64]. This is however the opposite of what we observe – obligate parasites and endosymbionts that suffered a decrease in population size and increase in mutation rate experiencing a decrease in (genetic) redundancy. We thus provide empirical support to the notion that the predictability of the environment is of paramount importance in the evolution of redundancy. Supporting our conclusion is the observation that modularity, a characteristic of biological systems that has been linked to robustness [2], has also been shown to vary with environmental predictability, with more modular networks being found in more unpredictable environments [65]. Note that genome reduction may not be a pre-requisite for loss of genetic redundancy, as even organisms like the marine bacteria of the genus Pirellula, inhabiting a predictable environment, have a remarkably small number of paralogues, while retaining very large genomes [31] Finally, many of the Reduced organisms that we studied here are causative agents of human diseases such as Lyme disease, leprosy, typhus, tularemia, pneumonia, among others. The realization that they all share a lack of robustness due to the loss of redundancy suggests new avenues for the identification of drugable targets. Instead of aiming to identify genes or pathways that are specific to the pathogenic organism, we can aim to target fragile parasite pathways in the context of robust host functions. The complete list of species used in this study is given as supplementary material (Tables S3 and S4 in text S1). It consists of 308 free living bacteria and 69 reduced genomes. Reduced genomes include obligate intracellular parasites (34 organisms) and endosymbionts (15 organisms), obtained from [15], [16], [66]. It further included parasitic bacteria like Mycoplasma sp. , which while not being intracellular are obligate parasites displaying signs of extreme genome reduction [51] (20 organisms). Essential genes in E. coli were obtained from [67] and from the PEC database (www. shigen. nig. ac. jp/ecoli/pec/). Functional class assignments were obtained from the COGs database [68], [69]. Analysis involving COGS included only genomes with more than 50% COG coverage: 176 free living and 54 reduced genomes. Functional classification with E. C numbers was obtained from KEGG [57]. We used domain architecture as defined in the Superfamily database [32] to identify protein families. Two proteins are considered part of the same family if they display the same N- to C-terminal domain architecture, ignoring gaps as described in [70]. Domain assignments were based on Superfamily release 1. 69 [71]. Sequence similarity was computed using BLAST [72] at a cutoff of E≤0. 01 and orthologues were identified as reciprocal best hits [73]. Considering a power law function y = γxα, the elasticity of y in relation to x is a constant: (dy/dx) (x/y) = α. The elasticity can be estimated using the linearization ln (y) = β1+β2ln (x), where β1 = ln (γ) and β2 = α. The filtered average family size is computed as (F/N) • (n/N) 2, where n is the number of organisms where it appears and N the total number of organisms. We simulated gene loss scenarios the following way. We randomly picked one free-living genome from the set of 309 as the start point. Then we used a log-normal distribution approximated to the Reduced genome size distributions to randomly generate an end point of the simulation, i. e. the final size of the artificially reduced genome. We then randomly picked genes from the start genome to be “lost”, until we reached the final size. The probability of gene loss was adjusted in six alternative ways. In S1 it was totally random and represents also the background of all other scenarios. In S2 the probability of loss is made to depend linearly on the number of protein domains, i. e. a protein with two domains was twice as likely to be lost as a protein with a single domain. In S3 we consider essentiality a property of the family [42]. We made the probability of loss depend on the protein family distribution of known essential genes in E. coli. Protein families rich in essential genes (>50%) had a 2 fold decrease in the probability of loss, and protein families with less than 50% had just the random background probability of loss. In S4 we adjusted the probability of loss to the functional class distributions in the reduced genomes. Functional classes that are more frequent in Reduced genomes (Figure S5 in Text S1) had its probability of loss reduced to half (functional classes F, J, L, O and U), and those functional classes that are less frequent in reduced genomes had double the probability of loss (E, K, P, Q, R, S and T). In scenario S5 we used KEGG pathway assignment to predict pathway participation and considered that once a gene was lost, members of the same pathway were three times more likely to be lost afterwards. Finally, bacterial genomes are frequently organized in operons, which results in functionally related proteins being coded by genes in close proximity on the chromosome. We considered this in scenario S6 where once a gene is lost, the probability of its adjacent genes being lost afterwards increases twofold. We estimate the probability of losing proteins in a given family Ploss as the ratio between the total number of elements lost in the family over the size of that family in FL. Ploss (FFL) = (FFL−FReduced) /FFL. FFL and FReduced are the total number of elements of the family in each class of genomes; for this analysis we only considered families that appear in 90% or more organisms in both classes.

Bacteria have found many niches in which to live, and one of them is inside eukaryotic cells. These intracellular bacteria include endosymbionts like Buchnera aphidicola, which provides its host, an aphid, with essential amino acids, as well as many pathogenic bacteria such as Mycobacterium leprae and Rickettsia prowazekii, the causative agents of leprosy and typhus, respectively. Even though they all evolved their intracellular lifestyle independently, all these bacteria lost a large number of genes as they adapted to their hosts, presumably because the rich environment where they found themselves no longer required such functions. For example, biosynthetic genes are frequently lost. It has been a matter of debate what decides whether a gene can be lost in evolution, and intracellular bacteria have been used as model systems to study these processes. In our study, we propose that when adopting an intracellular lifestyle, these bacteria extensively lost duplicated genes. We propose that this represents loss of copy redundancy that is possible because the host cell represents a predictable environment in which there is little pressure for the bacteria to retain these backups. In simplistic terms, if the road is always smooth, you are probably OK without a spare tire.

Abstract Introduction Results Discussion Methods

evolutionary biology/microbial evolution and genomics computational biology/systems biology computational biology/genomics

2011

Loss of Genetic Redundancy in Reductive Genome Evolution

Homology modeling predicts the 3D structure of a query protein based on the sequence alignment with one or more template proteins of known structure. Its great importance for biological research is owed to its speed, simplicity, reliability and wide applicability, covering more than half of the residues in protein sequence space. Although multiple templates have been shown to generally increase model quality over single templates, the information from multiple templates has so far been combined using empirically motivated, heuristic approaches. We present here a rigorous statistical framework for multi-template homology modeling. First, we find that the query proteins’ atomic distance restraints can be accurately described by two-component Gaussian mixtures. This insight allowed us to apply the standard laws of probability theory to combine restraints from multiple templates. Second, we derive theoretically optimal weights to correct for the redundancy among related templates. Third, a heuristic template selection strategy is proposed. We improve the average GDT-ha model quality score by 11% over single template modeling and by 6. 5% over a conventional multi-template approach on a set of 1000 query proteins. Robustness with respect to wrong constraints is likewise improved. We have integrated our multi-template modeling approach with the popular MODELLER homology modeling software in our free HHpred server http: //toolkit. tuebingen. mpg. de/hhpred and also offer open source software for running MODELLER with the new restraints at https: //bitbucket. org/soedinglab/hh-suite. Homology modeling is by far the most widely used computational approach to predict the 3D structures of proteins, and almost all protein structure prediction servers rely chiefly on homology modeling, as seen in the community-wide blind benchmark “Critical Assessment of Techniques for Protein Structure Prediction” (CASP) [1–3]. Homology modeling consists of four steps: (1) Finding homologous template proteins of known structure, (2) Selecting the best template or set of templates, (3) Optimizing the multiple sequence alignment (MSA) between query and template protein sequences, and (4) Building the homology model for the query sequence that resembles as closely as possible the structures of the templates, accommodating for deletions and insertions of query residues with respect to the template structures. During the last 15 years, much progress has been made regarding the sequence-based steps 1 to 3. This is mainly owed to the development of more sensitive and accurate methods for sequence searching and alignment that compare sequence profiles or profile hidden Markov models (Hmms) with each other [4–6]. In contrast, improvements to the last step have been marginal. This is illustrated by the fact that, although a number of tools for protein homology modeling exist, to our knowledge all are older than 12 years (see [7,8] for reviews). ModSeg/ENCAD [9] copies template coordinates and bridges gaps by short fragments that match the framework of the target structure. SWISS-MODEL [10] generates a core model by averaging template backbone atom positions. NEST [11] implements an artificial evolution algorithm where changes from the template structure such as substitutions, insertions and deletions are made one at a time, and each mutation is followed by an energy minimization. This process is repeated until the whole query is modeled. These tools rely on of various heuristics. MODELLER [12], with 7500 citations clearly the most popular and according to two studies [7,8] also the most successful homology modeling software to date, stands out by being based on a statistical approach to homology modeling. MODELLER is essentially unchanged at its core since its publication 22 years ago, while extensions such as refined energy functions [13] or loop modeling [14] have led to relatively minor improvements of its already excellent performance. We therefore believe Modeller’s success is owed to the consistent, statistical approach at its core. Modeller proceeds in two steps: (1) Derive from the MSA and template structures a list of restraints and (2) find the model structure that minimizes the restraint violations. Each restraint is a probability density function. The most important class of template-dependent restraints are the probability density functions for the spatial distances of pairs of atoms in the query protein. The true distance d will be distributed around the distance dt of the equivalent atoms in the template structure, where equivalent residues are those that are aligned to each other (Fig 1). MODELLER assumes for simplicity a Gaussian distribution for d. Its mean equals dt and its standard deviation is predicted based on the sequence similarity between query and template. The restraint minimization in the second step amounts to a maximum likelihood optimization, where the likelihood is approximated as the product over the density functions of the individual restraints. This factorisation of the likelihood assumes that the individual restraints represent information independent of each other, because in probability theory the joint probability of two random variables (X and Y) is the product of their probabilities, p (X, Y) = p (X) p (Y), if and only if they are independent of each other. Although the assumption of independence of restraints sounds rather drastic, the approximation turned out to work well in practice. To aggregate the information from several templates, however, MODELLER does not multiply the density functions of all restraints as probability theory would suggest. Instead, it relies on an empirical observation that the distribution of the target distance informed by multiple template distances is multi-modal. Thus, MODELLER reverts to a heuristic approach and computes an additive mixture of the density functions, each derived from an individual template, to restrain a single target distance based on multiple templates. Here, we develop a rigorous statistical treatment of multiple template homology modeling. We first show that the distance distributions for log (d) are very well described by two-component Gaussian mixture distributions. In contrast to MODELLER’s one-component densities, these two-component densities allow us to combine density functions by multiplication. Second, we derive an algorithm to compute weights that take the statistical dependence of the distance information from the templates into account. Third, we propose a heuristic scheme for template selection. We demonstrate that the new HHpred modeling pipeline and in particular the new constraints yield substantially improved model qualities. Our approach to multi-template homology modeling is based on the statistical approach to homology modeling introduced by Modeller. Our software computes improved spatial restraints and calls the Modeller software, which then reads in the restraints and finds a structure that optimally satisfies these restraints. We briefly recall Modeller’s approach of homology modeling here. We filtered the sequences from the PDB database of protein structures (May 2010) down to 20% and 70% maximum pairwise sequence identity and a minimal pairwise E-Value of 0. 1 (using scripts pdb2fasta. pl and pdbfilter. pl in the HHsuite package v2. 0. 16). For all sequences in the resulting pdb20 and pdb70 databases, we built multiple sequence alignments (MSAs) with our sensitive, iterative sequence search tool HHblits (v2. 0. 16) that is based on the pairwise alignment of profile hidden Markov models (Hmms) [15]. We used standard HHblits parameters with three search iterations against the uniprot20 database to get sufficiently diverse MSAs that are well suited to detect even remotely homologous proteins. The query sequences were picked from among the pdb20, and the template database was obtained from the pdb70 as explained below. We extracted three disjoint query sets from the pdb20, a test, a training and an optimization set, with 1000,1000, and 500 proteins, respectively. To achieve a good balance of easier and more challenging queries for modeling, we aimed to obtain the same distribution of query-template sequence identities as for the 108 queries in the CASP7 experiment shown in Supplemental Fig. S2 (which is similar to the distribution in CASP11, see Fig. S2). We computed the total amount of queries needed in each sequence identity bin (0%–5%, 5%–10%, …, 95%–100%). We then randomly picked query sequences from the pdb20 without replacement. For each picked query, we searched for possible templates in the pdb70 database and found the template most structurally similar to q according to TMalign (excluding the query itself) and recorded the sequence identity given by TMalign. q was then put into one of the three sets if the sequence identity bin for that set was not yet filled up. Otherwise, q was rejected. Finally, for each of the three query sets we constructed a template set by removing the sequences in the query set from the pdb70. We then searched with each query sequence q in one of the three sets through the corresponding template database using HHsearch, a slower and slightly more sensitive version of HHblits, resulting in a list tlist (q) of potential templates. Most model quality assessment scores, such as the GDT-ha, do not penalize incorrect regions and thus reward adding more templates to increase the fraction of the query structure for which restraints can be derived. [30] assessed the effect of using a single or multiple templates on model quality and concluded that most of the gains are due to increased coverage of query residues by template residues. We wanted to discriminate between improvements in model quality due simply to increased coverage and improvements owed to reducing statistical noise by increasing the number of distance restraints on “core residues”, conveniently defined here as residues covered by the alignment to the first, top-ranked template. We remove all non-core residues in the input alignment to Modeller. In that way, distance constraints can only be generated on cores. Then we evaluate the resulting models on core residues only and we compare the GDT-has with the general case. Table 2 shows that, first, using multiple templates leads to a clear improvement over single templates both in the core regions and overall. This shows that the effect of adding further templates to the first selected template does indeed improve model quality to a similar extent in the core and non-core regions. Similarly, the improvements due to our new two-component restraints are of the same order in the core regions (+2. 0%) as overall (+2. 5%), leading to a similar conclusion, that the new restraints improve the model to the same extent in the core and non-core regions. Our probabilistic multi-template modeling approach should have the advantage over the Modeller restraints of being more robust towards wrong restraints, because the new distance restraints become flat when log d deviates strongly from log dt, i. e. , when the restraint cannot be satisfied at all. Therefore, completely wrong restraints practically get ignored in the new approach. Note that this was not a design target of our method but it is simply a consequence of a correct statistical treatment. To test our hypothesis on the robustness of the new restraints, we modified the template selection as follows. For each query in the test set, we constructed three different template sets (Table 3). The three sets contained two good templates each, and 0,1 or 2 bad templates, respectively. The good templates were the top two templates according to the TMscores predicted by the neural network in Fig. S1 that also attained a true TMscore of > 0. 5. The bad templates were the lowest ranked templates with a true TMscore < 0. 3. The average model quality obtained with these three selections are shown in Table 3. As expected, the models built with the new restraints proved to be considerably more robust than the models built with the standard Modeller pipeline. CASP (Critical Assessment of Structure Prediction) is a community wide, double-blind experiment that takes place every second year to objectively test the performance of various predictors. HHpred regularly participates in the server based structure prediction category competing with 70–80 other servers. For CASP9 and CASP10, we integrated all methods described above into the HHpred pipeline. Depending on whether there existed a suitable template in the databases, all queries are subdivided into two categories: template based (TBM) and free (FM). Due to the ever increasing database sizes, most of the queries are TBM (121 vs. 26 in CASP9 and 111 vs. 15 in CASP10). As Fig 8 shows, for TBM HHpred is among the most accurate servers (top 1 in CASP9 and top 7 in CASP10 according to the official CASP ranking—all three servers differ only in minor technical details, see [31,32]). At the same time HHpred is faster by a factor of ∼ 350 compared with the other leading groups. Fig 8 summarizes the official results in the TBM category from two community-wide assessments of techniques for protein structure prediction, CASP9 (121 query proteins) and CASP10 (111 query proteins) [1,3]. The values used in the figure were downloaded from the official CASP website (http: //predictioncenter. org/). For detailed results, see Supporting Information. When replacing the new restraints with Modeller’s default restraints for the CASP10 set on the same selection of templates, the gdtts-score decreased by 3%. When considering HHpred’s performance in CASP9 and CASP10, note that assessors filtered out targets that will be too simple to predict by eliminating targets for which a high-confidence homologous template could be found using HHsearch. This procedure thus selectively biases the targets at the detriment of HHpred by eliminating targets that would be easy for HHpred to predict. Protein structure prediction is a mature field, in which the best methods differ only by a few percent in performance according to recent CASP benchmarks. Even so, great progress has been made in the last 10 to 15 years in template-based protein structure prediction, fuelled by advances in techniques for remote homology detection and alignment [6] and techniques for model quality assessment [3]. In contrast, most successful servers in CASP employ Modeller to build their 3D homology models, a software whose core has changed very little since its publication 22 years ago. This speaks to the enormous success of Modeller’s statistical approach to homology modeling. In this study we have shown how to generalize the statistical approach by taking account of alignment errors and treating restraints from multiple templates in a probabilistically satisfactory way. These theoretical insights have led to improvements in average model quality (around 6. 5%) that are somewhat smaller than what we expected initially. In hindsight, Modeller’s heuristic to derive multi-template restraints works surprisingly well. Also, since Modeller’s internal workings (e. g. the stochastic optimization) are optimized together with its own restraints, it might well be possible to improve on the presented results by specifically optimizing Modeller’s model building procedure with our new restraints. We note, however, that an average model score improvement of 4. 4% (m. ss. old versus m. mt. new in GDT-ts, see Table A in S1 Text) corresponds to the difference in GDT-ts scores between the 3rd best and 14th best server in CASP10 [5]. This is a considerable success in particular because our theoretical approach is quite general and can be transferred to other homology modelling methods and to the up-and-coming field of modeling large protein complexes from heterogeneous experimental data [33]. We noted during our tests that the positive impact of the new restraints on model quality is strongest when evaluated with the strictest score, GDT-ha, as compared to the less strict GDT-ts or TMscore (Table A in S1 Text). Here, strictness refers to how severely already small deviations of the model from the true structure are penalized. This observation shows that the improvements of our new restraints are to a substantial degree in the high-precision regime, i. e. , below 1 Å, by further improving regions of the model that are already fairly well modeled. Since the best-modeled regions are expected to largely coincide with the highly conserved and hence functionally most important parts of the protein, we expect the new restraints to have the strongest impact on the functionally most important regions of the model. We are convinced of the power of probability theory in describing quantitative phenomena under uncertainty. Modeller is an excellent case in point. An interesting idea is to carry the probabilistic view further by probabilistically integrating structural and sequence information. All approaches so far start from a fixed query-template alignment (or from a set of alternative alignments) and try to find the 3D model that is best compatible with the alignment. To allow information from the 3D modelling to be fed back to the alignment stage and vice versa, it seems promising to explore the joint posterior probability distribution of alignment and 3D structure. One way to do this would be by Markov Chain Monte Carlo Gibbs sampling of the alignment and the model structure from appropriate conditional distributions.

Since a protein’s function is largely determined by its structure, predicting a protein’s structure from its amino acid sequence can be very useful to understand its molecular functions and its role in biological pathways. By far the most widely used computational approach for protein structure prediction relies on detecting a homologous relationship with a protein of known structure and using this protein as a template to model the structure of the query protein on it. The basic concepts of this homology modelling approach have not changed during the last 20 years. In this study we extend the probabilistic formulation of homology modelling to the consistent treatment of multiple templates. Our new theoretical approach allowed us to improve the quality of homology models by 11% over a baseline single-template approach and by 6. 5% over a multi-template approach.

Abstract Introduction Materials and Methods Results Discussion

2015

Automatic Prediction of Protein 3D Structures by Probabilistic Multi-template Homology Modeling

Antigen B (AgB) is the major protein secreted by the Echinococcus granulosus metacestode and is involved in key host-parasite interactions during infection. The full comprehension of AgB functions depends on the elucidation of several structural aspects that remain unknown, such as its subunit composition and oligomeric states. The subunit composition of E. granulosus AgB oligomers from individual bovine and human cysts was assessed by mass spectrometry associated with electrophoretic analysis. AgB8/1, AgB8/2, AgB8/3 and AgB8/4 subunits were identified in all samples analyzed, and an AgB8/2 variant (AgB8/2v8) was found in one bovine sample. The exponentially modified protein abundance index (emPAI) was used to estimate the relative abundance of the AgB subunits, revealing that AgB8/1 subunit was relatively overrepresented in all samples. The abundance of AgB8/3 subunit varied between bovine and human cysts. The oligomeric states formed by E. granulosus AgB and recombinant subunits available, rAgB8/1, rAgB8/2 and rAgB8/3, were characterized by native PAGE, light scattering and microscopy. Recombinant subunits showed markedly distinct oligomerization behaviors, forming oligomers with a maximum size relation of rAgB8/3>rAgB8/2>rAgB8/1. Moreover, the oligomeric states formed by rAgB8/3 subunit were more similar to those observed for AgB purified from hydatid fluid. Pressure-induced dissociation experiments demonstrated that the molecular assemblies formed by the more aggregative subunits, rAgB8/2 and rAgB8/3, also display higher structural stability. For the first time, AgB subunit composition was analyzed in samples from single hydatid cysts, revealing qualitative and quantitative differences between samples. We showed that AgB oligomers are formed by different subunits, which have distinct abundances and oligomerization properties. Overall, our findings have significantly contributed to increase the current knowledge on AgB expression and structure, highlighting issues that may help to understand the parasite adaptive response during chronic infection. Echinococcus granulosus is the causative agent of cystic hydatid disease (CHD), a worldwide zoonotic infection that affects humans and livestock [1]. Antigen B (AgB) is the major protein secreted by the E. granulosus pathogenic larval stage (metacestode or hydatid cyst). Since its first description in 1971 [2], AgB has been the most studied E. granulosus protein due to its role in parasite biology and its potential for application in CHD control tools [3], [4]. AgB has been described as involved in several host-parasite interaction mechanisms that promote parasite establishment and survival in the intermediate host, such as protease inhibition [5], lipid binding [6] and immunomodulation [7], [8]. Furthermore, AgB is highly immunogenic in human infections, presenting a high diagnostic value for CHD [9], [10]. AgB is homologous to hydrophobic ligand binding proteins (HLBPs), a family of cestode helix-rich proteins that bind hydrophobic compounds [6]. It is an oligomeric lipoprotein composed of 8-kDa related subunits (AgB8 subunits) [2], [11], which are encoded by a multigene family that includes at least five members (AgB1-5) [12]–[16]. In SDS-PAGE, under reducing conditions, AgB dissociates into low-order oligomers of 8-kDa subunits (monomers, dimers, trimers, etc.) [11], [17]. In solution, AgB exists as high-order oligomers [2], [18], showing distinct populations of ∼160, ∼240 and >320 kDa. Despite being an extensively studied protein, several important aspects of the AgB molecular structure remain largely uncharacterized [19], such as its subunit composition and oligomeric states. It is not known which AgB subunits are expressed and secreted in the hydatid fluid of an individual cyst, as studies carried out so far analyzed AgB from a pool of cysts [17], [20]. The abundance of each 8-kDa subunit within a cyst, its oligomerization properties, and its contribution to define the distinct AgB oligomeric states are also still open questions. AgB subunits could present differential expression patterns within individuals [21] and/or throughout the parasite' s development [22], [23]. Furthermore, different 8-kDa subunits have distinct physical-chemical [18], immunological [9] and lipid-binding [6] properties. So, AgB subunit composition and abundance may determine distinct protein oligomeric states, biochemical interactions and biological roles. Therefore, these AgB structural aspects are expected to have repercussions on adaptive host-parasite relationships and on the outcome of AgB-based CHD immunodiagnostic methods [19], [21]. Thus, there is a clear need for further studies to elucidate which subunits are secreted in the hydatid fluid of a cyst and to characterize the oligomeric states formed by these subunits. This information could aid in a better understanding of the roles played by AgB during the host-parasite interaction and improve its application in advancing immunodiagnostic and therapeutic approaches for CHD. Recently, we demonstrated the self-assembly of three AgB recombinant subunits (rAgB8/1, rAgB8/2 and rAgB8/3) into homo-oligomers that have similar properties to those of parasite-produced AgB, validating them as tools for studying AgB structure [18]. In the present work, we investigated the subunit composition and oligomeric states of E. granulosus AgB. The subunit composition of AgB purified from individual bovine and human hydatid cysts was studied by mass spectrometry associated with electrophoretic analysis. The exponentially modified protein abundance index (emPAI) was employed to obtain information on the relative abundance of the 8-kDa subunits within the different AgB samples. Using the available AgB recombinant subunits, we assessed the in vitro oligomerization properties of these different 8-kDa subunits and performed a comparative structural characterization of the recombinant oligomers and AgB purified from hydatid cyst. E. granulosus bovine hydatid cysts were obtained from lungs of naturally infected animals slaughtered at Frigorífico Cooperleo, São Leopoldo, RS, Brazil. Animal slaughtering was conducted according to Brazilian laws and under supervision of the Serviço de Inspeção Federal (Brazilian Sanitary Authority) of the Brazilian Ministério da Agricultura, Pecuária e Abastecimento. Condemned viscera at post-mortem meat inspection due to the presence of hydatid cysts were collected at the abattoir and dissected in the laboratory, following protocols approved by the Ethical Committee of the Universidade Federal do Rio Grande do Sul. For AgB purification from single cysts, hydatid fluid samples from two fertile cysts (collected from different animals) were aseptically aspirated and individually processed for immunopurification. Hydatid fluid samples were processed according to Oriol et al. [2] and AgB immunopurification was performed as described previously [18]. Cysts used in this work were required to have a minimum volume of ∼200 ml hydatid fluid, which was necessary to obtain enough AgB for mass spectrometry and structural analysis. Maximum yields of 1 mg AgB were typically obtained from each bovine cyst. Human hydatid fluid was collected from a CHD patient after surgical aspiration of a lung cyst content performed at Instituto Hugolino Andrade, Santana do Livramento, RS, Brazil. The cyst was fertile and classified as type CE1 according to the standardized WHO classification [24]. The patient did not undergo any chemotherapy prior to surgery, and signed an informed consent for both the surgical procedure and the use of the aspired hydatid fluid for research purposes. The use of this biological sample was approved by the Ethical Committee of the Universidade Federal do Rio Grande do Sul. The hydatid fluid was clarified by centrifugation at 10000× g for 15 min at 4°C and concentrated 10-fold by vacuum centrifugation prior to electrophoretic analyses. AgB was not purified and the analyses were performed with raw human hydatid fluid, as it showed large quantities of AgB and minimal contamination with host proteins. Bovine and human cysts used in this work were genotyped as E. granulosus sensu stricto G1 (sheep strain) (for details, see Text S1). AgB recombinant subunits rAgB8/1, rAgB8/2 and rAgB8/3 were expressed in Escherichia coli as glutathione S-transferase fusion proteins, purified by affinity chromatography and recovered using thrombin cleavage as described previously [25]. Protein concentrations were determined using a Qubit quantitation fluorometer and Quant-it reagents (Invitrogen, Carlsbad, USA). For SDS-PAGE analyses, AgB samples (40 µg) were resolved on 15% gels, under reducing conditions, as described by Laemmli [26]. Native PAGE was performed in 4–20% Mini-PROTEAN TGX gels (Bio-Rad, Hercules, USA) using Tris-glycine (pH 8. 3) as running buffer. Protein samples (10 µg) were diluted with an equal volume of 2× native sample buffer (62. 5 mM Tris-HCl, pH 6. 8,40% glycerol, 0. 01% bromophenol blue) and run at a constant voltage of 100 V. Both denaturing and native gels were stained with Coomassie blue and scanned with an ImageScanner (GE Healthcare, Chalfont St. Giles, UK). LS measurements of AgB samples purified from bovine cysts were performed on a DynaPro instrument (Protein Solutions Inc. , Charlottesville, USA), as it allows for the use of smaller sample volumes, compatible with the low yields obtained from E. granulosus AgB purification. AgB was analyzed in PBS (phosphate buffered saline, pH 7. 4) at a concentration of 100 µg/ml, and DynaPro V. 5 software was used for data analysis. AgB recombinant subunits were analyzed on a Brookhaven Instruments standard setup (BI-200 M goniometer, BI-9000 AT digital correlator) with a He-Ne laser (λ = 632. 8 nm) as light source. Proteins samples were diluted in PBS to a final concentration of 1 mg/ml. After centrifugation (14,000 rpm, 10 min, 4°C), the supernatants were filtered through 0. 22 µm pore diameter membranes (Millipore, Milford, MA, USA) into dust-free cylindrical cuvettes in a laminar flow box. Measurements of both dynamic and static light scattering (DLS and SLS, respectively) were conducted in triplicate at a scattering angle of 90°. A water bath was used to control the temperature and LS was measured at 23°C and 37°C. The samples were equilibrated at each temperature for 10 min before the LS measurements. DLS and SLS data analysis was performed as described in Text S1. The recombinant AgB oligomers (0. 2 mg/ml in 25 mM Tris, pH 7. 5) were subjected to high hydrostatic pressure treatment at 25°C in either the absence or presence of 2-mercaptoethanol. The high pressure cell, equipped with optical windows, has been previously described [30]; it was purchased from ISS (Champaign, IL). The average size distribution of the proteins in solution was evaluated by exciting the samples at 320 nm and collecting the scattered light at 90° from 315 to 325 nm in an ISS K2 spectrofluorometer. For a compression/decompression cycle, the pressure was increased from 1 to 3000 bar in steps of 270 bar and then decreased in similar steps. At each step, the samples were allowed to equilibrate for 15–20 min before LS data collection. The secondary structure content of proteins before and immediately after the compression was monitored by circular dichroism (CD). The qualitative and quantitative subunit composition of AgB from individual bovine and human hydatid cysts was analyzed and compared using mass spectrometry. Total AgB subunit composition was analyzed by in-solution digestion of samples, while the specific composition of AgB low- and high-order oligomers was investigated by in-gel digestion of samples resolved in SDS-PAGE and native PAGE, respectively. Peptides corresponding to the AgB8/1, AgB8/2, AgB8/3, and AgB8/4 subunits were identified in all samples analyzed (Figure 1A and Table S1), while AgB8/5 subunit was not detected in any of them. Qualitative differences were observed between samples, with the identification of a variant for AgB8/2 subunit in one of the bovine samples (cyst 1). The detected AgB8/2 isoforms differ at their stop codon position, being the identification of the AgB8/2v8 variant based on the detection of the tryptic C-terminal peptide NLVEEKDDDS (Figure 1B). Mass spectrometry data from native PAGE revealed that the two AgB8/2 isoforms were expressed in the bovine cyst 1. To investigate the relative abundance of different subunits within AgB samples we used the emPAI (Figure 1A and Table S1), which revealed a consistent and significant relative overrepresentation of AgB8/1 subunit in all analyzed samples. AgB8/2 and AgB8/4 subunits were also well represented in bovine and human samples, especially in high-order AgB oligomers resolved by native PAGE. On the other hand, AgB8/3 subunit relative abundance varied between AgB samples from different hosts. In bovine samples, AgB8/3 was the less abundant subunit, whereas in the human sample it was found in relative high abundance based on in-solution digestion data. AgB was found in different oligomeric states, as detected by both native PAGE and light scattering. In native PAGE, AgB showed oligomers of different sizes, which appeared as a broad smear, with a more defined band of ∼550 kDa (Figure 2A). In DLS, AgB also showed different oligomeric states, which are represented by populations with hydrodynamic radii (Rh) of ∼4 nm, ∼100–200 nm and >2 µm. AgB samples heated at 37°C were also analyzed to evaluate the protein oligomeric states at physiological conditions, but no effect of temperature in AgB oligomerization was observed by native PAGE or DLS (data not shown). In addition to AgB identification, the MS analysis of native PAGE bands from the human hydatid fluid revealed that the ∼550 kDa band also contained peptides corresponding to antigen 5 (Ag5), another E. granulosus secreted protein. Moreover, the ∼660 kDa band observed in this sample was also identified by MS as Ag5. These results (data not shown) are suggestive that Ag5 is also able to form high-order oligomers, as AgB. In order to assess the oligomerization behavior of different AgB subunits, the homo-oligomers formed by the available recombinant subunits, rAgB8/1, rAgB8/2 and rAgB8/3, were also analyzed by native PAGE and light scattering. Both techniques detected a differential oligomerization behavior for AgB recombinant subunits, which formed oligomers with a maximum size relation of rAgB8/3>rAgB8/2>rAgB8/1. In native PAGE, the recombinant oligomers exhibited different size ranges, with rAgB8/3 forming the most heterogeneous oligomeric states, distributed over a wide range (∼100–550 kDa) (Figure 2B). This subunit also showed the most significant changes at 37°C, clearly exhibiting a higher oligomerization in response to temperature increasing. In addition, rAgB8/3 was the only subunit able to form higher-order oligomers (∼550 kDa) as those observed for E. granulosus AgB. The oligomeric states detected for rAgB8/3 subunit by DLS at physiological conditions, with calculated Rh of ∼4 nm, ∼100–200 nm and >2 µm, were also more similar to those observed for E. granulosus AgB by the same technique. The oligomers of different sizes formed by each recombinant subunit at 37°C can be observed from relaxation time distributions in Figure 3. The relaxation time (τ) is inversely proportional to the diffusion coefficient of the scattering molecules, and hence directly proportional to their size. Thus, a fast relaxation time (in µs) was related to the motion of smaller particles and a slow relaxation time was related to that of larger particles. At 23°C, all recombinant subunits appeared as oligomers of fast relaxation times, with calculated Rh of ∼4 nm. Upon temperature increase, rAgB8/2 and rAgB8/3 formed other oligomeric states, detected by slower relaxation modes. As also observed in native PAGE, rAgB8/3 was the only subunit able to form higher-order oligomers, which are represented by the slow relaxation mode of ∼6 µs. The different oligomerization properties of AgB subunits in solution was also confirmed by SLS data (data not shown), with rAgB8/3 subunit being more prone to oligomerization. We do not know whether the oligomeric states detected by DLS exactly correspond to those observed in native PAGE, but the differential oligomerization behavior of AgB recombinant subunits was detected by both techniques. Image techniques (TEM and AFM) were used in order to characterize the ultrastructure of AgB oligomers and its recombinant counterparts formed at 37°C. The differential oligomerization behavior of AgB recombinant subunits were also detected by microscopy experiments (Figures 4 and 5), with a rAgB8/3>rAgB8/2>rAgB8/1 oligomer size relation. In the three-dimensional AFM images, the AgB oligomers display a near-globular shape and showed heterogeneity in size and morphology both between and within samples (Figure 5A–D). Among recombinant subunits, the oligomeric states formed by rAgB8/3 subunit were more similar, both in size and morphology, to those observed for E. granulosus AgB (Figure 5C–F). To compare the structural stabilities of AgB recombinant oligomers, proteins were subjected to treatment with high hydrostatic pressures, during which their oligomeric states were monitored by LS changes (Figure 6A). The LS signal of rAgB8/1 oligomers decreased by ∼60% after compression, which indicates its partial dissociation into smaller species. The rAgB8/2 and rAgB8/3 oligomers dissociated more poorly (LS decreased only ∼30%), even at the highest pressure attained in our experimental setup. The CD spectra recorded before and immediately after the compression of the samples (Figure 6B) indicated that the secondary structure for rAgB8/2 and rAgB8/3 was less perturbed by the pressure treatment than for rAgB8/1. Thus, pressure-induced dissociation experiments showed marked differences in the stabilities of oligomers formed by AgB subunits, with those formed by rAgB8/2 and rAgB8/3 subunits presenting greater structural stability. However, despite of the differences in the stability of recombinant AgB oligomers, the dissociation appears to be irreversible for all of them, since the LS values were maintained after pressure removal (Figure 6A). As cysteine residues could contribute to the higher stability of the oligomers formed by rAgB8/2 and rAgB8/3 subunits, we also evaluated the effect of the reducing agent 2-mercaptoethanol on the stability of the AgB recombinant oligomers (Figure 7). The reducing agent had no effect on rAgB8/1 oligomer dissociation (Figure 7A), while the rAgB8/2 and rAgB8/3 oligomers exhibited greater dissociation in the presence of the reducing agent (Figure 7B and C, respectively). Even in the presence of 2-mercaptoethanol, the rAgB8/2 and rAgB8/3 oligomers were somewhat more resistant to pressure-induced disassembly than those formed by the rAgB8/1 subunit. Nevertheless, the results indicate that the differential stability of recombinant oligomers primarily results from disulfide bond formation. Few studies on Echinococcus AgB have addressed expression of the different AgB subunits in the hydatid cyst [22], [23] or identified the 8-kDa subunits involved in the formation of the AgB oligomers secreted in the hydatid fluid [17], [20]. Moreover, these previous studies analyzed a mixture of hydatid fluids from different cysts. In this work, we provided the first view of subunit composition for AgB samples purified from individual E. granulosus hydatid cysts from different hosts. Different experimental strategies, such as mass spectrometry and PAGE, allowed us to identify the subunits present in distinct oligomeric states of AgB. We also provided information on the relative abundance of AgB subunits using emPAI, a index that has been successfully employed to estimate relative abundance and stoichiometry of subunits on multiprotein complexes [31], [32]. AgB8/1, AgB8/2, AgB8/3 and AgB8/4 subunits were identified in all samples analyzed, with some quantitative and qualitative variations in AgB subunit composition within and between samples. The expression of different AgB subunits and their genomic variants may reflect an adaptive response of the parasite to different hosts and physiological conditions, diversifying AgB biochemical functions and antigenicity to promote its survival and evade host immune response [21], [33]. The AgB genes are highly polymorphic and several members are reported for each AgB subfamily [15], [21], [33]–[35]. This variation was detected here at protein level by the identification of an AgB8/2 variant, AgB8/2v8, which was previously genetically characterized in E. granulosus G1 strain [34]. The AgB2 subfamily was previously reported as subjected to selective pressure, which might be result of the direct interaction between AgB and host immune system [21], [34]. This can be addressed in future studies of the effect of different AgB8/2 epitopes in the modulation of host responses. Our results on AgB subunit composition are in agreement with the transcriptional data available for AgB genes, which show that AgB1, AgB2, AgB3 and AgB4 genes are expressed in Echinococcus metacestodes [22], [23]. We were not able to identify the AgB8/5 subunit in any of the samples studied, which suggests its absence or underrepresentation in the hydatid fluid. In agreement with our findings, AgB5 transcripts were detected at very low levels in the larval stage of E. granulosus and E. multilocularis, increasing its expression in the adult worm [22], [23], which suggests adult-specific functions for AgB8/5. An assumed AgB8/5 subunit was recently reported in a proteomic analysis of E. granulosus hydatid fluid [20], but the identified sequence, although initially named AgB8/5 [21], was latter assigned to the AgB3 subfamily [23], [36] and, therefore, does not correspond to the AgB8/5 subunit described by Mamuti et al. [16]. The detected AgB subunit relative abundance is also in agreement with the transcriptional data, which show that AgB1, AgB2 and AgB4 are transcribed at higher levels by the E. granulosus larval stage, whereas are poorly transcribed in protoscolex and other parasite developmental stages [23]. Therefore, both protein and RNA data suggest specific functions for AgB8/1, AgB8/2 and AgB8/4 subunits in AgB host-parasite relationships during metacestode chronic infection. Although it is not known which parasite tissue contributes more to the synthesis of AgB secreted in the hydatid fluid, our results show that the repertoire of secreted subunits resemble the expression profile of the germinal layer, suggesting that AgB secreted in the hydatid fluid is mostly produced by germinal layer cells. AgB3 gene, on the other hand, is expressed in all parasite stages and its level increases during development towards the adult worm [22], [23], which might indicate a more specific role for AgB8/3 subunit in adult worm differentiation. The observed discrepancies in AgB8/3 abundance between bovine and human cysts suggest that AgB genes, besides being developmentally regulated, may also vary their expression between parasite cysts and/or hosts. Comparing protein and RNA data from different cysts and hosts, one can note that the transcriptional profile of AgB genes described by Zhang et al. [23] for the germinal layer of a sheep cyst is more similar to the subunit abundance profile detected here for the human cyst than those observed for bovine cysts. Altogether, the issues raised here point to the need of comparative studies using a larger number of single-cyst AgB samples from parasites of different hosts, strains and physiological states (fertile and infertile) to produce a clearer and more comprehensive view on the AgB subunit expression at RNA and protein levels. In this sense, the present work represents a first step in this direction, providing and standardizing methodologies for preparation and analysis of AgB samples from single E. granulosus cysts. AgB subunit abundance can be correlated to some immunological features of 8-kDa subunits. The high antigenicity reported for AgB8/1 in Echinococcus infections [10], [37]–[39] could be related to its high expression in the parasite metacestode, as detected here at the protein level, and previously shown in transcriptional studies [22], [23]. The AgB8/2 subunit has been also described with a high diagnostic performance for CHD [9], [40] and was found here well represented in bovine and human cysts. The AgB8/4 subunit was also detected as relatively abundant in our cyst samples, but it has not been object of immunological characterization so far. In addition, AgB8/2 and AgB8/4 present 70% homology in their amino acids sequences [15] and are likely to share some antigenic epitopes, reinforcing the need to test AgB8/4 immunodiagnostic potential. On the other hand, AgB8/3 subunit showed lower antigenicity in preliminary tests [our unpublished results] and discrepant abundances between cyst samples. Further studies are needed to test possible implications of structural aspects of AgB oligomer architecture on subunit antigenicity, such as epitope masking or limited processing and presentation to T cells [18], [19]. In addition, immunoassays using combinations of AgB subunits, or mixtures of synthetic peptides containing major epitopes from different subunits, may improve the performance of serodiagnosis for echinococcosis [10], [41]. Since its first description by Oriol et al. [2], the oligomeric states of AgB have received little attention. Here, using AgB purified from hydatid fluid and recombinant 8-kDa subunits we characterized the E. granulosus AgB oligomers and demonstrated that they are not artifacts of the protein purification process and can be reproduced in vitro, under physiological conditions, using recombinant subunits. The results obtained in the present work for AgB structural analysis, besides confirming some previous findings [2], [18], have extended the structural characterization of recombinant oligomers and generated a useful comparison to those formed by E. granulosus AgB. The structural analysis of recombinant subunits also provided some insights into AgB oligomer formation and stability. Subunits found to be involved in the formation of AgB high-order oligomers were detected as having different oligomerization properties. Thus, considering subunit hetero-oligomerization, we could hypothesize a scenario where the more aggregative subunits act as nucleation centers for the formation of higher-order oligomers, to which less aggregative subunits could be attached. We also cannot exclude the possibility that post-translational modifications, absent in the recombinant proteins, could alter the oligomerization properties detected here for each 8-kDa subunit. AgB post-translational modifications, as previously described lipidation [2], may also be involved in the covalent association between subunits and in the oligomeric structure maintenance, since recombinant oligomers are non-covalently associated and completely dissociate under SDS-PAGE reducing conditions. Our structural analysis also points that AgB8/2 and AgB8/3 subunits could contribute to a higher stability of E. granulosus AgB oligomers, as they showed greater structural stability, which primarily results from disulfide bond formation. Other issue that remains elusive and deserves future investigation is the site of AgB subunit oligomerization. Secretory proteins are synthesized in the endoplasmic reticulum and oligomerization typically occurs within this cellular compartment, although, in some cases, oligomerization takes place in the intermediate compartment and Golgi apparatus [42]. Large protein oligomers, as those observed for E. granulosus AgB, were reported to be secreted by both lower and higher eukaryotes [43]–[45], with protein oligomerization occurring within the cisternae of Golgi apparatus. However, we do not know so far whether AgB higher-order oligomers can be formed within the secretory pathway of the germinal cells or they are formed only after secretion, in the hydatid fluid. Overall, this work provided important information on the AgB subunit composition, abundance and oligomerization, significantly increasing the current knowledge on AgB protein expression and structure. Using recombinant subunits we were able to structurally characterize the oligomers formed by different AgB subunits and to compare them to those formed by E. granulosus AgB. Although further studies will be required to completely elucidate AgB structure, our results will help understanding AgB roles in the host-parasite interplay during Echinococcus infection [3], [46], [47].

Antigen B (AgB) is the major secretory protein of the Echinococcus granulosus hydatid cyst, the causative agent of cystic hydatid disease. Structurally, AgB is a multisubunit protein formed by 8-kDa subunits, but it is not known which subunits are secreted by a single parasite (cyst) and how they interact in the formation of distinct AgB oligomeric states. Here, we investigated AgB subunit composition and oligomeric states in individual samples from bovine and human cysts. We identified AgB8/1, AgB8/2, AgB8/3 and AgB8/4 subunits in AgB oligomers of all samples analyzed. Quantitative and qualitative differences in the expression of AgB subunits were observed within and between samples. Using recombinant subunits as models, we showed that AgB subunits form distinct oligomeric states, with a rAgB8/3>rAgB8/2>rAgB8/1 maximum size relation. We also demonstrated by different experimental approaches that rAgB8/3 oligomers are more similar, both in size and morphology, to those observed for E. granulosus AgB. Overall, we provided experimental evidences that AgB is composed of different subunits within a single cyst, and that subunits have different abundances and oligomerization properties. These issues are important for the understanding of AgB expression and structure variations, and their impact for the host-parasite cross-talk.

Abstract Introduction Methods Results Discussion

medicine biochemistry infectious diseases veterinary diseases proteins zoonotic diseases genetics biology parasitic diseases zoology biophysics genetics and genomics veterinary science

2012

Echinococcus granulosus Antigen B Structure: Subunit Composition and Oligomeric States

Four related cows showed hairless streaks on various parts of the body with no correlation to the pigmentation pattern. The stripes occurred in a consistent pattern resembling the lines of Blaschko. The non-syndromic hairlessness phenotype observed occurred across three generations of a single family and was compatible with an X-linked mode of inheritance. Linkage analysis and subsequent whole genome sequencing of one affected female identified two perfectly associated non-synonymous sequence variants in the critical interval on bovine chromosome X. Both variants occurred in complete linkage disequilibrium and were absent in more than 3900 controls. An ERCC6L missense mutation was predicted to cause an amino acid substitution of a non-conserved residue. Analysis in mice showed no specific Ercc6l expression pattern related to hair follicle development and therefore ERCC6L was not considered as causative gene. A point mutation at the 5' -splice junction of exon 5 of the TSR2,20S rRNA accumulation, homolog (S. cerevisiae), gene led to the production of two mutant transcripts, both of which contain a frameshift and generate a premature stop codon predicted to truncate approximately 25% of the protein. Interestingly, in addition to the presence of both physiological TSR2 transcripts, the two mutant transcripts were predominantly detected in the hairless skin of the affected cows. Immunohistochemistry, using an antibody against the N-terminal part of the bovine protein demonstrated the specific expression of the TSR2 protein in the skin and the hair of the affected and the control cows as well as in bovine fetal skin and hair. The RNA hybridization in situ showed that Tsr2 was expressed in pre- and post-natal phases of hair follicle development in mice. Mammalian TSR2 proteins are highly conserved and are known to be broadly expressed, but their precise in vivo functions are poorly understood. Thus, by dissecting a naturally occurring mutation in a domestic animal species, we identified TSR2 as a regulator of hair follicle development. In 1901, the German dermatologist Alfred Blaschko proposed that congenital linear skin lesions could develop independently of the nervous system [1]. Blaschko observed a common non-random developmental pattern of the skin and described it extensively depicting the shape of the pattern lines [1,2]. The so-called lines of Blaschko run along the sides of the individual’s body, bending in a roughly S-shaped pattern toward the ventral part, forming a typical symmetrical V shape near the center of the back [3]. These lines become clinically manifest in the heterozygous state of various human X-linked inherited defects, such as incontinentia pigmenti, focal dermal hypoplasia, chondrodysplasia punctata, hypohidrotic ectodermal dysplasia, and Menkes syndrome [4,5, 6]. The inactivation of one X chromosome (XCI), which leads to mosaicism for cells with the mutant allele silenced, can explain different patterns of functional mosaicism in over a dozen X-linked conditions [5,6]. The pattern of cutaneous mosaicism can be tracked back to the type of cell affected, and its trajectory of migration and proliferation during embryogenesis [3,4]. Lines of Blaschko are due to ectodermal precursor cells which migrate and proliferate along these tracts. In female mammalian embryos, one of the two X chromosomes in each somatic cell is silenced in early development, albeit additional events can skew the inactivation [7]. Consequently, every female is a functional mosaic of cells, each exclusively expressing her maternal or paternal copy of X-chromosomal genes. In general, the effects of an X-linked gene mutation depend on XCI patterns. For genes subject to XCI, a mutation which affects males does not necessarily affect females who can be unaffected either due to random XCI or by selective skewing in favor of cells which express the normal allele [7]. Several forms of inherited alopecia have been described in domestic animal species (OMIA 001702–9913, OMIA 001702–9615, OMIA 001702–9796, OMIA 001702–9685, OMIA 001702–9825, OMIA 000031–9615, OMIA 000030–9685, OMIA 000030–9031, OMIA 000030–9940) [8], including hairlessness and X-linked phenotypes (OMIA 000543–9913) [9,10,11]. Our group has recently reported a family of horses in which females developed signs of a skin disorder reminiscent of human incontinentia pigmenti (OMIA 001899–9796) [10]. Notably, the affected horses showed congenital streaks of varying coat color which followed the lines of Blaschko, and a causative nonsense mutation was found in the X-chromosomal IKBKG gene [10]. In general, the dissection of naturally occurring spontaneous mutations in domestic animals can lead to important insights into developmental genetics, as has been shown for hairless dogs carrying a FOXI3 mutation (OMIA 000323–9615) [12]. In a dairy farm in Friuli (Italy), an X-linked inherited non-syndromic congenital hairlessness phenotype was detected in four cows showing hairless stripes in a consistent pattern resembling the lines of Blaschko. The condition was strikingly similar to the so-called streaked hairlessness phenotype reported 60 years ago in female Holstein cattle in North America, which was supposed to be X-linked dominant inherited with a lethal effect on hemizygous male embryos (OMIA 000542–9913) [11]. The goal of the present study was to identify the causative gene for bovine streaked hairlessness using a positional cloning strategy. The presence of skin lesions was detected in a total of four related female Pezzata Rossa cattle. The hairless lesions, present from birth, varied in their extent and size in the different animals but were all characterized by streaks of hairless areas following a vertical pattern. At the time of the first consultation the most severely affected animal was a 21-month-old pregnant heifer (case 1). Hairless streaks were present bilaterally along both sides of the animal (Fig 1A and S1 Fig). Their V-shaped symmetrical convergence at the level of the back resulted in a fishbone-like pattern (Fig 1B and 1C). On the right flank, approximately over the last three ribs, a larger area of hairlessness was also present (Fig 1A). Hairless streaks were also present on the head. The skin of the udder presented diffuse non-streaked hypotrichosis. The lesions occurred without any association to the coat color, both pigmented and unpigmented areas being affected (Fig 1C). Apart from the hairlessness, the skin of the affected areas was smooth, of normal color and without any crusts. No macroscopic intermediate aspect was present between the affected area and the surrounding skin. No abnormal cutaneous pain sensations by pressure, pricking or pinching stimuli, were observed at the level of the hairless areas as compared to the haired skin. Pruritus was also not apparent. The heifer showed no other clinical findings. The lesions remained practically unchanged during the three-year observation period, and no sign of hair regrowth was observed. During this period, the animal gave birth to three healthy calves, two males and one female. Both males were sold at the age of approximately one month and did not show any signs of hairlessness at that time. The female offspring of case 1 was examined for the last time at the age of 16 months and no skin abnormalities were detected, although similar, but less severe streaked hairlessness, was present in the dam (case 2) and in the granddam (case 3) of the aforementioned heifer (case 1). In the mother, the streaked lesions were limited to the rump and shoulders (S1 Fig), whereas, in the grandmother, the phenotype was diffusely evident at the level of the rump, back and hips (S1 Fig). The reported lesions had been present since birth and had the same characteristics as those described above (S1 Fig). Streaked hairless lesions were also present at the level of the rump, shoulders and the dorsal portion of the ribs of a forth case, a 15-month-old heifer (case 4), a half-sibling of case 2 on the side of their dam (S1 Fig). No other clinical signs were observed in these three additional cases. No alteration in the production of milk was reported but, with respect to the cows’ fertility, the owner reported that case 3 failed to conceive for five successive inseminations. The hair follicles in the biopsies from the haired skin were normally distributed, and size and shape were comparable with hair follicles in skin biopsies from non-affected cows (Fig 2A). In the skin biopsies from the hairless sites, the vast majority of the hair follicles and sebaceous glands were missing whereas the sweat glands, their ducts and the arrector pili muscles were present (Fig 2C). Dysplastic or miniaturized hair bulbs or remnant fibrous sheaths surrounding the bulb were occasionally present. In addition, remnants of infundibula were rarely seen. In the biopsies from the haired-hairless border, a mixture of normal hair follicles and dysplastic infundibuli were present (Fig 2C). The dysplastic infundibula were smaller than those of normal hair follicles, had an irregular outer contour and were often associated with the sebaceous glands and the ducts of the sweat glands (S2 Fig). The matrilinear descent of the affected animals, two of them being the female offspring of the oldest one and another being her granddaughter, is depicted in Fig 3A. Each one of the offspring was generated using different artificial insemination sires without any other comparably common ancestor of the four affected animals. Taken together, the segregation pattern of the observed phenotype can be explained by a monogenic X-linked inherited mutation causing the streaked hairlessness condition. Karyotype analysis of three of the affected animals (cases 1,2 and 3) and one healthy male offspring of case 1 was initially performed. The karyotypes appeared completely normal revealing no evidence for any visible numerical or gross structural chromosomal aberration (S3 Fig). To map candidate regions for the streaked hairlessness condition, we genotyped four affected cows, and a total of eight available normal family members for 777,962 SNPs (Fig 3A). A haplotype analysis searching for disease-linked haplotypes shared across the four affected animals was carried out. A 29. 2 Mb shared haplotype on BTA 7 (position 82,876,246 to the end of the chromosome) and an 11. 5 Mb shared haplotype on BTA 14 (position 21,284,128 to 32,783,095) were found. All four affected cows shared one single haplotype spanning the entire X chromosome (Fig 3A). In addition, all three non-affected offspring of case 1 were checked for the presence of the shared X haplotype and a recombinant X chromosome was detected in one son. A multipoint parametric linkage analysis revealing positive LOD scores on bovine chromosomes (BTA) 7,14 and X was carried out (Fig 3B and S4 Fig). In a critical interval of 118. 1 Mb on chromosome X (position 30,947,683 down to the end of the chromosome) the highest multipoint LOD score of 1. 405 was detected (Fig 3B and S4 Fig). The entire genome of one affected cow (case 1) was sequenced and the three genomic regions showing positive LOD scores in the linkage analysis were then focused on. Since the phenotype was mild and did not affect normal life, all variants present in the mapped regions including synonymous, nonsense and missense exon variants, and variants in the introns and splicing sites of annotated genes and intergenic polymorphisms were considered as potential causative mutations. A total of about 8. 8 million including 86,326 coding variants were called with respect to the reference genome (Table 1). A comparison was then made between all 361,134 DNA variants in the candidate regions present in the sequenced affected cow and 83 cow genomes of various cattle breeds which had been sequenced in our laboratory in the course of other studies. Thanks to this first step of filtering, the number of variants was reduced to 2593 including 21 coding variants of which all but one present on chromosome X. In a subsequent step, our membership in the 1000 bull genomes project was made use of [13] and the run4 variant database including 1119 genomes was used. This second filter step allowed the exclusion of 2564 variants remaining with 29 private sequence variants: 27 private variants located in intergenic and intronic regions on BTA 7 and two private non-synonymous coding variants located on the X chromosome in the excision repair cross-complementation group 6-like (ERCC6L) and TSR2,20S rRNA accumulation, homolog (S. cerevisiae) (TSR2) genes (Table 1, S1 Table). Due to the observed segregation pattern in the affected cattle family and the highest LOD score on the X chromosome a prioritization of the two non-synonymous coding variants located on the X chromosome was made. Collectively, these data do not strongly support the non-coding BTA 7 variants as causative mutations. In addition to the SNP and short indel variant calling, large deletions contained in the candidate regions were searched for using 41 sequenced control cow genomes which were selected in order to have a genome-wide coverage of more than 10×. Of the 11,784 deletions detected across the whole genome of the sequenced cow, 49 were private structural variants occurring only in the genome of the affected cow. One heterozygous deletion found exclusively in the affected animal was detected situated in one of the mapped regions, at position 128,716,121 in chromosome X. The 4039 bp deletion is 2271 bp upstream of the first exon of the membrane-bound transcription factor peptidase, site 2 (MBTPS2) gene. Subsequent PCR analysis confirmed the presence of this variant in case 1, in its sire and its three unaffected offspring, and its absence in the other family members including the other three affected cows. The first of the two remaining private variants was a missense mutation in the bovine ERCC6L gene (c. 54G>A) predicted to change an amino acid (p. A18T). The second private variant was a point mutation affecting the 5' -splice junction of exon 5 of the TSR2 gene (c. 441+226A>G). Both private variants were genotyped in all family members, and in two different cohorts of controls. The first cohort consisted of 1043 Pezzata Rossa cattle belonging to ten farms present in the same region including the farm of the four affected cows. All the Pezzata Rossa cattle were found to be free of streaked hairlessness. The second cohort consisted of 1682 animals of different cattle breeds from the DNA database present in our laboratory which had been collected during various studies. All four affected cows were heterozygous for both variants and the normal family members carried only the wild type allele. Both variants associated perfectly with the condition and were absent in all controls (Table 2). In silico analysis was then carried out on ERCC6L which predicted the p. A18T amino acid change as non-damaging with a Polyphen score of 0. 002 out of 1. The predicted altered protein sequence of the mutant ERCC6L protein was aligned with the homologs of several other mammalian species which showed that the affected residue was not conserved across mammals (S5 Fig). Interestingly, threonine is present in the ERCC6L protein sequence of the African elephant. Furthermore, expression analysis in mice showed no specific pattern related to hair follicle development (S6 Fig). Collectively, these data do not support ERCC6L as the causative gene. The TSR2 mutation was predicted to affect splicing because it altered the conserved splice acceptor sequence AG at the 3’-end of intron 4, which was changed to GG. An RT-PCR was carried out to test the consequences of the 5' -splice junction mutation of exon 5. Therefore, primers located in exons 3 and 5 of TSR2 were used to amplify cDNA from the affected and the unaffected skin of two cases, the normal skin of a related control (the first male offspring of case 1), and three unrelated controls (Fig 4A). The presence of two wild type transcripts was confirmed by Sanger sequencing in all tissues (Fig 4B). In the hairless skin of the affected cow, an additional prominent second band ~200 bp larger in size was detected. This additional band was also present in a much lower intensity in the normal haired skin of the affected cow. The RT-PCR products obtained from the hairless skin were cloned and Sanger sequencing of the various clones was performed. About 88% and ~2% of wild type transcript 1 and 2, respectively, and ~10% mutant transcripts were identified (Fig 4B and S7 Fig). The most common (~80%) mutant transcript 1 (mt1) was due to the retention of intron 4; splicing did not occur and exons 4 and 5 were separated by intron 4 in the transcript (c. 441_442ins226). A less frequently occurring (~20%) second mutant transcript (mt2) was the result of alternate splicing, thereby activating a cryptic splice acceptor site 7 bp downstream which led to skipping the first 7 nucleotides of exon 5 (c. 441_448del7) (Fig 4B and S7 Fig). Both mutant mRNAs contained a frameshift, and generated a premature stop codon predicted to truncate approximately 25% of the protein (mt1: p. Ala147Lysfs10*; mt2: p. Val146Leufs29*; S7 Fig). To verify the presence of the TSR2 protein in bovine skin, bovine fetal skin in different developmental stages, and hairless and normally haired skin of an affected and a control cow were used. Therefore, a species-specific antibody against the N-terminal part of the bovine protein was designed. A nuclear signal was detected in all epithelial cells in the hairless and the haired skin of the affected animal, and in the control cow (Fig 5). The TSR2 protein was strongly expressed in both cows within nuclei of epidermal and follicular keratinocytes, including cells of the hair bulbs as well as dermal papillae. In both cows, the protein expression was particularly strong in the root sheath. In the hairless skin areas of the affected cow, the root sheet was not present due to severe follicular atrophy. Nuclear expression was also observed in the majority of cell types present in the dermis including endothelial cells, epithelial cells of the sebaceous glands and the sweat glands, smooth muscle cells, infiltrating leukocytes and fibroblasts (Fig 5). The signal was not specific only for the hair follicle but also for the haired skin of both the case and the control which showed a denser signal on the upper part of the follicle toward the bulge (Fig 5). In addition, a nuclear TSR2 signal similar to that in adult cows was detected in all fetal samples (Fig 6). In the earlier fetal stage at day 177, TSR2 expression was detectable in all cells of the developing dermis. Interestingly, this signal was strongest in the hair bulb. At days 230 and 268, a strong signal appeared in the inner and outer root sheath of developing hair follicle (Fig 6). To further elucidate TSR2 expression in the hair follicle, in situ hybridization (ISH) was performed at different stages of murine hair follicle morphogenesis and the postnatal hair cycle. Tsr2 expression was detected in hair placodes using whole mount ISH at embryonic day 14. 5, at the onset of hair development (Fig 7A and 7C) whereas the sense probe gave only a faint background signal (Fig 7B and 7D). In situ hybridization on sections with a 35S-labeled Tsr2 probe also revealed low levels of expression in hair follicles during embryonic and postnatal growth phases as well as at the onset of anagen, the growth phase of the hair cycle (Fig 7E–7L). At all the stages analyzed Tsr2 was enriched in the epithelial compartment of the hair follicle at sites where actively proliferating cells reside: in the growing edge of early postnatal hair follicles and in the pool of transit amplifying cells of the cycling hair follicles at the beginning of anagen. No signal was detected when ISH was performed with the Tsr2 sense probe (Fig 7), confirming the specificity of the antisense probe. A rare non-syndromic hairlessness phenotype was observed in cattle which could be explained by an X-linked mode of inheritance. This disorder occurred across three generations of a single family of Pezzata Rossa cattle and showed a striking similarity to a sex-linked inherited condition described as streaked hairlessness [11]. Eldridge and Atkinson reported affected females in a pedigree of Holstein Friesian cattle showing approximately perpendicular areas devoid of hair on various parts of the body with the hairless areas occurring in consistent patterns which were highly variable in size [11]. In comparison with the disease phenotype in our study, the only difference lay in the fact that the owner of the affected cows reported no differences in cold endurance which represented a difficult feature to assess due the difference among breeds and the zones in which the animals had been raised. The four related cows reported in the present study showed hairless streaks on various parts of the body regardless of the pigmentation. Interestingly, the affected streaks were S-shaped on the sides with a typical V shape near the center of the back occurring in a consistent pattern resembling the lines of Blaschko. Genetic mapping confirmed the initially suspected X-linkage and this congenital anomaly therefore added another example to the list of X-linked conditions with visible skin manifestations [6]. The characteristic appearance of the skin in the affected females is most probably correlated with the X-inactivation, as the lines of Blaschko are typically visible in heterozygous females of X-linked disorders affecting hair development such as incontinentia pigmenti, focal dermal hypoplasia or hypohidrotic ectodermal dysplasia [4–6]. The phenotype presented showed no typical features of ectodermal dysplasia since only hair and no other ectodermal derived organs, such as eccrine glands or teeth, were affected. Notably, the anomaly was restricted to the regional absence of only hair follicles and sebaceous glands. The manifestation of X-linked phenotypes depends largely on the way in which cells subsequently divide and migrate, and is best studied in skin diseases [6]. The archetypal cutaneous pattern described by the dermatologist Alfred Blaschko [1] was later explained by the mosaicism which resulted from XCI in migrated ectodermal skin cells of females [4]. It was hypothesized that the four affected females who were heterozygous for private mutations on the X-chromosome showed varying phenotype expression affecting only small parts of their skin due to skewed X-chromosome inactivation (XCI). This is known to influence the appearance and severity of X-linked traits in heterozygous females by selective skewing in favor of cells which express the wild type alleles [6,7]. In the earlier report of bovine streaked hairlessness lethality in males carrying a copy of the putative X-linked mutation was assumed, thus supporting our hypothesis that the affected gene was subject to XCI. Evidence that the X-linked streaked hairlessness phenotype is likely caused by a disruptive mutation disturbing the normal splicing of the TSR2 gene was provided. During our study, positional cloning using linkage analysis and mutation analysis using whole genome sequencing were combined. Access to sequenced genomes of other cattle breeds and to the 1000 bull variant database [13] was very useful in detecting the disease-associated mutations. These filter steps allowed us to significantly reduce the number of associated variants within the critical regions. The investigation was not restricted to SNPs or short indels affecting annotated genes since the observed phenotype was mild and unclear in its definition. It was therefore taken into account that other types of mutations, such as larger structural variants, could cause the disorder. The 2. 3 kb deletion identified upstream of the MBTPS2 gene, a candidate gene for a skin condition [14], was finally excluded as potentially causative due to its presence in non-affected family members. Nonetheless, two private, perfectly associated single nucleotide sequence variants remained which were located in two X chromosomal genes: ERCC6L and TSR2. These two variants were subsequently genotyped in more than a thousand animals of the affected Pezzata Rossa breed but they remained private for the four affected cows and obviously occurred in complete linkage disequilibrium although they were located nearly 13 Mb apart. The ERCC6L gene encoded a DNA helicase which acted as an essential component of the spindle assembly checkpoint. The amino acid substitution occurred in a residue located in a non-conserved region, and the mutant residue was found in the wild type protein sequence of the African elephant. Furthermore, the experiments in developing mice showed no specific expression in hair follicles. For this reason, it was concluded that the missense mutation in ERCC6L was unlikely to be causative for the condition observed. The remaining mutation in the TSR2 splicing site was shown to lead to two mutant transcripts predominantly expressed in the hairless skin of the affected cows. Neither mutant transcript contained the terminal part of the TSR2 protein the function of which is unknown. To date, the only known putative functional WGG domain is situated in the N-terminal region of the protein. It was therefore concluded that this TSR2 variant present on the X chromosome represented a candidate causal mutation for the naturally occurring condition. The exact function of the TSR2 gene during hair follicle development had not been clarified until now. In order to validate whether TSR2 might be a reasonable functional candidate gene for the observed disorder, its expression in different stages of bovine and murine hair follicle morphogenesis and cycles was analyzed, including the time periods during which the ectodermal differentiation leading to the formation of hair takes place. Using tissue from the affected and control cows, TSR2 protein expression was detected in adult bovine skin. A clear signal was detected in the hair follicle, confirming the presence of the associated protein in the tissue affected by the condition. In order to investigate the presence of the protein during development, samples from bovine fetal skin from a previous study estimated as being from day 177,230 and 268 of gestation were used [15]. These three time points were chosen because they represented critical moments during hair follicle development: in the developing bovine embryo, one can detect a formed papilla between days 140–180, emerged hair between days 220–260 and the end of follicle length growth between days 240–280 [16]. A hair follicle is a dynamic self-renewing organ which periodically regenerates through cycles of regression (catagen), rest (telogen) and new growth (anagen). Hair follicle development is initiated during embryogenesis by the formation of an epithelial thickening (a placode) and an associated mesenchymal condensate (a dermal papilla). After the initial period, the hair follicle grows downwards into the mesenchyme and, once morphogenesis is completed, it enters the first hair cycle [17]. In mice, morphogenesis is completed by postnatal (PN) days 13–15, first catagen is initiated at ~PN17, and first anagen at ~PN20 [17,18]. The fact that Tsr2 mRNA was expressed in mouse hair follicles at the initial stage of development, in the growing hair follicles during embryogenesis and in anagen follicles in the adult skin at the site where proliferating progenitor cells reside was shown. Defects in cell proliferation during anagen could lead to impaired hair follicle down growth. Expression at the site of the proliferating cells in developing murine hair follicles suggests that Tsr2 could be important for hair growth. Currently, little is known regarding the cellular function of TSR2. Studies involving yeast have suggested a role in 20S rRNA processing [19,20]. Cytoplasmic cleavage of the 20S pre-rRNA to 18S is critical for the maturation of 40S subunits; the depletion of Tsr1, the paralog which is essential to ribosome biogenesis [21,22], and Tsr2 all lead to 20S accumulation [19,23–27]. Fassio et al. found TSR2 nonessential for yeast survival, but deletion resulted in slow growth with a doubling time of ∼2. 5 hrs in addition to a prominent 20S accumulation and a corresponding 18S deficit [20]. The paralog TSR1 is detected in yeast in both the nucleus and the cytoplasm, but is predominantly nuclear in exponentially growing cells [22–27]. A recent paper of Schütz et al [28] reported better insights into the function of the protein in yeast; it was shown that TSR2 bound released protein eS26, shielded it from proteolysis, and ensured its safe delivery to the 90S pre-ribosome. The authors defined the role of TSR2 protein as a nuclear carrier; its role is hypothesized to securely connect the nuclear import machinery with pathways which deposit r-proteins onto developing pre-ribosomal particles. A mutation within eS26 has been associated with Klippel-Feil syndrome in Diamond-Blackfan anemia [29–31]. A TSR2 missense mutation affecting the highly conserved predicted WGG domain (of unknown function) was reported to be associated with Diamond-Blackfan anemia with mandibulofacial dysostosis (Treacher-Collins syndrome) –a congenital anomaly involving absent external auditory canals and abnormal middle ears, micrognathia, unilateral cryptorchidism and a submucous cleft palate but no known hair phenotype [32]. Of note, the candidate mutation identified as causing streaked hairlessness in cattle did not affect the WGG domain. However it resulted in the formation of a C-terminal truncated version of the TSR2 protein. Notably, the C-terminal part of TSR2 is highly conserved among mammals, thereby suggesting a potential functional role of this domain, although no role has been inferred until now. We therefore speculate that the C-terminal part had a previously unknown important function during hair follicle development. In addition to its role in rRNA biogenesis, TSR2 is reportedly associated with other cellular processes. Behrends et al. identified TSR2 as one of the candidate interactors in the human autophagy system [33] whereas He et al. [34] reported that overexpression of TSR2 in human epidermal HEp-2 cells inhibited the transcriptional activity of NF-kappaB and induced HEp-2 cell apoptosis. The effect of the mutation appears to be circumscribed to the skin, even if TSR2 is supposed to be expressed ubiquitously. Cell or tissue specificity of the phenotype caused by a mutation in a gene expressed in the entire organism is not unknown, especially if some sort of compensation mechanism is not specifically available in the affected tissue [35]. In addition, probably skewed X-inactivation in favor of the cells expressing the wild type allele played an important role in the development and severity of the phenotype. The outcome of the study provided the first insights of the possible involvement of the TSR2 protein with a tissue or in a cell-specific manner. The TSR2 protein is potentially involved in several cell pathways, and the dynamics behind its relevance in several cell processes has yet to be unraveled. All animal research was conducted according to national and international guidelines for animal welfare. No permit number was necessary for the cattle as this study used naturally occurring cases. The bovine samples used were taken from different cattle farms in Italy, and all cattle owners agreed that the samples could be used in the study. The collection of fetal tissue, already used in previous studies [15], was carried out at a local government-authorized slaughterhouse in Switzerland since only a small number of pregnant cows are routinely slaughtered. All experiments involving mice were carried out in accordance with the guidelines and approval of the National Animal Experiment Board of Finland, the institute issuing the license is the Laboratory Animal Center of the University of Helsinki, and the license number is KEK13-020. Blood samples were collected from four affected Pezzata Rossa cows from the same farm. Genotyping of these cases was carried out using BovineHD BeadChip (illumina), including 777,962 evenly distributed SNPs at Geneseek (S1 Dataset). In addition, blood and semen samples were collected from eight cattle recorded as mates, parents and offspring of the affected cows (Fig 2A). A total of 1043 blood samples were collected from Pezzata Rossa cows from ten different farms in the region. The stored DNA samples from 1682 cattle belonging to several breeds previously subject of study, mainly Chianina, Romagnola, Simmental and Holstein Friesian were used. During the mutation analysis, 83 genomes of normal cattle from 17 genetically diverse Bos taurus breeds were used as local control cohort. The recent sequence variant database containing 1119 already sequenced genomes of the ongoing 1000 bull genomes project [13] was used as global control cohort during filtering for private variants of the sequenced affected cow. Eight millimeter skin punch biopsies were obtained from two affected cows and one normal offspring after subcutaneous injection of 2% lidocaine. The samples were collected at different sites, both at the level of the hairless streaks (lesional skin) and from grossly normal haired skin (unaffected skin), and from the border between haired and non-haired skin. All specimens were fixed in 4% buffered formaldehyde solution for histopathological examination or frozen at -80°C. After processing, they were embedded in paraffin, sectioned at 4 μm and stained with haematoxylin and eosin. PLINK v. 1. 07 software [36] was used to prepare the dataset for the linkage analysis using the—cow command to take into account the species specific number of chromosomes. The genotype data was pruned for the subsequently performed linkage analysis: (1) to remove SNPs with more than 10% missing genotype calls (—geno 0. 1); (2) to exclude uninformative SNPs with a minor allele frequency below 5% (—maf 0. 05); and (3) to exclude SNPs which exceeds the Hardy-Weinberg disequilibrium p-value of 0. 0001 (—hwe 0. 0001). MERLIN v 1. 1. 2 software [37] was used to analyze the dataset and carry out the linkage analysis. The—error was carried out in order to obtain a list of Mendelian errors and, hence, the SNPs to be excluded from the dataset. For all the autosomes, the multipoint LOD scores were calculated in a monoallelic autosomal dominant trait model, assuming complete penetrance. For the calculations, a frequency of 0. 15 for the mutated allele was assumed. In addition, the same parameters were used to analyze the X chromosome using MINX (part of the MERLIN package) which implements X-chromosome-specific versions of the functions provided by standard MERLIN. Due to the missing parents and the small number of cases, any result showing a positive LOD score was hypothesized to be suggestive of linkage. Graphs were traced with the—pdf command. Haplotypes were estimated using MERLIN by means of the—best command chromosome-by-chromosome (after extraction of each single chromosome from the dataset with PLINK using the—chr command). Haplotypes and markers were visualized using Haplopainter [38]. Heparinized blood samples were collected from one normal male calf and three affected females of the Pezzata Rossa cattle breed. The lymphocytes were cultured in 5 ml of RPMI-1640 medium containing 15% FCS, 1% L-glutamine (200 mM), 0. 6% heparin (50 mg/ml), 0. 8% pokeweed mitogen (80 μg/ml), 0. 1% penicillin (20. 000 U/ml) and 0. 1% streptomycin (20 mg/ml) for 72 h at 37°C. Two hundred microliters of Colcemide (10 μg/ml) were added to the cultures for the last 45 min of culturing. Incubation in a hypotonic solution of KCl (75 mM) at 37°C was carried out for 20 minutes, and the chromosomes were then fixed three times in a methanol: acetic acid solution (3: 1) and stored at -20°C. For each animal, 100 Giemsa-stained metaphases were analyzed using a Zeiss Axio Imager Z1 microscope. For each animal, ten metaphases were captured, and karyograms were prepared using IKAROS software (Metasystems). A fragment library with a 300 bp insert size was prepared and one lane of illumina HiSeq2000 paired-end reads (2x 100 bp) was collected; the fastq files were created using Casava 1. 8. A total of 767,575,378 100 bp paired-end reads were collected from a shotgun fragment library corresponding to roughly 28× coverage of the genome. The paired-end reads were then mapped to the cow reference genome UMD3. 1/bosTau6 and aligned using Burrows-Wheeler Aligner (BWA) version 0. 5. 9-r16 [39] with default settings. The mapping showed that 756,619,120 reads (98. 6%) had unique mapping positions. The SAM file generated by BWA was then converted to BAM and the reads were sorted by chromosome using samtools [40]. The PCR duplicates were marked using Picard tools (http: //sourceforge. net/projects/picard/). The Genome Analysis Tool Kit (GATK version 2. 4. 9, [41]) was used to carry out local realignment and to produce a cleaned BAM file. Variant calls were then made with the unified genotyper module of GATK. The variant data for each sample was obtained in variant call format (version 4. 0) as were raw calls for all samples and sites flagged using the variant filtration module of GATK. Variant filtration was carried out, following the best practice documentation of GATK version 4. The snpEFF software [42], together with the UMD3. 1/bosTau Ensembl annotation, was used to predict the functional effects of the variants detected. The genome data was made freely available under accession no. PRJEB8226 at the European Nucleotide Archive [43]. The Delly package [44] was used to detect structural variants in the cleaned BAM files. In order to avoid missing large inserts, deletions and false positives, all the variants detected were also manually inspected in the candidate region using 41 control genomes. The associated variants were genotyped by the re-sequencing of targeted PCR products using Sanger sequencing technology. The primers were designed using PRIMER3 [45]. The PCR products were amplified with AmpliTaqGold360Mastermix (Life Technologies), and the products were directly sequenced using the PCR primers on an ABI 3730 capillary sequencer (Life Technologies) after treatment with exonuclease I (NEB) and rapid alkaline phosphatase (Roche). The sequence data were analyzed using Sequencher 5. 1 (GeneCodes). Sequence alignment and mutation impact calculation for the ERCC6L mutant protein mutation was carried out with the prediction tool Polyphen 2 [46]. Sequence alignment was carried out using ClustalW [47]. The RNA was extracted from skin tissues using the RNeasy mini kit (Qiagen). The tissue was first finely crushed in TRIZOL (Ambion) using mechanical means, chloroform was then added and the RNA was separated by means of centrifugation. Additional passages were carried out as described by the manufacturer. The RNA was cleared of genomic DNA contamination using the Quantitect Reverse Transcription Kit (Qiagen). The same kit was used to synthetize cDNA, as described by the manufacturer. An RT-PCR was carried out using AmpliTaqGold360Mastermix (Life Technologies). The RT-PCR products were sequenced as described above. The products were ligated to TOPO TA cloning plasmids pCRII (Invitrogen), as described by the manufacturer. For whole mount ISH, E14. 5 mouse embryos were dissected, fixed in 4% paraformaldehyde PFA, and dehydrated using methanol series. Whole-mount in situ hybridization with a digoxygenin-labelled Tsr2 probe was performed according to a standard protocol using InsituProVS instrument (Intavis Bioanalytical Instruments) [48]. The Tsr2 antisense and sense probes corresponded to nucleotides 40–822 of NM_175146. 4. The probes were detected with BM Purple AP Substrate Precipitating Solution (Roche Applied Science). For radioactive ISH, mouse back skins were fixed overnight in 4% PFA, dehydrated in ethanol, embedded in paraffin and sectioned at 5 μm. Radioactive in situ hybridization with a 35S-UTP (Amersham) -labeled Tsr2 probe was carried out according to standard protocol [48]. To generate the expression plasmids encoding the wild type and mutant (mt1) proteins (pCI-W, pCI-M), the two relevant sequences were synthesized (Eurofins). The plasmids were HA-tagged (peptide YPYDVPDYA). Next, pCI-RFP-HA-W and pCI-RFP-HA-M, and the plasmids were generated by the PCR amplification vector and were subsequently cloned into the pCI-RFP-Linker-HA-cleaved plasmid. Competent XL10-Gold Ultracompetent Cells were transformed as described above, and the plasmid was recovered and used to transfect the Vero cells. Vero cells expressing the two constructs were grown in Dulbecco’s modified Eagle’s medium (Invitrogen) with 10% fetal calf serum at 37°C in the presence of 5% CO2. A positive and specific signal was obtained for the proteins translated from both transcripts from the wild-type and mutant vectors expressed in vero cells. The antibodies were designed against the N terminal part of the TSR2 protein and synthesized in rabbits (ProteoGenix). Western blots were carried out as previously described [49]. Transfected cells were washed twice with cold PBS before adding 150 μL of lysis buffer [10 mM Tris, pH 7. 4,150 mM NaCl, 1% deoxycholate, 1% Triton X-100,0. 1% sodium dodecyl sulfate (SDS) ] with a complete protease inhibitor (Roche). After incubation for 20 min at 4°C, the lysates were cleared by centrifugation at 5000 g for 15 min at 4°C, and the supernatant was mixed with an equal amount of Laemmli sample buffer (Bio-Rad) containing 100 mM dithiothreitol, subsequently boiled at 95°C for 5 min, and fractionated on 8% or 10% SDS-polyacrylamide gel under denaturing conditions. The separated proteins were transferred to nitrocellulose membranes by electroblotting. The membranes were then incubated with polyclonal rabbit anti-CDV antisera. Following incubation with a peroxydase-conjugated secondary antibody, the membranes were treated with an enhanced chemiluminescence (ECL) kit (Amersham) according to the manufacturer’s instructions. The transfected cells were washed with PBS and fixed with 500 μl 4% PFA (paraformaldehyde) for 20 min at room temperature. After a PBS wash, the cells were permeabilized with 500 μl to 1 ml of 2% Triton in PBS for 20 min at room temperature. After a PBS wash, they were incubated for one hour with primary antibody 1 μg/ml per well. After a PBS wash, secondary antibody goat anti-rabbit antibody Alexa fluor was diluted 1/1000 and incubated one hour before acquisition. Skin samples from normal fetuses collected at the slaughterhouse, and from affected (haired and hairless areas) and unaffected cows were embedded in Optimal Cutting Temperature compound (OCT) and were snap frozen by immersion in 2-methylbutane (59080; Sigma—Aldrich), which was chilled in liquid nitrogen. The frozen tissue blocks were stored at -80°C until cutting. Immunohistochemistry was carried out as previously described [50]. Briefly, cryostat sections were fixed in ice-cold acetone for 3 min and endogenous peroxidase activity was quenched by incubation with 3% hydrogen peroxide in methanol for 30 min. A protein block was obtained by applying 10% normal goat serum in PBS for 30 min. The slides were incubated with the primary antibody (the in-house produced polyclonal rabbit antibody) at 4◦C overnight. The positive reactions were detected with a LSAB-kit (Dako) according to the manufacturer’s instructions.

The identification of causal mutations of rare monogenic disorders provides an insight into the function of single genes. We herein report an example which demonstrates that the bovine species presents an excellent system for identifying these inherited phenotypes. The individual health status of modern dairy cows is well monitored, and emerging disorders are routinely recorded. An Italian breeder of ~500 Pezzata Rossa cattle reported a case of congenital streaked hairlessness. Three additional, closely related cows, showing similar hairless pattern following Blaschko’s lines were subsequently observed. A causative mutation was discovered in a previously uncharacterized rRNA processing gene. Cows possessing a single copy of this TSR2 mutation located on the X chromosome showed a mosaic skin pattern which is very likely due to the skewed inactivation of the X-chromosome, also known as lyonization. The expression of TSR2 was shown in skin and hair of cattle and mice. This study is the first to implicate an essential role for TSR2 during hair follicle development and reflects once more the potential of using rare diseases in cows to gain additional insights into mammalian biology.

Abstract Introduction Results Discussion Materials and Methods

2015

Hairless Streaks in Cattle Implicate TSR2 in Early Hair Follicle Formation

Large genome-wide association studies (GWAS) have identified many genetic loci associated with risk for myocardial infarction (MI) and coronary artery disease (CAD). Concurrently, efforts such as the National Institutes of Health (NIH) Roadmap Epigenomics Project and the Encyclopedia of DNA Elements (ENCODE) Consortium have provided unprecedented data on functional elements of the human genome. In the present study, we systematically investigate the biological link between genetic variants associated with this complex disease and their impacts on gene function. First, we examined the heritability of MI/CAD according to genomic compartments. We observed that single nucleotide polymorphisms (SNPs) residing within nearby regulatory regions show significant polygenicity and contribute between 59–71% of the heritability for MI/CAD. Second, we showed that the polygenicity and heritability explained by these SNPs are enriched in histone modification marks in specific cell types. Third, we found that a statistically higher number of 45 MI/CAD-associated SNPs that have been identified from large-scale GWAS studies reside within certain functional elements of the genome, particularly in active enhancer and promoter regions. Finally, we observed significant heterogeneity of this signal across cell types, with strong signals observed within adipose nuclei, as well as brain and spleen cell types. These results suggest that the genetic etiology of MI/CAD is largely explained by tissue-specific regulatory perturbation within the human genome. Coronary artery disease (CAD) and myocardial infarction (MI) remain among the leading causes of infirmity and death worldwide despite advances in and widespread adoption of medical therapies treating this disease. Studies have shown a large genetic component for CAD, with the heritability estimated to be 30–60% [1]. Large-scale genome-wide association studies (GWAS) have identified common single nucleotide polymorphisms (SNPs) at 45 loci associated with MI/CAD risk [2–8]. Although these newly discovered loci have led to important new biological insights for MI/CAD [9,10], the proportion of the heritability explained by these loci represents approximately 15% of the estimated heritability [8]. Therefore, a large proportion of the genetic effects are apparently not explained by known loci. This phenomenon has been similarly observed with GWAS for other complex diseases [11]. Previously, we modeled the genetic architecture of MI and CAD using GWAS data [12]. Using simulated genetic models, we inferred that a polygenic model comprised of thousands of associated common variants with small effects explains the majority of the heritability (proportion of total liability-scale variance explained is 0. 48,76% of family-study h2) for MI/CAD [12]. Furthermore, recent work has partitioned heritability of complex diseases into broad categories of the genome [13]. Pooling 11 common diseases, including CAD, Gusev et al. have shown that heritability is disproportionally represented in regulatory elements, specifically in DNase I hypersensitivity sites (DHS) (h2g = 79% for SNPs in DHS regions). However, a specific analysis of CAD yielded a wide interval of the true enrichment of DHS, with genotyped estimates of h2g = 10. 9 to 71. 3% (95% confidence intervals) for SNPs in DHS regions. Hence, other than being broadly distributed throughout the genome, the molecular consequences of MI/CAD-associated common variants largely remain undefined. Furthermore, it is unclear which cell or tissue types are influenced the most by MI/CAD-associated SNPs. We addressed these unresolved issues by leveraging data from the National Institutes of Health (NIH) Roadmap Epigenomics Project [14,15] and the Encyclopedia of DNA Elements (ENCODE) Consortium [16]. These projects have comprehensively catalogued biochemical functional regions such as those critical for transcription, transcription factor binding, chromatin structure and histone modification in different cell types, providing unique opportunities to scrutinize links between non-protein-coding DNA sequence variants and gene function. Studies have shown that SNPs in these functional DNA elements can regulate gene expression [17] and common disease-associated loci [18,19]. Using these data, we partitioned the genetic risk of MI/CAD into different categories, to discern drivers in specific cell types that may biologically influence MI/CAD. First, we investigated components of polygenicity and heritability in distinct genomic compartments. Second, we tested for differences in polygenicity, enrichment measures and heritability across diverse cell types within three histone modification marks. Third, we examined clusters of 45 loci discovered from recent GWAS meta-analyses [8] for MI/CAD mapping to the three histone marks in different cell types. Finally, we investigated whether specific biological networks were expressed differently in certain cell types. We imputed two GWAS datasets, the Myocardial Infarction Genetics Consortium (MIGen) and the Wellcome Trust Case Control Consortium (WTCCC) CAD, using reference haplotypes from the 1000 Genomes Project [20]. We imputed ~7 million SNPs with a high imputation quality metric (>0. 5) in both datasets. First, we investigated the relative contributions of different genomic compartments to MI/CAD risk. We partitioned the human genome into three distinct variant sets: “genic noncoding”, “genic coding” and “intergenic” (S1 Table) (see Materials and Methods). For each variant set, we performed two different analyses: 1) polygenic risk score analysis, where we test association of a genetic score comprised of multiple SNPs and 2) SNP-heritability analysis, where we estimate the variance in liability to MI/CAD [21] (see Materials and Methods). In the polygenic risk score analysis, we observed that the ‘genic’ variant set, defined as genomic regions within 10 kilobases (kb) upstream and downstream of a gene, showed a substantially stronger signal than SNPs in intergenic regions (defined as genomic regions outside of 10 kb of a gene) (Fig 1 and S1 Fig). Similar results were observed amongst genomic regions with window sizes of 20 kb and 50 kb (S2 and S3 Figs). Among this set of variants, both association (Fig 1) and the phenotypic variance explained by the polygenic risk scores before and after normalization by the number of SNPs (S1–S3 Figs) were the most significant in regions adjacent to the protein-coding DNA regions, called “genic noncoding” regions. We observed that polygenic risk scores were strongly associated with MI/CAD in “genic noncoding” regions (P<10−10), explaining 1 to 1. 5% of the phenotypic variance. By comparison, polygenic risk scores in regions within protein-coding DNA regions, called “genic coding” regions, were less strongly associated (P<10−5) and explained approximately 0. 5% of the variability. Similar patterns were observed after normalizing by the number of SNPs in the polygenic risk score. These patterns were particularly evident for P value thresholds <10−5 (see Materials and Methods). The association signal remained among the variant sets with discovery P value thresholds greater than or equal to 0. 05 after excluding regions within ±1 megabase of the 45 known SNPs associated with MI/CAD risk [8] (S4 Fig). We further examined the role of different genomic compartments on the heritability for MI/CAD risk by testing whether SNPs in the three compartments make up a large portion of the heritability for MI/CAD (Table 1 and S2 Table). Consistent with findings from our polygenic risk score analysis, we observed that most of the heritability resides within the “genic” regions. In a meta-analysis of the MIGen and WTCCC CAD studies, SNPs in “genic noncoding” regions explained approximately 58. 9% (variance in liability = 0. 25, P = 1×10−9) of the total heritability, resulting in a fold enrichment of variance of 1. 2. In contrast, the heritability of MI/CAD explained by SNPs residing within “genic coding” regions was estimated to be only 10% of the total (variance in liability = 0. 042, P = 0. 07). SNPs residing within “genic coding” regions accounted for only 0. 5% of the total number of variants, resulting in a high fold enrichment of variance of 19. 1. Despite the enrichment of variance estimates for both “genic coding” and “genic noncoding”, neither category statistically deviated beyond expectation (P = 0. 088 and P = 0. 23 respectively). On the other hand, a statistically significant depletion of variance in liability was observed in the “intergenic” regions compared to expectation (observed variance in liability = 0. 13, expected variance in liability = 0. 22 [0. 52 fraction of SNPs of total × 0. 42 total variance in liability], difference in observed and expected variance in liability P = 0. 0089). Similar results were observed amongst genomic regions with window sizes of 20 kb and 50 kb (S3 and S4 Tables). Heritability estimates with a prevalence of 3% of early-onset MI/CAD showed reduced heritability (variance in liability = 0. 35) explained by the genomic compartment whereas P values and enrichment of variance remained the same (S5 Table). Given the high polygenicity for MI/CAD explained by SNPs in noncoding regions surrounding protein-coding regions, we further examined whether polygenicity is stronger within specific regulatory elements. Using data from the NIH Roadmap Epigenomics Project (see URL), we specifically examined three histone modification marks that are indicative of active promoters (H3K4me3/H3K9ac) or enhancers (H3K27ac). A polygenic association signal comprised of SNPs with association P<0. 05 for MI/CAD was stronger in the histone marks, beyond what we expect by chance after randomly sampling “genic noncoding” regions outside of the marks (Mann-Whitney test P = 1. 1×10−95) (S5 Fig). We next tested for differences in polygenicity, enrichment and heritability estimates between different cell or tissue types within the three histone modification marks (Fig 2). We observed heterogeneity on the polygenicity of MI/CAD between cell types (Fig 2A). For example, SNPs in the H3K27ac and H3K9ac in bone marrow derived mesenchymal stem cell cultured cells and SNPs in the H3K4me3 regions in mesenchymal stem cell derived adipocyte cultured cells were amongst the strongest signals. Meanwhile, SNPs in any of the three histone marks in hematopoietic CD3, Treg, CD4, CD25, CD45RA primary cells were amongst the weakest signals. Enrichment analyses showed strong excesses for highly associated SNPs (P<10−5), compared with matched random SNPs, for mesenchymal stem cells, heart tissues such as fetal heart and ventricle, and intestinal mucosa including rectal and colonic mucosa, among others (Fig 2B). Cell-type specific effects were also generally consistent for heritability estimates (Fig 2C). The cell-type specific enrichment signals were strongest among strongly associated SNPs (Fig 2B). When tested for ENCODE data, statistically significant polygenicity and high heritability estimates were observed for variants residing in specific active chromatin regions, including enhancers and weak transcription regions (about 3 and 11% of reference genome, respectively) across several cell lines, including skeletal muscle and vascular cell lines (S6 Fig). Polygenicity and heritability estimates were substantially weaker for variants in inactive chromatin states than those in active states (S7 Fig), with the notable exception of the quiescent state. Heritability estimates for variants in a quiescent state were particularly high although the variants in this state accounted for >60% of the reference epigenome (S7 Fig). However, the enrichment of variance (% variance of total divided by % SNPs of total) in the quiescent state was low (on average 0. 76 [0. 50−0. 92] in different cell lines) compared with those for enhancers (on average 6. 5 [3. 2−10. 3]) and weak transcription (1. 9 [1. 1–2. 8]). In general, we note that ENCODE cell line chromHMM state results were largely driven by the sizes of these genomic compartments (i. e. % SNPs), which would be expected if alignment of MI/CAD genetic effects and ChromHMM stats were random. These observations have implications for fine-mapping loci discovered from large-scale GWAS. In particular, the results suggest that causal variants within GWAS loci are overrepresented in regulatory elements adjacent to protein-coding regions of the genome. Therefore, we investigated 45 independent, lead SNPs discovered from recent GWAS meta-analyses [8] for MI/CAD and tested whether a disproportionate number of these SNPs overlapped histone marks across diverse cell lines and tissues. We generated 10,000 random sets of SNPs that were selected to match the query GWAS SNPs based on similar minor allele frequency (±0. 05 frequency), number of SNPs in linkage disequilibrium (LD) with the query SNP (±10% of number of SNPs in LD with query SNP using r2>0. 5), distance to nearest gene (±10% of distance of nearest gene from query SNPs) and gene density (±10% of number of genes in loci around the query SNPs) [25]. We excluded two SNPs (rs3798220 and rs12205331) out of the 45 GWAS SNPs because we were unable to find appropriate matching null SNPs. We observed that 25 (58. 1%) out of 43 GWAS SNPs overlapped one of three histone marks. From the 10,000 random null sets, a median of 15 (34. 9%) out of the 43 random SNPs (1st quartile of 13 and 3rd quartile of 17) overlapped histone marks (S8 Fig). Only a very small fraction of random sets showed a higher number of GWAS SNPs overlapping with histone marks than what we observed (4 out of 10,000 random null sets) (P = 4×10−4). Because histone modification differed between cell types, we next examined whether the 45 MI/CAD GWAS SNPs yielded differential gene regulation across various tissues. We mapped the 45 GWAS SNPs, along with SNPs in high LD (r2≥0. 8), that reside within each of the three histone marks in different cell types. We observed distinct patterns between the different GWAS loci and cell types (Fig 3 and S9 and S10 Figs). For example, for the histone mark H3K27ac, 12 of the 45 loci were expressed in more than 80% of the cell types, whereas 13 of the 45 loci were expressed in less than 20%. Using HaploReg v2 [26], specific cell lines displayed enrichment with the 45 loci in strong enhancer regions. In particular, cell types related to adipose nuclei, spleen and brain tissue were amongst the strongest enrichment signals for inferred strong enhancer chromatin states (Table 2). We observed consistent results in these cell lines when also considering SNPs with less stringent significance levels using polygenic association analysis (Fig 2). HaploReg analysis for available ENCODE cell lines showed 24-fold enrichment with the 45 loci in enhancer regions in H1 cell lines (P = 3×10−5). Finally, to investigate whether specific biological networks are expressed in specific cell types, we examined connectivity of protein-protein interaction (PPI) networks among the 45 GWAS loci (45 GWAS SNPs, as well as SNPs in r2≥0. 8) in different cell lines. Consistent with the HaploReg v2 and polygenic association analysis, we observed high direct connectivity in a PPI network involving known lipid genes, particularly apolipoprotein E (APOE), apolipoprotein C3 (APOC3) and low-density lipoprotein receptor (LDLR) in adipose nuclei (P = 0. 002) and mesenchymal stem cell line derived adipocyte cultured cells (P = 0. 001), and apolipoprotein B (APOB) and Sortilin 1 (SORT1) in adult liver (P = 0. 009) (S11 Fig). Specific effects were observed in PPI networks in other cell types as well, including adult liver (S6 Table). We utilized several complementary human genetic approaches to partition the genetic risk of MI/CAD into different genomic categories and cell types. Three principal findings emerged: (1) genetic variants residing in noncoding regions flanking protein-coding genes make up a large proportion of the heritability for MI/CAD; (2) association signals are enriched in histone modification marks; and (3) clear cell-type specific effects emerged with genetic effects of MI/CAD-associated SNPs being enhanced in adipocyte cell lines. We highlight an important role for genetic variants residing within specific regulatory elements including promoters and enhancer regions, suggesting that the genetic risk for MI/CAD is largely driven by variants in key regulatory elements. Because we did not adjust for traditional risk factors of MI/CAD, our heritability estimates may include the fraction derived from these risk factors. Furthermore, our heritability estimate of 0. 43 for the total genome is consistent with previous family-based studies with estimates of 0. 3–0. 63 [1,28,29]. Our work is consistent with previous studies that showed that a significantly higher portion of GWAS SNPs overlap regulatory elements such as transcription factor binding regions and/or DNase I hypersensitive sites [18,30]. These findings have important implications for interpretation of GWAS signals, particularly for identifying causal variants. Our findings particularly highlight an important role for promoter and/or enhancer regions in fine-mapping GWAS signals. Our analyses further highlighted the genetic effects of MI/CAD-associated SNPs occurring in specific cell types, including mesenchymal and adipocyte cell lines. Adipocytes may affect cardiovascular homeostasis by regulating diverse peptides and nonpeptide compounds, which have been implicated in cardiovascular disease pathogenicity [31]. Furthermore, obesity is a casual risk factor for MI/CAD [32]. Here, we demonstrate that alteration of chromatin-mediated gene regulation by DNA sequence variation at key genomic regions within adipocytes is an important determinant of MI/CAD risk. Our results are consistent with studies that have shown that adipocytes play an important role in the pathogenesis of obesity, with adverse effects on inflammation, hemodynamics, and cardiovascular function [33]. We propose that the adipose-cardiovascular axis is not only an acquired mediator of MI/CAD risk but also a critical genetic determinant of MI/CAD risk in the general population. Furthermore, we highlight other loci that are not known to be involved in cardio-metabolism but appear to map to histone marks in the adipocyte cell lines. For example, SNPs in the MIA3 and MRAS loci are highly associated with CAD [2–4] but have not been found to be associated with any cardio-metabolic intermediate trait [34]. Notably, we observe enrichment of MI/CAD-associated SNPs disrupting regulation in brain and spleen tissue raising new hypotheses in the pathogenesis of human atherosclerosis. Hypothalamic-pituitary regulation of the autonomic nervous system has a diverse array of impacts on cardiovascular disease determinants including blood pressure, heart rate, sodium regulation, metabolism, and physiologic responses to stress [35]. Furthermore, proinflammatory mediators within the spleen may have an important role in MI [36]. Because the spleen consists of a multitude of cell types, including immune B/T cells, macrophages, monocytes, endothelial cells, smooth muscle cells from larger arterioles, chromatin data of specific cell types in the spleen may be useful to identify specific functional roles related to MI. The mechanism by which these tissues contribute to a higher enrichment to MI/CAD associated SNPs may be through tissue-specific regulatory perturbation in functional regulatory regions (histone marks and/or chromatin enhancer states) adjacent to protein-coding regions, specifically in adipose, heart, brain and spleen tissues. Thus we propose that genetic determinants of early-onset myocardial infarction have biologic roles within distinct tissue types and these observations should prioritize experimental modeling strategies. We note the following limitations. We note that the case samples in MIGen and WTCCC CAD were ascertained based on early-onset MI/CAD that was not fatal (male cases < 50 years old and female cases < 60 years old). Therefore, the genetic architecture and some of the tissue types highlighted in this study may be less relevant to the more general, broader CAD phenotype. Furthermore, we observed strong enrichment signals in some tissues (such as heart and intestinal mucosa) that may serve as a proxy for tissues that are biologically relevant to MI/CAD (such as smooth muscle in arterial walls). This may be due to tissues with similar cell types (i. e. smooth muscle cells in different tissues) having similar open chromatin structures. We note that we observed relatively weak signals for immune-related tissues and cell types despite a recent study suggesting that there was enrichment of tissue-specific eSNPs associated with CAD in pathways related to the immune system[37]. We have not examined our results in the context of biological pathways relevant to specific cell types (for example, immune/inflammation pathways in immune cells), and further investigation on this is warranted. In summary, we have shown that disease-causing variants for MI/CAD are enriched in promoters and enhancer regions flanking protein-coding genes. Functional data from the NIH Roadmap Project and ENCODE provide key links to specific cell types such as mesenchymal stem cell derived adipocyte cultured cells, heart, brain and spleen cells with risk variants for MI/CAD. Our results highlight the importance of investigating the noncoding regions of the genome in genetic studies and suggest a key role of tissue-specific regulatory mechanisms on the etiology of MI/CAD. Identifying critical nodes that are significant drivers of a substantial portion of human disease can both inform biological investigation and prioritize therapeutic efforts. We obtained GWAS data for 5,903 samples from MIGen [5] (2,905 cases and 2,998 controls) and 4,837 samples from WTCCC [38] (1,914 cases and 2,923 controls). For quality control, samples with extreme heterozygosity, gender mismatch or sample call rate <95% and SNPs with Hardy-Weinberg equilibrium P<10−6, minor allele frequency <1% or SNP call rate <98% were excluded prior to imputation. We prephased MIGen GWAS data using the MaCH software and imputed into the 1000 Genomes Phase I Integrated Release Version 3 Haplotypes panel using minimac [39]. We imputed WTCCC CAD data using IMPUTE2 [39–41]. To control for imputation quality, we removed low-frequency SNPs with minor allele frequency <0. 5% and used SNPs with high imputation quality metric (minimac rsq or IMPUTE2 info) >0. 5. We tested for association with additive tests for imputed dosage data using the SNPTEST [40,42] and PLINK [43] software. Gender, age and principal components for population structure [44] were used for covariates in the association analysis. A categorization of SNPs was adopted in order to compare their relative aggregate properties (S1 Table). “Genic” regions were defined as ±10 kb of the 3′ or 5′ untranslated regions (UTR) of a gene. “Intergenic” regions were those that were beyond ±10 kb of the 3′ or 5′ UTR of a gene. “Genic coding” variants were those that code amino acid sequence (nonsense, missense, synonymous). “Genic noncoding” variants were those that were resided outside of the coding region but within the “genic” window size. This includes the region beyond the 3’ or 5’ UTR, as well as the introns. 19. 5% of variants in the “genic noncoding” region are also observed in the DHS regions as defined by Gusev et al [45]. Genomic compartments for the “genic coding”, “genic noncoding”, and “intergenic” regions were defined to be non-overlapping. Other window sizes for “genic” regions of ±20 kb and ±50 kb were also tested. To examine polygenic effects of specific chromatin marks, we obtained histone modification marks (histone H3 lysine 4 trimethylation or H3K4me3, histone H3 lysine 9 acetylation or H3K9ac, histone H3 lysine 27 acetylation or H3K27ac) in diverse cell types or tissues that are indicative of active enhancers or promoters from the NIH Roadmap Epigenomics Project [14,15] data repository (see URL). We identified histone mark using the MACS v1. 4 software [46], with a P value cutoff of 10−5 and a false discovery rate cutoff of 0. 01. We ran the histone mark test files with control files matched by individual if available in order to increase specificity. We also obtained chromatin core 15-state data for 16 ENCODE cell lines (Epigenome ID E114-E119) (see URL). We examined eight active chromatin states (active transcription start site [TSS] proximal promoter states [active TSS, flanking active TSS], a transcribed state at the 5′ and 3′ end of genes showing both promoter and enhancer signatures [transcription at gene 5′ and 3′], actively transcribed states [strong transcription, weak transcription], enhancer states [enhancers, genic enhancers], and a state associated with zinc finger protein genes [ZNF and repeats]) and seven inactive states (constitutive heterochromatin, bivalent regulatory states [bivalent TSS, flanking bivalent TSS/enhancers, bivalent enhancers], repressed PolyComb states [repressed PolyComb, weak repressed PolyComb], and a quiescent state) inferred by ChromHMM [15]. Polygenic risk score (PRSi) for individual i for a given variant set was calculated as PRSi = ∑j∈SNPsβj×dij, where βj >0 is the natural log of odds ratio for the risk allele of SNP j from the association result in the MIGen discovery set and dij is the dosage (0−2) for that same allele of individual i from the WTCCC CAD validation set, as previously described [12]. Association of polygenic risk scores with disease status was tested using the Wald test and variance explained was estimated by Nagelkerke’s R2 from logistic regression [47]. Variance explained after normalization was calculated using Nagelkerke’s R2 divided by the number of SNPs. We used different discovery P value thresholds in MIGen (association P<5×10−8,5×10−7,5×10−6,5×10−5,5×10−4,0. 005,0. 05,0. 1,0. 2,0. 3,0. 4,0. 5,1) to define various polygenic risk scores (Fig 1). The proportion of phenotypic variance explained by each variant set was estimated using the restricted maximum likelihood method [48] implemented in the Genome-wide Complex Trait Analysis (GCTA) software, transformed to the underlying liability scale assuming a prevalence of 6% for CAD [22,23]. We removed one of each pair of individuals with estimated relatedness larger than 0. 05 (grm-cutoff 0. 05 in the GCTA software). Given a previous observation that LD does not substantially influence polygenic risk score analysis [49] or heritability explained by genotyped SNPs [13], and that imputed SNPs do not produce biased heritability estimates compared to genotyped SNPs [13], we included all SNPs in our variant sets (Table 1). We performed heritability estimates independently in the MIGen and WTCCC CAD studies, and then a meta-analysis across both studies using as weights the inverse variance (Tables 1 and S2–S4). We also estimated heritability with a prevalence of 3% for early-onset MI/CAD (S5 Table). We performed analyses on SNPs within three histone marks (H3K27ac, H3K4me3, H3K9ac) in different cell types. We performed polygenic risk score analysis as described above. We used a discovery P value threshold of <0. 05 in MIGen to form the polygenic risk score and then validation in WTCCC CAD (Fig 2A). We performed enrichment analysis by comparing the proportion of significant variants passing a specific association P threshold of a variant set with that of a baseline set. We tested different association P thresholds P<5×10−7,5×10−6,5×10−5. For the baseline set, we examined variants that resided in the intergenic regions, were not conserved (Genomic Evolutionary Rate Profiling score [50] <5) and did not overlap with any of the studied regulatory elements (Fig 2B). We performed heritability analysis as described above. We restricted heritability analysis to variants within histone marks for the indicated cell line using the restricted maximum likelihood method [48] implemented in the GCTA software (Fig 2C) [22,23]. For all approaches, we performed analyses restricting to SNPs residing in the three histone marks (H3K27ac, H3K4me3, H3K9ac) that were present in the different cell types For enrichment analyses, we selected 45 independent SNPs that were shown to be significantly associated with CAD in a large-scale GWAS meta-analysis (Table 2) [8]. We also included tag SNPs in strong LD (r2≥0. 8 in 379 individuals from European populations [85 CEU, 98 TSI, 93 FIN, 89 GBR, 14 IBS] from the 1000 Genomes Project [20]) with the 45 GWAS SNPs. We utilized the NIH Roadmap and ENCODE data and two mammalian conservation algorithms, GERP and SiPhy-omega, implemented in HaploReg v2 [26] (see URL) to examine if the 45 GWAS loci are enriched in regions of inferred strong enhancer chromatin states [27] in specific cell types. The background set for enhancer enrichment analysis was “All SNPs in 1000 Genomes phase I data”. For hierarchical clustering analysis, we mapped all 45 MI/CAD GWAS SNPs, along with SNPs in high LD (r2≥0. 8) (same SNP set in enrichment analysis), to each of the three histone marks (H3K27ac, H3K4me3 and H3K9ac), which are associated with active regulatory regions, in different cell types (Fig 3). Hierarchical clustering was performed using the heatmap function in R (R Project for Statistical Computing). Reordering of the rows and columns to produce the dendrogram was based on the presence or absence of a SNP residing in a histone mark in different cell types. We tested for direct connectivity of genes in GWAS loci in specific cell types. We tested 45 MI/CAD GWAS SNPs, in addition to SNPs in high LD (r2≥0. 8) (same SNP set in enrichment analysis), that overlapped with the three histone marks (H3K4me3, H3K9ac, H3K27ac) in a specific cell type. DAPPLE [51] was utilized to test for direct connectivity in PPI networks. Gene regulatory regions were defined as within 110 kb upstream of transcription start site and 40 kb downstream of transcription end site of each of the 45 lead SNPs or tag SNPs were included in the analysis. We performed 1,000 permutations to obtain empirical significance for the observed connectivity compared with the expected connectivity. NIH Roadmap Epigenomics Project, http: //www. roadmapepigenomics. org ENCODE Chromatin Core 15-State data, http: //egg2. wustl. edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/coreMarks/jointModel/final/ HaploReg v2, http: //www. broadinstitute. org/mammals/haploreg/haploreg. php

Coronary artery disease (CAD) and its subcomponent, myocardial infarction (MI), are the leading causes of infirmity and death worldwide. Large-scale genetic association studies have identified many genetic markers associated with CAD and MI. However, it has been difficult to determine the precise functional effects of these markers. Furthermore, it is unknown which cell types are biologically important in the development of MI/CAD. By intersecting findings from large-scale genetic association studies with functional genomic annotations, we show that genetic markers located in genomic regions that regulate expression of genes make up a large proportion of the genetic risk of MI/CAD. Furthermore, we show that this effect is particularly strong in certain tissues, including adipose, brain and spleen tissue. These results highlight the role of tissue-specific regulatory mechanisms in the genetic etiology of MI/CAD.

Abstract Introduction Results Discussion Materials and Methods

2015

Disproportionate Contributions of Select Genomic Compartments and Cell Types to Genetic Risk for Coronary Artery Disease

Human adenoviruses have been studied extensively in cell culture and have been a model for studies in molecular, cellular, and medical biology. However, much less is known about adenovirus replication and pathogenesis in vivo in a permissive host because of the lack of an adequate animal model. Presently, the most frequently used permissive immunocompetent animal model for human adenovirus infection is the Syrian hamster. Species C human adenoviruses replicate in these animals and cause pathology that is similar to that seen with humans. Here, we report findings with a new Syrian hamster strain in which the STAT2 gene was functionally knocked out by site-specific gene targeting. Adenovirus-infected STAT2 knockout hamsters demonstrated an accentuated pathology compared to the wild-type control animals, and the virus load in the organs of STAT2 knockout animals was 100- to 1000-fold higher than that in wild-type hamsters. Notably, the adaptive immune response to adenovirus is not adversely affected in STAT2 knockout hamsters, and surviving hamsters cleared the infection by 7 to 10 days post challenge. We show that the Type I interferon pathway is disrupted in these hamsters, revealing the critical role of interferon-stimulated genes in controlling adenovirus infection. This is the first study to report findings with a genetically modified Syrian hamster infected with a virus. Further, this is the first study to show that the Type I interferon pathway plays a role in inhibiting human adenovirus replication in a permissive animal model. Besides providing an insight into adenovirus infection in humans, our results are also interesting from the perspective of the animal model: STAT2 knockout Syrian hamster may also be an important animal model for studying other viral infections, including Ebola-, hanta-, and dengue viruses, where Type I interferon-mediated innate immunity prevents wild type hamsters from being effectively infected to be used as animal models. Human adenoviruses (Ads) are non-enveloped, double-stranded DNA viruses that are classified into over 60 types (Ad1, Ad2, etc.), which in turn are grouped into 7 species (A to G) (for a review on Ad biology, see [1,2]). Infection with the more frequent types is very common; about 40 to 60% of children are seropositive for Ad1,2, or 5 [3,4]. Ads cause a variety of diseases in humans; the symptoms range from respiratory to enteric, ocular and urinary, depending on the type of Ad. Generally, in immunocompetent adults the course of the infections is mild, and it resolves without the need for medical intervention (for a review of Ad epidemiology and pathology, see [3,5]). Infection with a specific type causes long-term immunity to that type. In certain circumstances, Ads can cause serious illness. Some types are more pathogenic than others: Ad8 and other types can cause epidemic keratoconjunctivitis (EKC), a disease that can result in lasting vision defects, even blindness. Other Ad diseases are more dependent of the health status of the host: in military recruits, Ad4,7, 3,21, and 14 cause serious acute respiratory disease that in some cases can lead to death. The most extreme case of the host’s influence on the illness caused by Ads is seen with immunocompromised patients. For these patients, the same types that cause self-resolving illness in healthy adults can cause serious, often life-threatening illness. The reason is that the dysfunctional immune system of these people cannot clear the primary virus infection, and thus it can develop into a much more severe systemic disease. In allogeneic hematopoietic stem cell transplant patients with a rising load of Ad in peripheral blood (as determined by quantitative PCR [qPCR]) despite antiviral therapy, the mortality can approach 100% [3,6, 7]. The host’s response to Ad infection has been extensively studied in cell culture and in vivo; the in vivo studies have been nearly always performed using non-permissive mice, usually with replication-defective Ad vectors. It was demonstrated that Ad infection induces vigorous innate immune response (for a recent review, see [8]). As part of the innate immune response, Ad infection results in the production of Type I interferons (IFNs). The production of IFNα and IFNβ by a variety of mononuclear cells was demonstrated in cell culture [9,10]. In recent studies in cell culture, cGAS/STING was shown to be the sensor for Ad DNA in the cytoplasm leading to the IFN response [11,12]. With this sensor, upon DNA binding, cyclic GMP-AMP synthetase synthesizes cyclic guanine-adenine dinucleotide (cGAMP). cGAMP binds to the STING adaptor protein, which competes with TANK-binding kinase leading to phosphorylation of IFN response factor 3 (IRF3). IRF3 translocates to the nucleus to induce transcription of IFN and IFN-stimulated genes (ISGs). These Ad studies were conducted with murine endothelial and Raw264. 7 macrophage lines infected with replication-defective Ad vectors [12] and with various permissive human cell lines infected with wild-type (wt) Ad [11]. It is not known whether Syrian hamsters have this DNA sensor. In vivo in mice, myeloid dendritic cells in the spleen were shown to be the most prolific producers of Type I IFNs [9]. In these cells, the induction of IFNα and IFNβ is mediated by a hitherto unknown cytosolic sensor [9]. This seems to be unique, because plasmacytoid dendritic cells are reported to produce most of Type I IFNs during other virus infections, and the induction is mediated by a TLR9-dependent pathway, signaling via IRF7 [13]. In addition, different types of Ad seem to have different potential to induce a Type I IFN response [14], and this differential induction can lead to differential activation of NK cells [15]. These discrepancies are confounded by the fact, as mentioned, that the mouse is not an adequate model for studying human Ads, inasmuch as human Ads replicate poorly in mice. Thus, immune responses initiated by actual virus replication cannot be studied in mice. Our laboratory and others have pioneered the use of Syrian hamsters as an animal model to study the replication and pathogenesis of human species C Ads in vivo, as well as a model system to investigate oncolytic Ad vectors. Species C human Ads replicate in most organs of Syrian hamsters after intranasal or intravenous injection, and cause pathology that is similar to that seen with humans [16–29]. For immunocompetent hamsters, the immune response clears Ad infections by approximately 1 week post challenge, and without exception the animals recover from the infection [28,29]. Hamsters immunosuppressed by high doses of cyclophosphamide are unable to eliminate the infection, and develop enhanced pathology, mirroring the pathology caused by Ad infections of immunocompromised patients [30]. We have shown that the immunosuppressed Syrian hamster model can be used to test the efficacy of anti-adenoviral drugs in vivo [30–33]. In the latter studies we have shown that brincidofovir (formerly named CMX001), cidofovir, ganciclovir and valganciclovir are very effective both prophylactically and therapeutically in treating disseminated Ad5 infections. With several studies mentioned above, cyclophosphamide was used to immunosuppress the animals. However, there are some significant drawbacks in using cyclophosphamide, chief among them is that it exerts pleiotropic effects. Recently, a Syrian hamster strain with the STAT2 gene inactivated was generated with CRISPR/Cas9-mediated site-specific gene targeting, and the authors showed that STAT2 knockout (STAT2 KO) hamsters do not express the full length STAT2 protein [34]. STAT2 is a crucial element of the Type I and Type III IFN signal transduction pathway (reviewed in [35,36]). STAT2-null mice are defective in inducing ISGF3 target genes and are more susceptible to viral infections [37]. Here, we demonstrate that the Type I IFN pathway is disrupted in STAT2 KO hamsters, which makes these animals extremely sensitive to intravenous infection with Ad5. At 3 days post challenge, the virus burden in the organs of STAT2 KO animals is 1000-fold higher than in wt hamsters. Our results underscore the importance of innate responses in fighting Ad infections, and may provide relevant information to the management of Ad infections in immunocompromised patients. Sixteen STAT2 KO and 16 wt hamsters were injected with Ad5 as described in the Methods section. Four animals of each strain were scheduled to be sacrificed at 1 and 3 days post infection (p. i.), and 8 hamsters were scheduled to be sacrificed at 7 days p. i. However, 6 of the 8 Day 7 STAT2 KO hamsters were moribund at around 4 days p. i. (Fig 1A). These animals were sacrificed, and for some subsequent analyses their data were grouped with those of animals sacrificed according to schedule at 3 days p. i. In figures depicting such combined results, data obtained from moribund animals is discernible from data gathered during scheduled sacrifice. Altogether, there were 6 treatment-related deaths in this study, all in the Ad5-infected STAT2 KO group (Fig 1A). For the animals killed at the scheduled time point at 3 days p. i. , gross pathology revealed that 2 of the 4 Ad5-injected STAT2 KO hamsters had pale, mottled liver, while no significant findings were noted for the other two hamsters. The hamsters that were sacrificed moribund (all STAT2 KO) presented pathology characteristic of advanced Ad infection (yellow, mottled liver, enlarged gall bladders). No pathology was observed with the wt hamsters at any sacrifice time point. Histopathological analysis of the liver confirmed that the peak of pathology was at 3 to 4 days post challenge (Table 1 and Fig 1C). Ad5-infected STAT2 KO hamsters sacrificed according to schedule at this time showed moderate multifocal mononuclear infiltration, multifocal hepatocellular necrosis of minimal to mild severity and, marked diffuse decreased vacuolation. The pathology was even more obvious with the hamsters that were sacrificed moribund at 3 and 4 days post challenge: these animals presented with moderate to marked hepatocellular necrosis with nuclear inclusion bodies, which is characteristic of Ad5 infections of the liver [28,38]. At 3 days post challenge, the only significant findings for the Ad5-infected wt hamsters were mild to moderate decreased vacuolation and minimal to mild mononuclear infiltration (Table 1 and Fig 1C). At 7 days post challenge, minimal mononuclear infiltration and marked decreased vacuolation was reported for the surviving STAT2 KO hamsters, while no significant microscopic lesions were found in the liver of wt animals (Table 1). A STAT2 KO animal that was sacrificed moribund at 6 days post challenge had lesions of similar nature and severity as the animals sacrificed moribund at 3 to 4 days post challenge. Consistent with these observations, the Ad5-infected STAT2 KO animals had significantly higher serum alanine transaminase (ALT) levels than wt ones at 3 days post challenge (Fig 1B). By 7 days post challenge, ALT levels returned to normal levels in all surviving animals. Liver samples were collected at necropsy and were analyzed for infectious virus burden using a cell culture assay. At 3 days p. i. , STAT2 KO hamsters had 100- to 1000-fold higher virus load in their liver than wt animals (Fig 2A). By Day 7, all surviving animals had undetectable to unquantifiable levels of virus load in the liver. To ascertain whether this elevation in virus load at 3 days p. i. was the result (of enhanced virus replication or delayed clearance of the released progeny virus, we determined the relative number of late adenoviral transcripts in the isolated liver samples as a surrogate for ongoing Ad replication. Ad late transcripts were measured by RT-qPCR using primers that are specific to the tripartite leader present in most mRNAs in the major late transcription unit. We found that there were a 100- to 1000-fold more late viral mRNA copies in samples collected from STAT2 KO animals than in samples from wt ones (Fig 2B), indicating that the increased virus load was the result of increased virus replication in the liver of STAT2 KO hamsters. Immunohistochemical staining for Ad fiber, a late Ad protein, corroborated the findings depicted in Fig 1A and 1B, inasmuch there were markedly more cells staining for the Ad fiber protein in the liver of STAT2 KO hamsters than in the liver of wt animals at 3 days p. i. (Fig 2C). The multifocal distribution of hepatocytes staining for Ad fiber (Fig 2C) coincides with the multifocal distribution of hepatocellular necrosis found during histopathological examination (Fig 1C). This finding suggests that Ad5 is the causative agent of the pathology observed in these animals. Most remarkably, when we analyzed the lung and the kidney for infectious virus load, we found that the these organs from STAT2 KO animals had virus load commensurate to the liver, while only a very low virus load was detectable in the lung and kidney of wt hamsters (Fig 3). If the Type I IFN response is impaired in the STAT2 KO hamsters, then we would expect that transcription of IFN-stimulated genes (ISGs) would be impaired. RT-qPCR was utilized to detect changes in the expression levels of ISGs in the liver, namely protein kinase R (double-stranded RNA-dependent protein kinase, PKR), oligoadenylate synthetase (OAS), and IFN-induced GTP-binding protein (Mx2). Baseline levels of PKR and Mx2 were significantly elevated in the livers of wt hamsters compared to the STAT2 KO animals (Fig 4A and 4C). The difference was much more striking after Ad5 infection: for the wt hamsters’ transcripts, all three ISGs assayed were extremely elevated at 1 and 3 days post challenge, while no such elevation was observed for the STAT2 KO animals (Fig 4A–4C). The induction of ISGs was similarly impeded in the spleen (Fig 4D–4F). To assess whether disruption of the Type I IFN pathway contributes to increased Ad replication in cell culture, we prepared primary kidney cultures of wt and STAT2 KO hamsters. Primary kidney cells were used because Ads replicate in hamster kidneys in vivo, thus these cells of epithelial origin are expected to support Ad replication in vitro. Further, established protocols exist for preparing such cells from harvested hamster kidneys. We infected both kidney cultures with either a mutant Ad named dl331 that is deleted for the VA RNA I sequence or with the parental Ad (dl309) in which the VA RNA I gene is intact. VA RNA I is a small Ad-coded RNA that counteracts the effect of PKR in Ad-infected cells [39], and it is required for the efficient translation of viral RNAs at the late stage of infection [40]. In human cells, the deletion mutant dl331 is very susceptible to IFN-mediated inhibition of virus replication, while dl309 is somewhat resistant [39]. When Ad-infected wt hamster kidney cells were treated with human IFNα, which is known to be active in Syrian hamsters [41], the yields of dl331 decreased approximately 100-fold, while there was a more modest but detectable decrease in the yields of dl309 (Fig 5A). However, no reduction in the yields of either virus was seen in IFNα-treated STAT2 KO kidney cells (Fig 5A). When we tested for the induction of ISGs in human IFNα-treated primary kidney cells, we found that PKR was efficiently induced in wt but not in STAT2 KO cells (Fig 5B), confirming that human IFNα activates the Type I IFN pathway in hamster cells. These data indicate that the disrupted Type I IFN pathway of the infected cells contributes to the increased Ad replication in the STAT2 KO animals. The data also indicate that the VA RNA I of human Ad5 is functional in primary hamster kidney cells. Besides affecting the antiviral state of the infected cell, Type I IFNs can have an indirect antiviral effect by modulating the functions of other immune cells [42]. One such cell population is natural killer (NK) cells, the cytotoxic function of which is increased by Type I IFN [43]. At 3 days post challenge, infiltration of the liver by NK cells was significantly reduced in STAT2 KO hamsters compared to wt animals, as judged by the changes in the abundance of mRNAs for KIRL1. 1, an NK cell marker (Fig 6A). The decrease in KIRL1. 1 expression was particularly marked in animals sacrificed moribund. We also examined if obstructing the Type I IFN signaling pathway influences the expression of IFNα, inasmuch as it is known that the expression of IFNα depends on a virtuous cycle maintained by paracrine stimulation [37]. We found that there was a significant delay in the increase of IFNα serum concentration with the STAT2 KO hamsters. The baseline concentration of IFNα in the serum was significantly lower in these hamsters and did not reach the concentration of IFNα seen in the serum of wt hamsters until 3 days post challenge (Fig 6B). There was no statistically significant difference between the maximal IFNα concentrations reached in the two stains of hamsters (Fig 6B). Neutralizing antibody (NAb) levels were determined in serum collected at sacrifice. For both STAT2 KO and wt hamsters, NAb levels were elevated in response to Ad infection at 3 days post challenge (Fig 7A). However, for the STAT2 KO hamsters the NAb titers kept increasing after this time, while the NAb levels for wt hamsters reached a plateau by this time (Fig 7A). This finding was corroborated by data from another experiment, in which approximately 10-fold more NAb was raised in STAT2 KO hamsters by 7 days post challenge than in wt ones (Fig 7B). When testing the liver for infiltrating T lymphocytes by immunohistochemical staining for CD3, we found that the baseline T-cell numbers in the liver were similar for both hamster strains; however, at 1 day post challenge, T-cell numbers in the liver dropped with wt hamsters, while perivascular infiltration had started with STAT2 KO hamsters by this time (Fig 8A). At 3 and 7 days post challenge, T-cell infiltration in STAT2 KO hamsters was similar to that in wt animals (Fig 8A). We determined by flow cytometry the number and ratio of CD8+ (Fig 8B) and CD4+ (Fig 8C) T-cells infiltrating the liver, and found that in vehicle-treated hamsters there was no difference between the two hamster strains. At 1 day post challenge, there was a decrease in the number of both CD8+ and CD4+ T-cells in the liver of wt animals, while no such decrease was seen in the liver of STAT2 KO animals (Fig 8B and 8C). This finding corroborates the observation that the number of CD3+ cells declined in the liver of wt hamsters at 1 day post challenge (Fig 8A). A similar decrease in both CD8+ and CD4+ T-cell numbers was observed in the spleen of wt hamsters (Fig 8D and 8E). At 3 days post challenge, there was an increase in the number of both CD8+ and CD4+ T-cells in the liver of both strains of hamsters; however, there was a significantly larger number of CD8+ cells in the liver of wt animals compared to STAT2 KO hamsters (Fig 8B). Similar observations were made for the spleen (Fig 8D and 8E). At 7 days post challenge, the number of both populations of T-cells declined in the liver and spleen (Fig 8) of both hamster strains. To test whether there was any difference between the number of Th1 and Th2 lymphocytes infiltrating the liver, we determined the changes of IFNγ and IL-4 transcripts. After challenge, IFNγ mRNA levels were elevated in the liver of both strains of hamsters; peaking at 3 days post challenge, and starting to decline at 7 days post challenge (Fig 9A). Samples collected from moribund STAT2 KO animals had significantly lower levels of IFNγ mRNA levels in their liver at 3 days p. i. (P = 0. 0159; compare the open and closed square symbols) (Fig 9A). Conversely, a slight decrease over time after challenge was seen in IL-4 mRNA levels in the liver, with no significant differences between the two hamster strains (Fig 9B). In the spleen, baseline IFNγ mRNA levels were significantly higher in STAT2 KO hamsters than in wt ones, and reached significantly higher peak values at 3 days post challenge (Fig 9C). No significant changes were seen in IL-4 mRNA levels in the spleen (Fig 9D). We have also determined the expression levels for the Class I major histocompatibility complex (MHC-I) mRNA in the liver. We found that at 3 days post challenge, there was a small but statistically significant decrease in the abundance of MHC-I mRNA in the liver of STAT2 KO hamsters compared to the wt ones (Fig 9E). The immune response to human Ad infections has been studied extensively using mice, and valuable information was collected as to the response to the early phases of viral infection, i. e. virus attachment, entry, endosomal escape, etc. [8,44,45]. However, the experiments conducted in mice could not investigate the immune response induced by events in the late phase of infection, i. e. DNA replication, expression of late proteins, cell death and lysis, because human Ads replicate very poorly in mice. Our work is the first of its kind, using a permissive animal model to investigate the effect of Type I IFNs on the replication of, and the pathology induced by human Ads. Ads, like other viruses, are believed to employ various mechanisms to thwart the effects of IFNs as indicated by studies in cell culture [1,5]. It was demonstrated that the Ad VA RNA I effectively inhibits the action of PKR in cell culture [39], thus abrogating the antiviral effect of IFNα. As anticipated, we have shown that the replication of a mutant Ad deleted for the VA RNA I gene is inhibited in IFNα-treated primary hamster kidney cells, while there was only a modest suppression of the replication of a wt Ad by IFNα (Fig 5). Ad has other mechanisms to suppress the function of Type I IFNs. E1A expression inhibits the expression of ISGs by suppressing the transcriptional co-activator p300 [46] and by suppressing the Jak-STAT signal transduction pathway [36,47–50]. In Ad-infected cells, E1A was reported to overcome the IFN-mediated inhibition of replication of vesicular stomatitis virus [51], the IFN-induced resistance to NK cell lysis [52], and to suppress Jak-STAT signaling [53,54]. The Ad E1B-55K protein functions to inhibit the transcription of some ISGs as shown by studies in cells infected with Ad E1B-55K mutant viruses [55,56]. E1B-55K expressed alone did not have this inhibitory function. In studies with Ad mutants that do not synthesize the E4orf3 protein, IFN blocked viral DNA replication and early gene expression [57,58]. The E4orf3 protein apparently exerts its anti-IFN function by causing the components of promyelocytic leukemia protein (PML) nuclear bodies (PML-NB) to rearrange into nuclear tracks [57]. PML-NBs, which contain a large number of cellular proteins, some of which are induced by IFN, function in multiple cellular responses including antiviral defense [59]. It is interesting and somewhat puzzling that in the studies discussed above, in which an Ad mutant lacks only one of the anti-IFN gene products, namely VA RNA I, E1A, E1B-55K, or E4orf3 but expresses the other gene products, treatment of infected cells by IFN has such strong anti-Ad effects. Many of these studies were conducted in cancer cell lines, some of which may have defects in IFN-induced effects. However, we observed this same phenomenon in our studies with dl331 (lacks VA RNA I) in primary hamster kidney cells. Considering that Ad5 has at least 4 functions that are reported to counter IFN antiviral effects, it is somewhat surprising that we found that control of Ad5 replication by Type I IFNs is of such high importance in vivo in the Syrian hamster model. In the liver of STAT2 KO hamsters, human Ad5 replicated to a 100- to 1000-fold higher titer compared to their wt counterparts (Fig 2), and the virus load was also much higher in the lung and kidney of the STAT2 KO animals (Fig 3). As might be expected, this high virus load resulted in higher morbidity and mortality (Fig 1A). As to the reasons for this elevated Ad replication, we have shown that the Type I IFN signal transduction pathway is not functional in STAT2 KO hamsters; these animals cannot up-regulate the expression of ISGs upon virus infection (Fig 4). In the infected cells, this failure to up-regulate the expression of ISGs will most likely lead to increased virus replication. Further, in the STAT2 KO animals, the immunomodulatory function of Type I IFNs may be impaired as well. At 3 days post challenge, we found less infiltrating NK cells in the livers of Ad5-infected STAT2 KO hamsters than in the livers of wt ones (Fig 6A), though the levels of IFNγ transcripts, one of NK-cells’ main antiviral product, were similar in the livers two strains of hamsters (Fig 9). Another interesting finding pertains to the efficacy with which all surviving STAT2 KO animals cleared the virus. This, the abundance of neutralizing antibodies (Fig 7) and infiltrating CD3+ cells (Fig 8) in the liver indicate that the adaptive immune responses are intact. This is somewhat at variance with results presented by Zhu et al. that Type I IFN signaling on both B cells and CD4 T cells is critical for eliciting a neutralizing antibody response to replication-defective Ads in mice [60]. A possible explanation for this discrepancy may be that the adaptive immune response in a permissive model is probably stimulated by more factors, like the inflammation caused by the lysis of infected cells and the abundance of viral antigens than in a non-permissive one. The abundance of viral antigens (100- to 1000-fold higher virus load than wt hamsters in the livers and probably other organs) may be a factor in the very high neutralizing antibody levels seen with STAT2 KO hamsters. Although more CD8+ T cells infiltrated the liver of Ad5 infected wt hamsters than STAT2 KO animals (Fig 8), the IFNγ and IL-4 profile of the liver-infiltrating lymphocytes was identical for wt and STAT2 KO animals (Fig 9), thus it is not likely that a Th2-type skewing of the adaptive immune response is responsible for the elevated neutralizing antibody responses in STAT2 KO hamsters. The slight elevation of MHC-I mRNA levels in the liver of wt hamsters over their STAT2 KO counterparts (Fig 9E) might have resulted in better retention of infiltrating CD8+ T lymphocytes and thus might have contributed to the larger number of CD8+ cells seen in the liver of wt animals. It was reported earlier that intravenous injection of large doses of Ads can lead to leukopenia both in humans, non human primates, and mice [61–63]. We have seen a similar phenomenon in wt hamsters: at 1 day post challenge, the number of CD3+ cells declined in the liver, as did the number of CD4+ and CD8+ cells in the liver and spleen (Fig 8). However, no similar decline in leukocyte numbers at 1 day p. i. was seen with STAT2 KO hamsters (Fig 8). This difference between wt and STAT2 KO animals suggests that the Type I IFN production in response to Ads causes leukopenia. This conclusion is supported by findings that Type I IFN treatment of mice and hepatitis C patients causes leukopenia [64–67]; thus, a similar mechanism may be at work during Ad infection. By 3 days p. i. , the number CD3+ lymphocytes rebounded with wt hamsters, and the number of CD8+ T cells infiltrating the liver, as well as the number of CD4+ and CD8+ T cells in the spleen were elevated compared to that of vehicle-treated hamsters (Fig 8). At 3 days p. i. , a similar increase in the number of these lymphocyte subsets was observed with the STAT2 KO animals, albeit to a lesser degree (Fig 8). However, this difference does not result in a defect in clearing the virus infection (Fig 2). This is the first time that data regarding the immune response with a STAT2 KO Syrian hamster, or indeed any kind of gene modified hamster, has been made available. We have verified that the Type I IFN pathway is disrupted with these animals, as evidenced by their inability to increase the expression of the ISGs PKR, OAS, and Mx2. We have characterized the immune response to Ad infection in these hamsters, and described the resulting pathology. We believe that this hamster strain is a valuable resource for researchers using Syrian hamsters as an animal model, and that our data will help investigators in their research. Further, our data highlight the importance of the Type I IFN response in controlling systemic Ad infections, and as such our results may have some clinical importance. A549 human lung adenocarcinoma cells were purchased from the American Type Culture Collection (ATCC) (Manassas, VA), while HEK293 human embryonic kidney cells were purchased from Microbix (Mississauga, Ontario, Canada). Both cell lines were cultured in Dulbecco’s modified Eagle’s medium (Sigma-Aldrich, St Louis, MO, USA) with 10% fetal bovine serum (FBS) (complete DMEM) at 37°C. Ad5 wt500, a wt human Ad5 isolate, was derived in our laboratory by plaque purification from an Ad5 stock purchased from ATCC. The dl309 virus [68] was obtained from Elizabeth Moran; dl331 [69] was obtained from Tom Shenk. dl331 contains a deletion that blocks expression of VA RNA I, while dl309 is considered to be the phenotypically wild-type parental virus for dl331. The titer of the viruses was determined by plaque assay. The hamster strain (STAT2 KO) homozygous for a +1 frameshift mutation in the N-terminal domain of the STAT2 gene was reported previously [34]; wt hamsters were purchased from the same supplier from which the parental hamsters of the STAT2- strain originated (Charles River Laboratories, Wilmington, MA). With a genomic PCR assay developed in house, both hamster strains were found to carry a latent hamster polyoma virus infection [70]. In some of the STAT2 KO animals this infection resulted in tumors (hemangiomas and lymphomas). All animals with lymphomas and ones with large hemangiomas were excluded from our study. All studies were approved by the Institutional Animal Care and Use Committee of Saint Louis University and were conducted according to federal and institutional regulations. Two groups of hamsters were established; one with STAT2 KO animals (10 males and 14 females) and the other with wt ones (females only). There were 24 hamsters in both groups. For the STAT2 KO animals, male and female hamsters were equally distributed to all treatment groups. The animals were anesthetized with a ketamine/xylazine mixture, and PBS or Ad5 was injected i. v. (via the jugular vein). Half of the animals in both the STAT2 KO and wt groups received vehicle (PBS), while the remaining hamsters were injected with 2x1011 plaque forming units (PFU) /kg of Ad5. Four Ad5-infected and 4 vehicle-treated hamsters from both groups were sacrificed at 1,3 and 7 days after challenge. The body weights and signs of morbidity of the animals were recorded daily. Hamsters that became moribund before Day 7 were sacrificed as needed. Besides animals judged moribund by observation, we sacrificed all hamsters that lost more than 20% of their original body weight. By Day 4 post challenge, we lost all the Ad5-infected STAT2 KO animals scheduled to be sacrificed at 7 days post challenge. To compensate for this loss we altered the design of the study, and injected the remaining 4 wt and 4 STAT2 KO vehicle control animals (no procedures were performed on these hamsters at this point) with 1. 6x1011 PFU/kg of Ad5 with the intent of sacrificing these hamsters at Day 7. However, 2 of the 4 STAT2 KO hamsters died at Day 4, thus, we had only two animals in this group for the Day 7 sacrifice time point. Data obtained from animals sacrificed moribund at Days 3 and 4 were grouped with the data collected from animals sacrificed according to schedule at 3 days post challenge, while data collected from a single hamster sacrificed moribund at 6 days post challenge were grouped with data from the group sacrificed at 7 days post challenge. All data collected from moribund animals is clearly marked when used. At necropsy, the animals were bled out and liver, lung, and kidney was collected. Virus was extracted from the liver and was quantified by the 50% tissue culture infectious dose (TCID50) assay in HEK293 cells as described previously [30]. A portion of the collected tissues was preserved in formalin and processed for histopathology (Seventh Wave Laboratories, St. Louis, MO). Immunohistochemical staining was performed by the Histopathology and Tissue Shared Resource of Georgetown University, using 1: 1000 dilution of the Adenovirus Ab-4 (4D2) (Lab Vision, Fremont, CA) and 1: 200 dilution of the CD3-ε (M-20) (Santa Cruz Biotechnology, Santa Cruz, CA) antibodies to stain for the Ad fiber and hamster CD3 proteins, respectively. Serum was assayed for alanine transaminase levels (Advanced Veterinary laboratory, St. Louis, MO). The serum concentration of hamster IFNα was determined using a hamster IFNα ELISA kit (MyBiosource, San Diego, CA), according to the manufacturer’s instructions. Total RNA from liver and spleen was extracted from each hamster by homogenizing a fraction of collected tissues in RNALater buffer (Qiagen, Valencia, CA) and then extracting the RNA using the RNeasy mini kit (Qiagen). All RNA samples were treated with RNase-free DNase followed by RNA cleanup to eliminate DNA contamination. The RNA yield was determined on a NanoDrop-2000 spectrophotometer. For RT-qPCR, two micrograms of each RNA and 50 pM of oligo (dT) primer were used for in vitro reverse transcription (RT) using High Capacity cDNA Reverse Transcription kit (ABI, Forster city, CA). SYBR-green based qPCR was used to specifically detect target gene mRNA. Primer sequences for PKR, Mx2, IFN-γ, and IL-4 were described previously [71]. The primers for OAS-3 (F, 5’-AGGTGCTTAAGGTGGTTAAGGG-3’; R, 5’-TGCTCAGAGAAGTGCTGGAAG-3’), ribosomal protein S6 kinase polypeptide1 (RPS6KB1) (F, 5’-TCAGACCGGTGGAAAACTCTAC-3’; R, 5’-TGATGCAAATGCCCCAAAGC-3’), and KiRL1. 1 (F, 5’-CCTTTACTCTTGCTGCTATGC-3’; R, 5’-TTTGACTCTTGATCCCTGTTG-3’) were designed using Primer3 software (Whitehead Institute for Biomedical Research, Cambridge, MA) and synthesized by Integrated DNA Technologies (Coralville, IA). The PCR was set up in a 20 μl volume containing 1x SYBR select master mix (ABI), 250 nM forward and reverse primers, and 3 μl of the diluted RT template. Quantification was done in triplicate for each sample using an ABI model 7500 genetic analyzer with the following cycling parameters: 1 cycle at 50°C for 2 min, 1 cycle at 95°C for 10 min, followed by 40 cycles of 95°C for 15 sec and 60°C for 1 min. Dissociation curve analysis was performed following PCR to verify the specificity of the amplified product. The data were analyzed using the ΔΔCt method. Housekeeping gene RPS6KB1 was used as an endogenous control for normalization. Briefly, the Ct of each target gene in a treated hamster was first normalized to the Ct of the endogenous control (ΔCt) and then compared to the same normalized gene in a mock-treated (calibrator) hamster to determine the ΔΔCt. The final value is displayed as the relative fold change between the Ad5-infected and mock-treated hamsters. To generate primary hamster kidney cells, kidneys were collected from wild-type and STAT-2 KO hamsters. Kidney tissue was minced and trypsinized. The material was passed through a mesh sieve to give a cell suspension. Kidney cells were centrifuged and plated onto cell culture dishes in DMEM containing 10% FBS. After 4 days, cells were trypsinized and plated into 12-well plates at 8x104 cells per well. After an additional 5 days, cells were left untreated or treated with 250 or 1000 U/ml of human IFNα (PBL Assay Science; cat. #11105–1) for 24 h, and then infected with 25 PFU/cell of dl309 or dl331. Cell numbers at infection were 1. 29x105 cells/well (wild-type kidney cells) or 8. 39x104 (STAT-2 KO kidney cells). Infections were done in duplicate for each IFNα concentration; cells remained in IFNα during and after infection. Monolayers were washed at 24 h p. i. and 1 ml DMEM/10% FBS was added per well. Cells were freeze-thawed and harvested. Samples were sonicated for 4 min. Infectious virus yield was determined by TCID50 assay on HEK 293 cells. Liver and spleen were collected into complete DMEM and stored on ice until processing. Single cell suspensions were made by forcing the organs through a 100 μm cell strainer (Fisher Scientific, Waltham, MA) using the blunt end of a syringe plunger. The cell suspensions were washed in PBS. A leukocyte-enriched preparation of the liver homogenate was produced as described before [72]. Briefly, an equal volume of the liver suspension was layered onto room temperature Ficoll-Paque Plus gradient (GE Healthcare, Little Chalfont, United Kingdom) and centrifuged for 30 min with no brake. The supernatant enriched in white blood cells was collected, diluted with PBS 1: 1, pelleted, and washed in PBS. The cell pellets were re-suspended in a final volume of 300 μl of PBS. For the spleen, single cell suspensions were pelleted as described above, and re-suspended in 7 ml Pharm Lyse buffer (Beckton Dickinson, Franklin Lakes, NJ). The samples were incubated for 15 min at room temperature and then diluted 1: 1 with PBS containing 2% fetal bovine serum (FACS buffer). The samples were then washed in DPBS, and the cell pellets were re-suspended in a final volume of 300 μl PBS. All cell suspensions were transferred to 96 well tissue culture plates, and kept on ice until staining. For staining, the plates centrifuged at 1200 rpm for 5 min, and the supernatants were discarded. The cells were incubated with Live/Dead Fixable Aqua (Life Technologies, Grand Island, NY) cell stain for 30 min in the dark and on ice. The cells were washed twice with FACS buffer and then re-suspended in FACS buffer containing fluorochrome-conjugated antibodies. Cells were stained with a 1: 2000 dilution of PE conjugated anti-mouse/rat MHC class II (I-Ek; clone 14-4-4S), a 1: 200 dilution of FITC conjugated anti-rat CD8b (clone 341) and a 1: 200 dilution of A700 conjugated anti-mouse CD4 (clone L3T4) (all from E-Bioscience, San Diego, CA). After 1 h incubation, the plates were washed twice with PBS, and the cell pellets re-suspended in 2% paraformaldehyde solution. After 5 min incubation, the cells were washed in PBS and re-suspended in FACS buffer. The samples were stored at 4°C for less than 24 h before being analyzed on a BD Biosciences LSR II. The data was acquired using FACSDiva software (BD) on the LSR II, and the data analysis was done at a separate computer using FlowJo. Serum samples were incubated for 30 min at 56°C to inactivate complement. Serum samples (in four replicate wells) were diluted twofold in a 96-well plate in complete DMEM. Ad5 was added to the dilutions of sera at 100 PFU per well, and after 1 h incubation at 37°C with the virus, A549 cells were added to each well at 5x105 cells per plate. The plates were incubated at 37°C for 10 days, after which live cells were stained with neutral red (30 mg/ml in PBS). The cell-bound dye was extracted with 100 ml of acidified ethanol solution (50% ethanol and 1% acetic acid in H2O), and the absorbance was measured at 550 nm on a Synergy 4 microplate reader (BioTek, Winooski, VT, USA). The NAb titer was calculated as the reciprocal dilution causing 50% inhibition of viral cytopathic effect. Statistical analysis was performed using GraphPad Prism (version 4) software (GraphPad Software, La Jolla, CA). The overall effect was calculated using the Kruskal-Wallis test, and comparison between groups was performed using the Mann-Whitney U test. P values of 0. 05 were considered significant. All animal studies were approved by the Institutional Animal Care and Use Committee of Saint Louis University (protocol# 2015). The studies were conducted according to regulations in the Animal Welfare Act, the PHS Policy on Humane Care and Use of Laboratory Animals, and according to the recommendations of the Guide for the Care and Use of Laboratory Animals.

The biology of human adenoviruses has been studied extensively; however, much less is known about the replication and pathogenesis of the virus in a permissive host. Our laboratory pioneered the use of Syrian hamsters to study the pathogenesis of human adenoviruses. Syrian hamsters are permissive for species C human adenoviruses, which replicate in these animals and cause illness akin to that in humans. Hereby, we report findings with a new Syrian hamster strain (STAT2 KO hamsters), in which the Type I interferon pathway, an important part of the innate immune response to virus infection, is disrupted. This is the first genetically modified Syrian hamster strain ever reported. We show that these animals are very sensitive to infection with type 5 human adenovirus (Ad5). Ad5 replicates to 100- to 1000-fold higher titers in STAT2 KO hamsters than in wild-type ones, and this increased infection causes enhanced pathology. However, the adaptive immune response to the virus infection seems to be intact with the STAT2 KO hamsters, and surviving animals clear the virus effectively. The data reported here may be of interest to researchers focusing on adenoviruses, and also to those who utilize the Syrian hamster as their animal model for other purposes.

Abstract Introduction Results Discussion Materials and Methods

2015

STAT2 Knockout Syrian Hamsters Support Enhanced Replication and Pathogenicity of Human Adenovirus, Revealing an Important Role of Type I Interferon Response in Viral Control

Clonally derived bacterial populations exhibit significant genotypic and phenotypic diversity that contribute to fitness in rapidly changing environments. Here, we show that serial passage of Salmonella enterica serovar Typhimurium LT2 (StLT2) in broth, or within a mouse host, results in selection of an evolved population that inhibits the growth of ancestral cells by direct contact. Cells within each evolved population gain the ability to express and deploy a cryptic “orphan” toxin encoded within the rearrangement hotspot (rhs) locus. The Rhs orphan toxin is encoded by a gene fragment located downstream of the “main” rhs gene in the ancestral strain StLT2. The Rhs orphan coding sequence is linked to an immunity gene, which encodes an immunity protein that specifically blocks Rhs orphan toxin activity. Expression of the Rhs orphan immunity protein protects ancestral cells from the evolved lineages, indicating that orphan toxin activity is responsible for the observed growth inhibition. Because the Rhs orphan toxin is encoded by a fragmented reading frame, it lacks translation initiation and protein export signals. We provide evidence that evolved cells undergo recombination between the main rhs gene and the rhs orphan toxin gene fragment, yielding a fusion that enables expression and delivery of the orphan toxin. In this manner, rhs locus rearrangement provides a selective advantage to a subpopulation of cells. These observations suggest that rhs genes play important roles in intra-species competition and bacterial evolution. Bacteria often reside in complex communities such as biofilms in which cells from multiple species touch one another in a three-dimensional network [1]. These environments provide opportunities for cellular interactions, yet the mechanisms underlying contact-dependent competition and cooperation have been largely unexplored until recently. A diverse family of YD-peptide repeat proteins mediates at least two distinct forms of contact-dependent competition in Gram-negative and -positive bacteria [2]. The Rhs (rearrangement hotspot) proteins of Gram-negative enterobacteria [3], [4] are large (∼1,400–1,700 residues) toxic effectors that appear to be exported through the type VI secretion machinery. Related WapA (wall-associated protein A) proteins from Gram-positive bacteria are somewhat larger (∼2,200–3,600 residues) [5] and are likely exported through the general secretory pathway [2]. Rhs and WapA proteins are both characterized by sequence-diverse C-terminal regions (Rhs-CT and WapA-CT) that vary considerably between different strains of the same species. Analysis of several Rhs-CTs and WapA-CTs from Dickeya dadantii 3937 and Bacillus subtilis subspecies revealed that these domains contain the toxin activities responsible for intercellular growth inhibition. All rhs and wapA genes are closely linked to small downstream open reading frames that encode RhsI and WapI immunity proteins, respectively. These immunity proteins are also sequence-diverse and only protect against their cognate Rhs-CT (or WapA-CT) toxins. Thus, Rhs and WapA represent related, yet distinct, delivery platforms for polymorphic toxin domains [2]. Because different strains typically express unique rhs-CT/rhsI (wapA-CT/wapI) alleles, these systems collectively form a complex network of toxin/immunity pairs that are thought to mediate inter-strain competition for environmental resources [2]. The rhs loci of Enterobacteriacae often contain one or more additional rhs-CT/rhsI gene pairs located downstream of the main rhs/rhsI pair. These modules have been termed “orphan” toxin/immunity pairs, because the rhs-CT coding sequences resemble displaced fragments from full-length rhs genes [6]. Orphan rhs-CT genes often contain some coding sequence for portions of the conserved N-terminal regions, but orphan fragments are much smaller than full rhs genes and usually lack translation initiation signals. Therefore, it is unclear whether orphan rhs-CT genes are expressed, raising the question of whether these auxiliary elements are functional. Here, we show that repeated passage of Salmonella enterica serovar Typhimurium LT2 (StLT2) produces “evolved” lineages that deploy the orphan Rhs-CT toxin to inhibit the growth of ancestral cells. We provide evidence that the rhs locus undergoes rearrangement to fuse the rhsmain and rhs-CTorphan genes, thereby providing a mechanism to express and export the Rhs-CTorphan toxin domain. These results indicate that rhs rearrangement provides a selective advantage to a subpopulation of cells, suggesting that rhs plays an important role in clonal selection and bacterial evolution. In an effort to isolate StLT2 strains with increased fitness, we serially passaged cells for ∼1,000 generations in LB medium [7]. Analysis of six independently evolved cultures revealed that each lineage outcompeted ancestral StLT2 cells in co-culture experiments (Figures 1A & S1A). Remarkably, we observed the same competitive advantage in four of eight StLT2 lineages that were obtained by passage through multiple mouse hosts [8] (Figures 1A & S1B). This competitive advantage was not due to faster growth rate, because four of the evolved lineages grew more slowly than the ancestral strain (Figure S2). To further explore this phenotype, we tested whether evolved lineages inhibit ancestral cells in a contact-dependent manner. We co-cultured evolved and ancestral cells using trans-well culture dishes, in which the two populations are separated by membranes of different porosities [9]. The growth of ancestral cells was inhibited when the populations were separated by a cell-permeable 8. 0 µm filter, but not when cell contact was prevented with a 0. 4 µm filter (Figure 1B). These results indicate that evolved cells must be in close proximity to target cells in order to inhibit growth. This phenomenon is reminiscent of Rhs-mediated growth inhibition, which we recently characterized for D. dadantii 3937 [2]. StLT2 contains a single rhs locus, which contains a full-length “main” rhs gene (STM0291) and an “orphan” rhs gene fragment (STM0292) (Figure 2). Both rhs genes are closely linked to small open reading frames representing potential rhsI immunity genes (Figure 2), although the predicted rhsImain immunity gene found downstream of rhsmain is not annotated in the genome sequence NC_003197. To determine if the rhs region is responsible for the observed growth inhibition, we tested whether over-expression of either rhsImain or rhsIorphan immunity genes provided protection against evolved StLT2 lineages. Parental StLT2 cells overexpressing rhsImain were still inhibited by the evolved lineages, but overexpression of the rhsIorphan gene fully protected targets from growth inhibition (Figure 1A). These data strongly suggest that evolved StLT2 cells gained the ability to deliver Rhs-CTorphan toxin into neighboring cells. We next tested each rhs/rhsI gene pair to confirm that they encode functional toxin and immunity proteins. Nucleotides 3608 to 4095 of rhsmain and nucleotides 269 to 741 of rhs-CTorphan were cloned under the control of the arabinose-inducible PBAD promoter. The predicted rhsI immunity genes were cloned using a compatible plasmid under control of the IPTG-inducible Ptrc promoter. These plasmids were then introduced into StLT2 cells to evaluate toxin and immunity functions. Induction of either rhs-CTmain or rhs-CTorphan in StLT2 resulted in rapid growth arrest (Figure 3A). In each instance, growth inhibition was neutralized by expression of the cognate rhsI immunity gene. However, co-expression of non-cognate immunity genes did not alleviate growth arrest (Figure 3A), demonstrating that RhsImain and RhsIorphan immunity proteins are specific for their cognate toxins. We obtained essentially identical results upon expressing the rhs main toxin and immunity genes in E. coli cells (Figure 3B). These results indicate that Rhs-CTorphan is capable of inhibiting bacterial growth and support a model in which evolved StLT2 lineages deploy the orphan toxin to inhibit the ancestral strain. The rhs-CTorphan sequence does not encode a full-length Rhs protein, raising the question of how this toxin is synthesized and exported from evolved cells. The rhsmain and rhs-CTorphan coding regions share 95% sequence identity over 522 base-pairs (Figure 2), raising the possibility that homologous recombination in the evolved lines generates a new full-length rhs gene that encodes the Rhs-CTorphan toxin domain [10]. Bacteria expressing this Rhs chimera would have a growth advantage if rhsIorphan expression is low in ancestral cells. However, the proposed recombination event would also delete the rhsImain gene, rendering the evolved cells sensitive to inhibition by siblings expressing the main Rhs-CT toxin. Therefore, we hypothesized that rhs recombination occurs subsequent to duplication of the locus such that evolved cells retain the rhsImain immunity gene (Figure 2). To test this hypothesis, we analyzed chromosomal DNA from evolved and ancestral lineages by Southern blot. DNA was digested with HincII, which cleaves between the rhsImain and rhs-CTorphan coding sequences, and probed with a labeled DNA fragment that specifically hybridizes to rhsmain (Figure 2). We detected a unique junction fragment representing fusion of rhs-CTorphan to the upstream rhsmain gene in StLT2 lineage 2, which displayed the highest level of growth inhibition of all lineages (Figures 1 & 4A). The wild-type rhs locus was also detected in lineage 2 (Figure 4A), which is consistent with rhs region amplification, but may also indicate distinct populations of recombinant and non-recombinant cells. Orphan rhs recombinants were not detected in the other evolved lineages by Southern blot analysis (Figure 4A). Because the growth inhibition phenotype varied in magnitude between the different evolved strains, it is possible that only a fraction of the evolved StLT2 cells are rhs recombinants. If so, then the proportion of recombined rhs loci in the DNA sample may be below the detection limit of Southern analysis. Therefore we analyzed each evolved lineage with quantitative real-time PCR (qPCR) to measure the relative levels of rhsmain-rhsorphan junction sequences. All five of the evolved lineages contained 10- to 1,000-fold more rhsmain-rhsorphan junction than ancestral StLT2 (Figure 4B), consistent with the ability of these strains to deploy Rhs-CTorphan toxin. Because only a fraction of the passaged cells appeared to display growth inhibitory activity, we asked whether inhibitor-cell clones could be isolated from each population. As a control, we first isolated colonies from an overnight culture of the ancestral strain and tested these clones for growth inhibition activity. None of the ten ancestral clones tested were inhibitory, suggesting that the proposed rhs rearrangements occur at low frequency. By contrast, approximately 30–90% of the clones isolated from the culture-evolved lineages and ∼20% of the clones from mouse-evolved lineage 1 showed inhibition activity against ancestral cells (Figure 5A). However, no inhibitor clones were isolated from mouse-evolved lineage 2 (Figure 5A). Strikingly, the inhibition activity of these clones varied considerably. For example, competitive index values ranged from 10−1 to 10−5 for competitions between ancestral cells and inhibitory clones isolated from evolved lineage 2 (Figure S3). Although their potencies varied, it appears that each inhibitor-cell clone deployed the Rhs-CTorphan toxin because ancestral cells could be protected through over-expression of rhsIorphan, but not rhsImain (Figure 5B). The presence of DNA fragments corresponding to both ancestral and recombinant rhs loci in lineage 2 (Figure 4A) suggests that either the rhs region was duplicated or there are distinct populations of recombinant and non-recombinant cells. In the latter case, single colonies isolated from the inhibitory lineages would contain only the rhs- rhs-CTorphan junction and not the rhs-CTmain sequence. However, PCR analysis of the single colonies with inhibitory activity in Figure 5B showed that each contained both ancestral and recombinant rhs loci. In addition, sequence analysis of the recombinant PCR product verified that recombination occurred between the regions of homology shared by rhsmain and rhs-CTorphan. Together, these data demonstrate that the evolved populations are heterogeneous with respect to Rhs-CTorphan mediated inhibition activity. Furthermore, these results suggest that the inhibition phenotype of a given culture may be due entirely to a minor subpopulation of potent inhibitor cells. Because inhibitor cells represent a subpopulation in the evolved cultures, the other non-recombinant cells in the cohort are presumably resistant to the Rhs-CTorphan toxin. To test this hypothesis, we isolated non-inhibitory clones from each of the evolved cultures and tested them in competition co-cultures against their respective evolved lineages. As predicted, each of the non-inhibitory clones was either fully- or partially-resistant to its cohort lineage (Figure 5C). These cells likely carry uncharacterized resistance mutations that may prevent cell-cell contact, block the delivery of Rhs-CTorphan toxin, , or increase the immunity of these cells to Rhs-CT toxin. To directly detect Rhs-CTorphan expression in the evolved lineages, we examined cells by immunofluorescence microscopy using polyclonal antibodies against the Rhs-CTorphan toxin. Rhs-CTorphan antigen was detected on the surface of some cells within evolved lineages 1,2 and 3 as well as mouse-evolved lineages 1 and 2 (Figures 6A & S4). In contrast, the Rhs-CTorphan signal was undetectable on the surface of both ancestral StLT2 cells and cells carrying a deletion of the rhs-CTorphan (Figures 6A & S4). We then quantified the fraction of cells with Rhs-CTorphan antigen on the cell-surface using flow-cytometry. Evolved lineages showed a 2- to 20-fold increase in the fraction of Rhs-CTorphan-positive cells compared to ancestral StLT2 cells (Figures 6B & S5). Mouse-evolved StLT2 showed very low expression of Rhs-CTorphan antigen on cell surfaces (Figure 6B), consistent with the modest growth inhibition observed for these lineages (Figure 1). Based on Southern blot and RT-qPCR analyses, it seems likely that surface expression of Rhs-CTorphan requires rhs locus rearrangement to generate a chimeric rhs gene. In accord with this conclusion, we also found that over-expression of rhs-CTorphan from a multicopy plasmid does not increase Rhs-CTorphan antigen levels on the cell surface (Figures 6B & S5). Therefore, we sought to detect the predicted Rhs fusion protein using antisera to the Rhs-CTorphan toxin. Western blot analysis revealed an immuno-reactive protein at ∼150 kDa in culture evolved lineages 2 and 3 (Figure 6C). This product corresponds to the expected size of the Rhs fusion protein. Moreover, we were unable to detect the 29 kDa product encoded by rhs-CTorphan in the ancestral and evolved lineages (Figure 6C). Together, these data strongly suggest that the rhs-CTorphan reading frame must recombine with rhsmain to be expressed. The results presented here show that serial passage of StLT2, in either laboratory media or within a natural host, leads to enrichment of cells that express Rhs-CTorphan toxin. Analysis of the rhs locus indicates that evolved cells undergo recombination between rhsmain and rhs-CTorphan, forming a gene fusion that allows the Rhs-CTorphan toxin domain to be deployed. Rearranged rhs genes are detected at low levels within the evolved populations, indicating that only a fraction of cells are recombinant inhibitors. A number of observations argue that this subpopulation of cells is responsible for growth inhibition activity. First, the relative competitive advantage of each evolved lineage is correlated with its level of recombinant rhs junctions and surface expression of Rhs-CTorphan antigen. More importantly, ancestral cells are fully protected when they over-express the rhsIorphan immunity gene. Because Rhs immunity proteins are highly specific for their cognate toxins, this latter result demonstrates that Rhs-CTorphan toxin is indeed deployed by the evolved lineages. This result also indicates that ancestral StLT2 cells do not normally express rhsIorphan immunity genes under laboratory conditions. The number of inhibitor cells within each lineage is not known, but can be estimated to be <2% of the population based on flow cytometry measurements of Rhs-CTorphan antigen on cell surfaces. However, we note that this assay may underestimate the actual number of recombinant inhibitor cells because Rhs effectors are likely exported through type VI secretion systems [2], [11]. Although recent studies indicate that the N-terminal PAAR domain found within many Rhs proteins forms the tip of the type VI injection structure [12], other structural studies show that Rhs-peptide repeats form a chamber capable of encapsulating toxin domains [13]. Therefore, much of the Rhs-CTorphan antigen may be inaccessible to antibody until it is delivered to target cells. In accord with this model, we only detect Rhs-CTorphan where two bacteria make contact with one another and never on the surface of individual cells. Regardless of the absolute number of recombinants or rhs expression levels, our results suggest that a small number of inhibitor cells are capable of inhibiting a large excess of ancestral cells. The same phenomenon has been observed during bacterial contact-dependent growth inhibition (CDI), in which each CDI+ cell is able to inhibit 100–1,000 target cells over a few hours [9]. Presumably, the unstructured environment in shaking broth culture promotes a series of transient cell-cell interactions, thereby enabling toxin delivery to multiple ancestral cells. Chromosomal duplications and amplifications occur frequently in bacteria, typically at rates of about 0. 1% per generation for any given locus [14], [15]. However, there is a cost to maintaining amplified regions, and gene duplications are lost during segregation at frequencies up to 10% per generation [16], [17]. Therefore, positive selection is required to retain multiple gene copies. If the amplified region can be stabilized, then the additional gene copy can diverge towards a new function, thus providing a mechanism for evolution [16], [18]. Rearrangement of rhs loci represents a previously unrecognized mechanism for bacteria to exploit chromosomal amplifications for adaptation. We propose that, subsequent to duplication, homologous recombination occurs between rhsmain and rhs-CTorphan to generate a novel chimeric rhs element. This recombination would necessarily delete one copy of the rhsImain, but the other copy would remain and ensure that recombinant cells retain immunity to the Rhsmain toxin should it be deployed by neighboring non-recombinant siblings. This model also predicts that evolved recombinant cells could undergo homologous recombination to restore the original rhs locus (see Figure 2, reverse of the duplication step). Thus, rhs rearrangement could be exploited transiently under conditions where it confers a selective advantage, but rapidly revert back to the ancestral genotype as environmental circumstances dictate. Analysis of over 150 Salmonella genomes shows that rhs-CT toxin sequences are diverse with at least 57 distinct sequence types (Figure S6A & Table S1). This is a common feature of rhs genes in other bacteria as well and suggests that Rhs mediates inter-strain competition. All Salmonella serovars contain at least one rhs gene, located on pathogenicity islands SPI-6 or SPI-19 [19], [20]. Approximately 50% of these serovars contain at least one predicted rhs orphan sequence, with some strains containing as many as eleven modules. There is generally high conservation of Rhs-CTmain and Rhs-CTorphan. sequences within a given serotype. For example, all sequenced Typhi isolates contain the same Rhs-CTmain and Rhs-CTorphan. sequences, whereas these CT sequence types are only found in one other serotype, thus suggesting that different toxins are linked to serotype and/or the type of infection. However, orphan rhs-CT sequences in one serotype can be present within the main rhs gene of another serotype. For example, the StLT2 Rhs-CTorphan toxin studied in this work is part of the full-length main Rhs in Salmonella enterica serovar Saintpaul SARA23 and some Newport isolates (Figures S6A & S6B). These observations and the association with horizontally transferred elements suggest that rhs genes are exchanged between different serovars and contribute to the evolution of toxin diversity. Given that Rhs toxins are encoded on pathogenicity islands, it seems likely that these systems also play important roles in Salmonella growth and fitness during pathogenesis. Indeed, StLT2 mutants lacking a chromosomal region containing rhs-CTorphan are outcompeted by wild-type cells in mice [21], and StSL1344 mutants lacking rhs are completely attenuated in pig and cattle models of infection [22]. These observations raise the possibility that rhs locus rearrangement occurs commonly during infections. Intriguingly, StLT2 produces distinct intracellular infection foci, each originating from one or only a few clones [23]. Similarly, analysis of mice orally infected with Yersinia pseudotuberculosis indicates that only a few bacterial clones are able to disseminate from the intestines to the spleen and liver [24]. Clonal invasion has also been reported for Yersinia enterocolitica infections [25], but the mechanisms underlying these apparent dissemination bottlenecks are unknown. Most Yersinia species contain rhs loci with associated orphan gene pairs, raising the possibility that clonal expansion through rhs recombination and growth selection may be a general feature of many enterobacterial infections. Rearrangement could function as a stochastic switch that enables some cells to deploy Rhs-CTorphan and thereby “differentiate” into cells that are specialized for tissue invasion or immune modulation. Although Rhs-mediated inhibition clearly occurs between bacteria, it is also possible that Rhs toxins act directly as virulence factors. The C-terminal region of RhsT from Pseudomonas aeruginosa was recently shown to be delivered into mouse host cells [26]. In the process, the Rhs fragment activates the inflammasome and contributes to pathogenicity. Bacterial strains were derived from Salmonella enterica serovar Typhimurium LT2 (StLT2) and are listed in Table S2. Bacteria were grown in LB medium [27] supplemented with 50 mM potassium phosphate (pH 7. 3). Bacteria were incubated at 37°C with shaking at 200 rpm. Where appropriate, media were supplemented with antibiotics at the following concentrations: ampicillin (Amp), 200 mg/L; chloramphenicol (Cam), 17 mg/L; kanamycin (Kan), 80 mg/L; and tetracycline (Tet), 5 mg/L. Six independent lineages of StLT2 were obtained by serial passage for ∼1,000 generations [7]. Each lineage was passaged daily by dilution of 1. 5 µL of overnight culture into 1. 5 mL of fresh LB medium. Each evolved lineage was sampled periodically (100–150 generations) and stored at −80°C. All growth competitions were conducted using ancestral StLT2 marked with the flhC: : cat allele, which confers Cam resistance. Non-inhibitory clones isolated from the evolved cultures were transduced with the flhC: : cat allele prior to testing for resistance. Ancestral and evolved cells were co-cultured in LB medium supplemented with 50 mM potassium phosphate (pH 7. 3) at 37°C with shaking. At time 0 h, ∼106 cfu (1 µL of overnight culture) from evolved and ancestral cultures were suspended in 2 mL of fresh LB (pH ∼7. 3) and plated for viable cell counts before shaking incubation for 24 h at 37°C. After 24 h of co-culture, viable cell counts were determined by plating onto LB agar (to enumerate evolved and ancestral cells) and LB agar supplemented with Cam (to enumerate ancestral cells). The competitive index was calculated as the ratio of ancestral∶evolved cells at time 24 h divided by the cell ratio at 0 h. Ancestral StLT2 flhC: : cat cells were also supplemented with either rhsImain or rhsIorphan on the chromosome under control of the lac promoter and on plasmid pBR322 under the tet promoter. Chromosomal rhsIorphan and plasmid-borne rhsIorphan individually provided partial protection against the evolved lineages (data not shown), but both copies were required for full immunity. Proximity-dependence of growth inhibition was determined as described previously [9]. Cells were grown to OD600 ∼0. 3, then transferred to a trans-well culture plate (BD diagnostics) that separates the two populations with filter containing 0. 4 µm (no-contact) or 8. 0 µm (contact) pores. Trans-well culture plates were seeded at an evolved∶ancestral cell ratio of 1∶1 and incubated at 37°C with shaking for 24 h. Cultures were then plated onto selective media to determine viable cell counts and to calculate competitive indices. All oligonucleotides used in this study are presented in Table S3. The rhsImain and rhsIorphan genes were amplified from ancestral StLT2 chromosomal DNA using oligonucleotides 2337/2338 and 2340/2544 (respectively) and ligated to plasmid pBR322 using EcoRV and SalI restriction sites. The immunity genes were also placed under the lac promoter at the glmS locus using bacteriophage λ Red-mediated recombination [28]. Integration constructs containing rhsI genes flanked by a Kan-resistance cassette and glmS-derived homology regions were constructed by overlapping end-PCR as described previously [29]. The following primer pairs were used to amplify: upstream glmS homology (2666/2676), lac promoter (2677/2678 for rhsImain and 2677/2682 for rhsIorphan), rhsImain (2679/2680) or rhsIorphan (2683/2684), Kan-resistance cassette (2618/2619) and downstream glmS homology (2681/2667). The final PCR product was electroporated into StLT2 cells that express Red recombinase proteins, and transformants were selected on LB supplemented with Kan. Integrated immunity genes were verified by PCR analysis using primers 2666/2667 and subsequent DNA sequencing. The flhC: : cat and STM0292: : kan alleles were generated by PCR using primers 2436/2437 and 2410/2490 to amplify the cat/kan cassettes of plasmids pKD3 and pKD4, respectively. Each PCR product was integrated into the StLT2 chromosome by Red-mediated recombination. To evaluate toxin activity and the specificity of immunity, individual rhs-CT and rhsI sequences were cloned under the control of inducible promoters on compatible plasmids. The rhs-CTmain and rhs-CTorphan coding sequences were amplified with primers Sty-rhs (E1203) -Nco/Sty-rhs-Xho and Sty-rhs (E1203) -Nco/Sty-orph-rhs-Xho (respectively) and ligated to plasmid pCH450 [30] using NcoI and XhoI restriction sites. The rhsImain and rhsIorphan genes were amplified and ligated to a derivative of plasmid pTrc99A using KpnI and XhoI restriction sites. Rhs-CTorphan was expressed and purified as a non-toxic variant fused to His6-tagged thioredoxin. The his6-trxA sequence was amplified from plasmid pSH21P: : trxA [31] using primers pET-Sph and trxA-Bam-TEV-Kpn. The product was digested with SphI/BamHI and ligated to plasmid pET21b to generate plasmid pSH21P: : trxA-TEV. The coding sequences for Rhs-CTorphan residues 112–247 and RhsIorphan were amplified using primers Sty-rhs (D1225) -Kpn/Sty-orph-rhsI-Xho) and the His208Ala mutation made by mega-primer PCR using oligonucleotide Sty-CTo1-H208A. The final product was digested with KpnI/XhoI and ligated to plasmid pSH21P: : trxA-TEV to generate plasmid pCH10068. The resulting construct was used to overproduce His6-TrxA-Rhs-CT (H208A) orphan fusion protein. Chromosomal DNAs were isolated using the Sigma genomic DNA kit and digested with HincII restriction endonuclease. Digested DNAs were resolved by electrophoresis on 0. 7% agarose gels at 34V for 15 h and blotted onto nylon membranes by capillary transfer. A probe to nucleotides 2969–3128 of rhsmain was generated by PCR using oligos 2226/2227 and labeled with [32P]-labeled using the Prime-It Random Primer Labeling Kit (Agilent Technologies). Southern blots were visualized by phosphor imaging. Fragment sizes were calculated using a standard curve based on HindIII digested λ ladder (New England Biolabs, USA) run on the same gel. The proportion of rhs recombination junctions was determined by quantitative real-time PCR (qPCR) using oligonucleotides 2226/2231 using the cycle threshold Ct-value method according to the manufacturer (Bio-Rad). Fluorescence was monitored on-line using the MyIQ iCycler real-time PCR system (Bio-Rad). The rhs-rhsorphan junction DNA levels were calculated relative to bamA DNA (oligos 1981/1990) in each sample and normalized to the level of junction DNA in ancestral cells. His6-TrxA-Rhs-CT (H208A) orphan fusion protein was overproduced in E. coli CH2016 and purified by Ni2+-affinity chromatography as described [32]. The Rhs-CT (H208A) orphan domain was released by TEV protease digestion and used for antiserum production in rabbits (CoCalico Biologicals). Non-specific antibodies were removed by incubation with carbonyl-diimidazole-activated agarose beads linked to soluble protein from E. coli strain CH2016 [33]. Briefly, protein-linked beads were resuspended in 0. 5 mL of antiserum (1∶5 dilution) and mixed by rotation for 1 h at room temperature followed by additional incubation for 3 h at 4°C. This process was repeated at least four times with fresh beads. Evolved lineages were grown to mid-log phase in LB medium supplemented with 50 mM potassium phosphate (pH 7. 3) and cells were collected by centrifugation and frozen at −80°C. Cell pellets were resuspended in NuPage LDS-sample buffer (Invitrogen) at 70°C and treated with benzonase to degrade nucleic acids. Cell lysates were run on 3–7% NuPage Tris-acetate gradient gels (Novex) for the detection of the Rhsmain-Rhs-CTorphan chimera, or on 4–10% Precise Tris-glycine gradient gels (Thermo Scientific) to detect Rhs-CTorphan. Gels were electrotransferred to nitrocellulose membranes and the blots incubated with polyclonal antisera against Rhs-CTorphan (1∶1,000 dilution) and secondary anti-rabbit 800CW antiserum (1∶10,000 dilution). Immunoblots were visualized using an Odyssey CLx Infrared Imaging System (LI-COR). Cells were incubated overnight with 4% formaldehyde in 0. 15 M phosphate buffered saline (PBS, pH = 7. 2). Cells were washed three times with PBS and incubated with polyclonal antibodies to Rhs-CTorphan (1∶50 dilution) for 30 min. Cells were washed with PBS before incubation with secondary anti-rabbit Alexa-Fluor480 antibodies (1∶500 dilution) (Invitrogen) for 30 min on ice. After washing with PBS, cells were applied to poly-D-lysine coated slides, treated with Fluoro-gel II/DAPI (Electron Microscopy Sciences) and visualized by fluorescence microscopy. The fraction of evolved cells expressing Rhs-CTorphan on the surface was determined by flow cytometry. Antibody-labeled cells were analyzed (50,000 events for each sample) with an Accuri C6 flow cytometer with gates set to include bacteria-sized particles. StLT2 Δrhs-CTorphan cells were used to assess non-specific binding of the Rhs-CTorphan antisera. The fraction of cells with surface Rhs-CTorphan antigen was calculated as the ratio of green fluorescent particles in the population after subtracting background fluorescence observed with StLT2 Δrhs-CTorphan cells.

Salmonella Typhimurium is a bacterium that causes intestinal diseases in a number of animals including humans. In mice, this pathogen invades tissues, causing symptoms similar to typhoid fever. In an effort to understand the evolution of this pathogen, we grew S. Typhimurium in either liquid broth or in mice for many generations and examined the resulting “evolved” strains to determine if they were different from the original “parent” culture. We found that many of these evolved strains inhibited the growth of the parent after they were mixed together, and that this growth inhibition requires that the evolved and parental cells are in close contact. Genetic analysis showed that this contact-dependent growth inhibition requires Rhs protein, which has a toxic tip. Salmonella is normally resistant to its Rhs toxin because it also produces an immunity protein that blocks toxin activity. However, evolved cells have undergone a DNA rearrangement that allows them to express a different Rhs toxic tip that inhibits growth of the parental cells, which lack immunity to it. This allows the evolved cells to outgrow the original parental cells. Our work indicates that populations of Salmonella are dynamic, with individuals battling with each other for dominance.

Abstract Introduction Results Discussion Materials and Methods

bacteriology cell death organismal evolution microbial mutation genome evolution cell processes population genetics microbiology developmental biology mutation molecular cell biology cell growth microbial evolution molecular genetics population biology microbial growth and development bacterial pathogens genomics microbial physiology medical microbiology microbial pathogens molecular evolution salmonella bacterial physiology cell biology natural selection genetics biology and life sciences gene duplication computational biology evolutionary biology bacterial evolution evolutionary processes

2014

Selection of Orphan Rhs Toxin Expression in Evolved Salmonella enterica Serovar Typhimurium

Communication between cells is a ubiquitous feature of cell populations and is frequently realized by secretion and detection of signaling molecules. Direct visualization of the resulting complex gradients between secreting and receiving cells is often impossible due to the small size of diffusing molecules and because such visualization requires experimental perturbations such as attachment of fluorescent markers, which can change diffusion properties. We designed a method to estimate such extracellular concentration profiles in vivo by using spatiotemporal mathematical models derived from microscopic analysis. This method is applied to populations of thousands of haploid yeast cells during mating in order to quantify the extracellular distributions of the pheromone α-factor and the activity of the aspartyl protease Bar1. We demonstrate that Bar1 limits the range of the extracellular pheromone signal and is critical in establishing α-factor concentration gradients, which is crucial for effective mating. Moreover, haploid populations of wild type yeast cells, but not BAR1 deletion strains, create a pheromone pattern in which cells differentially grow and mate, with low pheromone regions where cells continue to bud and regions with higher pheromone levels and gradients where cells conjugate to form diploids. However, this effect seems to be exclusive to high-density cultures. Our results show a new role of Bar1 protease regulating the pheromone distribution within larger populations and not only locally inside an ascus or among few cells. As a consequence, wild type populations have not only higher mating efficiency, but also higher growth rates than mixed MATa bar1Δ/MATα cultures. We provide an explanation of how a rapidly diffusing molecule can be exploited by cells to provide spatial information that divides the population into different transcriptional programs and phenotypes. Cells communicate with each other by detecting and responding to external cues and stimuli. In cellular systems one can find several examples where cells coordinate growth in specific regions by sensing and responding to small differences in the concentrations of substances that provide positional information [1], [2]. Such signaling among cells is a common feature of cell populations as well as multicellular organisms, where cells often operate close to the physical limit of gradient or concentration detection [3]–[5]. The pheromone response of budding yeast (Saccharomyces cerevisiae) is an example for a eukaryotic cell communication system. Yeast cells occur either in the haploid forms MATa and MATα, or as MATa/α diploid. Haploid and diploid cells are both able to replicate vegetatively. Mating of two haploid cells with opposite mating types yields diploid cells, while haploid cells are formed through spore formation in meiosis [6], [7]. Mating is initiated by the secretion of mating type-specific pheromones, called a-factor and α-factor, which are sensed by haploid cells of the opposite mating type and trigger the mating response [8], [9]. During the mating response, yeast cells arrest their cell cycle in G1 phase and elongate in the direction of the pheromone signal by forming directed mating projections called “shmoos” [6], [10]. Yeast can sense pheromone gradients as well as absolute concentration levels. Nevertheless, yeast cells are not capable of chemotaxis and, thus, mating requires the haploid cell types to signal their location particularly to nearby potential mating partners. One way for MATa cells to regulate the extracellular α-factor is the secretion of Bar1, an aspartyl protease, which degrades α-factor [11], [12]. This leads to the paradox that MATa cells degrade the signal they need to receive. Theoretical investigations hypothesized that the major role of Bar1 is to sharpen the pheromone gradient [13], [14]. However, direct visualization of the resulting spatial α-factor concentration profiles between secreting and receiving cells is impossible due to the small size of diffusing molecules and because such visualization requires experimental perturbations such as attachment of fluorescent markers, which change diffusion properties and activities. Therefore, reaction-diffusion (RD) models have been used to simulate the pheromone distribution on the basis of physical properties of the molecules [13]–[16], but neither the validity of the model predictions nor the effect of the pheromone distribution have been tested experimentally. Except of the publication of Jin et al. [15], where it has been suggested that Bar1 promotes avoidance of the same mating type and accurate gradient detection. Using a microfluidic device it had been shown how MATa cells avoid each other when exposed to an artificial unidirectional gradient, which was reproduced quite vividly by simulations of an RD model. However, the assumptions and choice of parameters were in contrast to other works [14], [16] (compare Text S1). In general, experimental validation of RD models is complicated, especially on the molecular level [17]. Moreover, recent theoretical findings support the theory, that secretion of Bar1 in the extracellular medium does not help to align gradients in the direction of the opposing mating type [16]. In summary, the role of Bar1 is still controversially discussed. Also, none of the models proposed so far has investigated interactions of more than a few cells or the four haploid spores in an ascus, even though mating occurs not only inside the ascus, but also in a cell population which was shown by new findings where a remarkably high outcrossing rate from asci was reported [18]. This indicates that mating yeast cells interact with quite a number of potential mating partners in a natural environment. Furthermore, a recent study has shown the potential of simple secrete and sense motifs to exhibit surprising effects on the population level [19]. Therefore, we designed a method to identify the most likely α-factor distribution within mixed haploid yeast populations of thousands of cells directly from confocal microscopic images with fluorescently tagged marker proteins. Here, an RD model is used to simulate interactions of a few hundred cells at the same time. In MATa cells, the protein Fus1, which is strongly expressed upon pheromone stimulation, is tagged with GFP to record pheromone pathway activation and serves as a proxy for Bar1 induction. Therefore, the experimentally observed pheromone activation level of each MATa cell is integrated into the model and compared to the experiments. MATα cells are modified to express mCherry from the TDH3 promoter to indicate their mating type and location. In a unique way we coupled physical RD models with experimental imaging in order to quantify the spatial distribution of extracellular α-factor. The use of simple marker constructs, that altered neither α-factor nor Bar1, served for minimal interference with the biological system. We used this approach to directly estimate the influence of Bar1 on the distribution of α-factor in a mixed yeast population and suggest a novel function of Bar1 to enable the coordination of mating and growth in a yeast population in vivo. We combined image analysis with spatiotemporal mathematical modeling to determine spatial concentration distributions of Bar1 and of α-factor. Figure 1 introduces the concept of the approach: In order to obtain in vivo conditions during microscopy, which were suited for the described methodology, confocal microscopic images were taken from synchronized haploid cells or from equally mixed haploid MATa/MATα cell populations. The cell culture samples were spun down with low g-force on glass bottom dishes in order to have immobile cells without using concanavalin A coated dishes (which we found to alter the population response). This protocol essentially yielded sedimented cells in the same way as they would be present in any laboratory or naturally occurring non-agitated medium, with the difference that sedimentation here was achieved under controlled conditions. Location, shape, and mating type of the cells were extracted from out-of-focus images in the brightfield and mCherry channel (Figure S1 in Text S1). The cells' spatial arrangement on the images was transferred into locally refined triangular meshes for the model (Figure S2 in Text S1) and used to calculate the extracellular spatial distributions of Bar1 and α-factor. In mathematical terms the problem (ii) is described as a pure extracellular reaction-diffusion process for α-factor and Bar1 with distinct boundary conditions (Figure S3 in Text S1). We formulated an RD model by the following set of partial differential equations: This system covers the extracellular dynamics of α-factor concentration, , and the activity of Bar1 protease, , over time t and position in space, . The equations quantify two types of processes: (1) diffusion of both α and B, where and are diffusion constants and Δ denotes the Laplacian, and (2) degradation of α by B. Boundary conditions at the cell surfaces define secretion of α-factor by MATα cells and induced secretion of Bar1 by MATa cells. The diffusive flux of α-factor on the cell surface and exterior system boundaries is given by: Here, is a constant specifying the secretion of α-factor. The vector points towards the cell interior and, therefore, the diffusion flux at the boundary takes a negative value for the secretion of molecules and a positive sign for the absorption of molecules. The induction of Bar1-activity at each MATa cell is calculated from the average α-factor concentration at the surface of this cell, which is represented by a Hill-curve (compare Figure 1 E, for details see Figures S4 and S5 in Text S1): The expression is the average concentration of α-factor at the i-th MATa cell at time, which promotes the secretion of Bar1. We assume a zero Bar1 flux on the MATα cells and the system exterior: The remaining exercise was to identify the parameter values of the RD model. Diffusion constants for the proteins were directly calculated from protein properties (size and density). Thus, only two parameters had to be identified by parameter estimation: the activity of Bar1 and the secretion rate of α-factor, as described below. To determine the activity of Bar1, we used a three-step procedure. First, we performed an initial calibration to quantify the α-factor concentration perceived by individual MATa cells. We used MATa cells carrying a BAR1 deletion (bar1Δ) and the pheromone response marker Fus1-GFP and synchronized them in G1 phase, where they are responsive to pheromone. Their response to varying concentrations of α-factor (artificially added, in the absence of MATα cells) was quantified as Fus1-GFP fluorescence intensity on microscopic images. Since bar1Δ strains do not secrete the protease, the local α-factor concentration was equal to the applied concentration. Fluorescence intensity of Fus1-GFP in correlation with α-factor concentration was recorded as a calibration curve (see Figure S4 in Text S1). The calibration curve was then applied to mixed haploid cultures to determine the perceived α-factor concentration for each MATa cell on the image (see Figure 1). In mathematical terms, the calibration yielded the boundary value of α-factor concentration at the surface of a MATa cell and a functional relation between this α-factor concentration and the induced Bar1 expression. Second, the steady state activity of Bar1 was quantified by stimulating wild type MATa cells carrying the Fus1-GFP marker in G1 phase with given concentrations of α-factor. Since there were no MATα cells, the α-factor secretion rate was equal to zero at this stage. For the model, these data were used in a mathematical optimization to quantify the activity of Bar1 for wild type cells. In order to verify that Fus1 can in fact be used as a proxy for Bar1, we used a strain expressing qVenus fluorophore under the control of the Bar1 promoter (Bar1pr-qVenus, [20]) while maintaining wild type Bar1 activity (see Figure S5 in Text S1 and Materials and Methods). The induction occurred with the same kinetics and Hill-coefficients as the Fus1 induction, but with a slight delay of the Bar1 expression verifying our used induction kinetics for Bar1. The Bar1pr-qVenus construct was able to quantify the expression levels of Bar1, but not the Bar1-induced degradation rate of α-factor in the extracellular medium, which is why we preferred the use of the Fus1-GFP data. Third, the α-factor secretion rate could be calculated by parameter optimization from images of mixed haploid yeast cultures containing MATa Fus1-GFP cells and MATα mCherry cells and the information we obtained for the Bar1 activity. Due to the lack of evidence for an extracellular protease activity in MATα cells, we neglected potential differences in a-factor induction of α-factor. In practice, the induction of α-factor secretion should be nearly homogeneous on a single image, but may vary for different images. We obtained secretion rates from significant fits (F-test p<0. 05) of 9 images (see Text S1). The mean fitted value of the α-factor secretion rate of 865 molecules per second and MATα cell is in good agreement with recent experimental measurements (550 molecules per second and cell measured as basal secretion with 2. 5 - 4-fold maximal induction) [21], [22]. The fully parameterized model could now be used to efficiently calculate the entire α-factor distribution on arbitrary images. We validated the obtained model and parameters by predicting the Fus1-GFP fluorescence on an image of a larger mixed yeast population not used for model fitting (F-test p<2. 2e-6). The power of this combined imaging and modeling approach is illustrated in Figures 2 and 3 and in Movie S1, which shows a mixed haploid population during growth and mating next to the simulation of the distribution of α-factor on a computational grid generated directly from the corresponding microscopic image. We observed large differences in the estimated local α-factor concentrations between wild type cell populations and cell populations with a bar1Δ background. Dense wild type cell populations showed a strongly localized α-factor distribution at sites of high MATα cell density, with α-factor concentration quickly declining with distance. Consequently, MATa cells far away from a cluster of MATα cells experienced significantly lower local α-factor concentrations than close-by cells, and hence were often non-permissive for induction of the pheromone response (Figure 3). Populations with bar1Δ background showed an almost uniform distribution of very high pheromone concentrations, resulting in global pathway activation as evidenced by high Fus1-GFP expression. Nevertheless, the global (over-) activation led to reduced mating events. We wanted to see whether this behavior arises in general and independently of the exact spatial composition of the culture. Thus, we performed a computational study using randomly generated cell populations mimicking the ones observed microscopically with varying cell densities (Figure 4). Each virtual population was simulated both with wild type Bar1 secretion and in bar1Δ background. We tracked key parameters such as the average α-factor concentration, the pheromone gradients perceived by the individual MATa cells (calculated as the average difference in α-factor concentration a cell would sense at its shmoo tip and the opposing cell site), and the maximum information content of the α-factor distribution. Information (or Shannon entropy) quantifies the “surprisal” of a specific α-factor concentration, i. e. how likely an observed α-factor concentration is given an overall α-factor distribution obtained for many cell populations. Virtual wild type populations exhibited a strong gain for the information content of the α-factor distribution as the population size increases (Figure 4A). This was accompanied by increasing α-factor gradients across MATa cells (Figure 4B). In contrast, populations not secreting Bar1 showed information contents close to zero (Figure 4A) as well as insignificant pheromone gradients (Figure 4B), both independently of population density. We noted that the overall pheromone concentration remained within a range of up to 20 nM in wild type, but in the mutant linearly increased with population density (see Figure S6 in Text S1). This observation indicates that the gradients and, thus, the reachability of nearby mating partners can only be detected faithfully in cell populations secreting Bar1, particularly in high cell densities. Additionally, we simulated various scenarios where a high-density subpopulation was placed next to a low-density subpopulation (Figure 4C–D). Here, the wild type is capable of limiting the α-factor distribution to the corresponding subpopulation, leaving the low-density subpopulation unaffected by the high local α-factor concentration of the high-density subpopulation. The same also holds true for random cell distributions (see Figure S7 in Text S1). Again, this behavior was not observed in the absence of Bar1, showing that Bar1 activity restricts the distribution of α-factor. Hence, only subpopulations with high local cell densities and small intercellular distances, as required for successful mating, were exposed to α-factor concentrations permissive for mating. We could validate this model prediction concerning the dependency of mating success on culture density experimentally by incubating mixed MATa Rpl9A-GFP/MATα mCherry populations in cell densities varying from 0. 5–10 million cells/ml for various time-points up to 5 h. We incubated the cells in Petri dishes of 36 mm diameter to ensure that the experimental density in the resulting cell layer is in agreement with the simulated density (density and distance calculations can be found in Text S1). Subsequent cell counting during bead-normalized flow cytometry allowed us to quantify the absolute number of each cell type along with the number of diploid cells in each sample (Figure 5). Here we also observed a strong dependency of diploid formation on cell density where efficient mating was predominantly observed in cell concentrations higher than 5 million cells/ml, which coincided with cell distances permissive for mating (Figure 5A). At the same time we could observe a pronounced growth phase taking place in parallel with diploid formation (Figure 5B), giving support to the finding that there was indeed a variety of different phenotypes (mating and growing cells) occurring at the same time in the same culture. Figure 5C visualizes the cell densities used in the measurements for comparison and Figure S7 in Text S1 shows the simulated levels of α-factor under these conditions, with representation of α-factor distribution and Bar1 activity distribution for one selected cell density. Our results suggested that Bar1 acts by restricting the activity of α-factor to sites, where successful mating is possible, leaving the remaining areas free for continued growth. In order to test the validity of this prediction and its dependency on Bar1 we observed mating between MATa cells (here marked with Rpl9A-GFP) and MATα cells (marked with mCherry) and quantified their growth rates with FACS analysis (Figure 6) and OD measurements in wild type and in bar1Δ populations during incubation (Figure 7). Mating rates were quantified by the rate of diploid formation. To measure the diploid formation rate, we used flow cytometry for mixed populations of MATa and MATα (where we took care to not destroy extracellular gradients before the actual measurement) to quantify the fractions of MATa, MATα and MATa/α diploids over time for a fixed number of cell counts (Figure 7A). We found no difference in the rate of diploid formation between wild type and bar1Δ cultures before completion of the first cell cycle (<120 min). This observation is in agreement with our results that positive effects on the perceived pheromone gradients require higher cell densities (Figure 4A, B). However, after passing the first cell cycle, the relative fraction of diploids is clearly larger in the wild type cultures than in the mutant, consistent with the general view that Bar1 activity helps to reveal the position of mating partners [3], [13], [14] Looking at population growth during mating, we found strong differences between wild type and bar1Δ cultures (Figure 7B). For bar1Δ cultures, the global activation of the pheromone response in effectively all MATa cells of the population led to an almost complete loss of population growth (Figure 7B, C). This also caused a characteristic population phenotype with many pheromone-stimulated MATa cells being significantly larger than normal MATa cells and showing multiple mating projections (Figure 7E, F). This phenotype was never encountered in unperturbed wild type mixtures of MATa and MATα cells, but could be induced by swirling them rapidly to inhibit cell fusion (Figure S8 in Text S1). Thus, this phenotype appears associated with induction of pheromone response in vivo under conditions where a cell cycle arrest has been induced but successful mating is inhibited. Wild type cultures exhibited significant growth on the population level despite the higher rate of diploid formation and a normal phenotype of MATa cells (Figure 7B, D–E). The effect of Bar1 secretion on haploid growth rates was even more prominent when looking at the MATa/MATα ratio in the population (Figure 7C). There is no known secretion of an extracellular protease described for MATα cells equivalent to Bar1. Co-cultured wild type MATa cells strongly outperform MATα cells in growth during mating to an extent that within 5 hours MATa is the predominant haploid cell type in the population. This cannot be observed in bar1Δ background where the MATa/MATα ratio remains constant, presumably because both haploid cell types are equally inhibited in growth. In summary, secretion of Bar1 enables a high mating rate on a population level, but also strongly optimizes the population growth rate by avoiding unnecessary cell cycle arrest when mating is improbable (Figure 7F). In order to further validate our findings about the role of Bar1 in the mating process, we mixed MATα cells with different ratios of wild type and bar1Δ MATa cells. We measured the amount of haploid and diploid cells after 4 h of incubation by FACS analysis (Figure 8). For labeling of MATα cells we again used mCherry, whereas MATa BAR1 wild type cells were labeled with Rpl9a-GFP and bar1Δ MATa cells with Rpl9a-TagBFP2. For equal ratios of wild type cells of both mating types at time 0 h we obtained a diploid fraction of 14% at time 4 h (leftmost columns). However, MATα cells assumed about 23% while MATa cells reached more than 62% of the total population, again supporting the observation that a part of the MATa population engages in mating and other cells continue to grow. For equal ratios of wild type MATα and bar1Δ MATa at start, we obtained only 3. 1% diploids and roughly equal ratios of the haploid cells after 4 h. This is in agreement with the view that essentially all cells stop growing and start to prepare for mating. Microscopic imaging confirms that all cells are shmooing under this condition (Figure 8B). When mixing 50% of MATα with different ratios of wild type bar1Δ MATa cells, we obtain diploid cells of both types at ratios as could roughly be expected from the mixes with either wild type or mutant MATa. Strikingly, for small ratios (5%) of wild type MATa cells at 0 h, we see that bar1Δ MATa mate more frequently (4. 5% of total population) than in pure mutant mixes (3. 1%), indicating that they profit from the Bar1 secreted by wild type MATa cells. Figure 8C shows a microscopic image of a mixture of 5% wild type MATa and 45% bar1Δ MATa cells together with 50% MATα at the beginning. Here, the bar1Δ MATa cells exhibit clearly lower levels of shmooing compared to the 50%/50% mix in Figure 8B. A few successful mating events leading to diploids are indicated by white arrows. Again, the experiments mixing wild type cells with cells not secreting Bar1 – the bar1Δ or cheater cells – confirm the role of Bar1 which is even supportive for those cells not actively secreting it, but experiencing its effect on α-factor levels and gradients. The combination of spatial modeling and a series of in vivo experiments employing images of mating cells allowed us to quantitatively describe the distribution of the pheromone α-factor in the intercellular space and the role of the protease Bar1 in shaping the pheromone pattern. We found that secretion of Bar1 is a highly cooperative mechanism. Haploid MATa cells secrete individually small quantities of Bar1 molecules. These few molecules are quickly distributed across the population by diffusion and generate Bar1 activity that strongly influences the diluted distribution of α-factor. For example, a low cell density of about 106 cells/ml induces a degradation rate of α-factor in the range of 10−3 molecules per second which corresponds to almost no degradation of α-factor, however, high cell densities can create substantial degradation by the Bar1 activity where single α-factor molecules are cleaved within a second after secretion (compare Text S1). Thus, individual MATa cells need to secrete only very small numbers of Bar1 protease. The global concentration of Bar1 appears to be fine-tuned to ensure a highly informative α-factor distribution. This effect is based on the interplay of the secretion of α-factor, its diffusion, the activation of Bar1 transcription and α-factor degradation by Bar1. Thus, we observed steep gradients in the distribution of α-factor only in high densities of MATa cells where the joint degradation by Bar1 limits α-factor diffusion sufficiently. Our conclusions differ from what has been shown in previous publications where the effect of Bar1 is demonstrated either completely theoretical for one or a few cells [13], [14], [16] or experimentally for an artificial setup in a microfluidic device without mating partner [15]. First, we observed little to no effect of Bar1 in a setting with only few cells since the Bar1 activity is insufficient to degrade α-factor before it diffuses a large distance. Our results also indicate that Bar1 rather acts on the level of subpopulations than on a few individual cells. Second, we also did not observe an effect reported earlier that Bar1 secretion leads to “self-avoidance” [15]. Third, we modeled and experimentally quantified the induction of Bar1 due to stimulation with α-factor (compared to constant Bar1 levels as in [13] or constant secretion as in [14], [15]). Remarkably, the simple circuit of pheromone secretion and degradation by jointly secreted Bar1 is able to produce highly dynamic behavior in yeast populations. In general, the α-factor concentration profile recovers the regions where mating has a high chance of success, but quickly drops at all other regions, thus allowing cells to continue growth when there is only little chance for successful mating (see Figure 7E, F). This is further regulated by a stimulated Bar1 production, which depends on the extracellular α-factor concentration. Stimulation of Bar1 secretion might be a strategy to adapt the zone of influence of α-factor to varying cell densities and numbers of MATα cells in the population. Cells will react to high α-factor concentrations with strong secretion of Bar1, which culminates in a steady state permissive for efficient mating (also see Text S1 and Movie S1). A lack of Bar1 in mixed haploid populations has crucial influence on the population phenotype since it leaves many cells in a prolonged cell cycle arrest in G1 phase along with activation of the pheromone response. Moreover, the temporally extended stimulation leads to larger cells with many mating projections. This result is further supported by the observation that MATa cells strongly outperform MATα cells in growth, an effect depending on Bar1. Thus, Bar1 is beneficial for growth as well as diploid formation because it enables continued growth for large parts of the population, but it also provides an enhanced ability to interpret the extracellular pheromone signal at sites where many cells cluster into locally dense subpopulations. While the gradient-enhancing effect of Bar1 has been reported before, we additionally connect it to the requirement of high local cell densities [13]. This indicates that the gradient enhancing ability of Bar1 has evolved to take place selectively in dense yeast populations and is not a treat of individual yeast cells in large volumes. Our observations suggest that the yeast populations segregated into a spatial pattern with localized regions of diploid formation and other regions of continued growth. As a consequence, the entire population is divided into two different work programs. Both of those programs are performed in parallel and Bar1 is sufficient to induce this separation by forming a locally varying α-factor pattern. Depending on the overall cell density, the locations reserved for mating show high local concentrations of α-factor, whereas the locations reserved for growth show negligible concentrations (compare Figures 5 and 7). However, one may wonder how this cooperation arose during evolution since cheater MATa cells not producing Bar1 can also easily exploit it. A possible explanation lies within the spatial structure of the population, since the fitness benefit conferred by Bar1 is locally restricted. Haploids can only arise from a small set of initial spores, which form subpopulations by budding and mating type switching. As a consequence, cells profiting from the collective secretion of Bar1 are likely to be genetically related. Under these circumstances there is indeed evidence that cooperation can be conserved during evolution [23], [24]. Our experiments show that cheater cells not producing Bar1 can indeed profit from the Bar1 secreted by non-cheater wild type MATa cells in a mixed population (compare Figures 7 and 8). However, the non-cheater cells consistently outperform the cheater cells in mating as well as growth, indicating that the non-cheater cells maintain an advantage even in a mixed population of cheaters and non-cheaters. The analyzed regulatory circuit of combined pheromone and protease secretion is not only observed in Saccharomyces cerevisiae, but is also found in other fungi [25]–[27]. Furthermore, a similar mechanism is known in Dictyostelium discoideum secreting phosphodiesterase (PDE) during detection of cAMP [28], [29]. Taking this into consideration, the described mechanism might be a general strategy to separate a cell population into subpopulations with different transcriptional programs. Our methodology of quantifying the distribution of extracellular morphogens in the absence of direct measurement also has potential applications in other problems of cellular communication and pattern formation. A reduction from the computationally very expensive 3D problem (especially for parameter estimation) to an integrated 2D problem is feasible for any cells that sediment to the bottom of the containing volume under non-agitated conditions. However, this computational tool could be used to model the behavior of any culture in a non-moving liquid film such as on the surface of fruits or any controlled fermentation such as wine or beer production where the liquid is kept still for some time. This makes the method applicable for clinical research as well since biofilm formation involving quorum sensing is a major complication when fighting bacterial infections [30], [31]. Here, it might be helpful to quantify the distribution of quorum signals in order to find possible ways to optimally disrupt the system. Wild type MATa reporter strains used in this study are Fus1-GFP and Rpl9A-GFP. They are based on BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) and part of the yeast GFP collection [32]. The MATα reporter strain expressing mCherry under control of the TDH3 promoter (MATα can1Δ STE2pr-SpHIS5 lyp1Δ: : STE3pr-LEU2 his3Δ1 leu2Δ0 ura3Δ0 met15Δ hoΔ0: : TDH3pr-mCherry-NATMX4) was a friendly gift of Alexander DeLuna [33]. As mutant MATa reporter strains we used two different strains: bar1Δ Fus1-GFP and bar1Δ Rpl9a-TagBFP2. The first mutant was created by deletion of the BAR1 gene in the Fus1-GFP strain mentioned above, the second was cloned by tagging of the Rpl9a Gene with Tag-BFP2 in the BY4741 bar1Δ strain. Rpl9a tagging was used because of the high expression level and also since it is not known to be involved in the mating process. Yeast strains were cultivated at 30°C in synthetic medium (0. 17% yeast nitrogen base without amino acids, 0,5% ammonium sulfate, 2% glucose, 55 mg/l adenine, 55 mg/l L-thyrosine, 55 mg/l uracil, 20 mg/l L-arginine, 10 mg/l L-histidine, 60 mg/l L-isoleucine, 60 mg/l L-leucine, 40 mg/l L-lysine, 10 mg/l L-methionine, 60 mg/l L-phenylalanine, 50 mg/l L-threonine and 40 mg/l L-tryptophane). BAR1 deletions in MATa reporter strains were inserted by homologous integration of a URA3 cassette in the BAR1 locus (bar1Δ0: : URA3). PCR amplification of the URA3 cassette from plasmid template pESC-Ura (Stratagene) was done by sequential amplification with the primer pairs 1/2 and 3/4 shown in Table 1. This was followed by transformation and selection on agar plates with synthetic medium lacking uracil. Verification of the BAR1 deletion was done with a physiological assay based on growth inhibition by α-factor pheromone [34]. The bar1Δ Rpl9a-TagBFP2 reporter strain was cloned by PCR amplification of a TagBFP2 loxP-Ura3-loxP transformation cassette with primer pairs 5 and 6 from Table 2. As PCR template we used vector EKP232. EKP 232 was cloned by ligation of TagBFP2 into PstI site of pUG72 [35]. The qVenus expressing strain under control of the Bar1 promoter (Bar1pr-qVenus) was cloned by using the plasmid pSP 34 from Serge Pelet [20]. Bar1 promoter region [−500 bp] was amplified from genomic DNA (strain BY4741) by PCR using primer pair 9 and 10 as well as the first 51 bp of the BAR1 gene using primer pair 7 and 8 shown in Table 3. PCR product of promoter and gene were mixed and used as template for a fusion PCR with a PmlI restriction site between promoter and gene. Included in forward and reverse primer were restriction sites for SacI and PstI respectively. The fusion PCR product was ligated into SacI/PstI site of pSP 34 resulting in plasmid EKP252. Plasmid EKP252 was linearized using PmlI restriction enzyme and used for transformation and homologous integration into the BAR1 locus under preservation of the BAR1 gene. Positive clones were selected in minimal medium lacking leucine, α-factor induced qVenus expression was controlled microscopically, and wild type Bar1 activity was verified by a physiological assay based on growth inhibition by α-factor pheromone [32] and comparison with Bar1 wild type and deletion strains. Microscopic images were acquired with an inverted FluoView 1000 microscope (Olympus, Tokio, Japan) equipped with a 60× (1. 2 N. A) water-immersion objective and a climate chamber (Tokai Hit, Japan). GFP was excited with a 488 nm argon laser and mCherry with a 559 nm laser diode. Fluorescence emission was detected in the range 500–545 nm and 570–670 nm, respectively. The Bar1pr-qVenus construct was excited with 515 nm and detected between 530 nm and 630 nm, for Rpl9a-TagBFP2 we used 405 nm excitation and as detection range 425–475 nm. For mating experiments, MATa Fus1-GFP wild type and bar1Δ reporter strains as well as the MATα reporter strain (TDH3pr-mCherry), were cultivated to mid logarithmic phase and mixed equally. Mating was followed over indicated time periods microscopically while microscopic samples were kept in cultivation medium at 30°C. Image acquisition for α-factor calibration curves was done with synchronized cultures of Fus1-GFP wild type and bar1Δ. Cultures were synchronized in G1 phase by elutriation with a Beckman Coulter JE-5. 0 elutriation system. Synchronized cells were incubated with α-factor pheromone for 3 hours at 30°C. Afterwards cells were spinned down on the surface of a glass bottom dish (MatTek Corporation, Ashland, US) by centrifugation at 100× g using self-built accessories. For Fus1-GFP wild type, α-factor pheromone concentrations in the range between 0 µM–100 µM were used and for bar1Δ we used 0. 1 nM - 1 µM. Mean fluorescence intensity of Fus1-GFP was analyzed as described in the Computational Techniques (see Text S1). For validating the employment of Fus1-GFP as proxy for the mating response pathway we used a strain expressing qVenus under control of the Bar1 promoter in BAR1 wild type background. Non-synchronized cells were incubated with α-factor pheromone as described for Fus1-GFP BAR1 wild type cells and analyzed in the same way. A comparison of the results is shown in Figure S5 in Text S1. Growth of equally mixed MATa and MATα reporter strains, as well as a haploid control strain was analyzed by measuring optical density at 600 nm with a Photometer (Eppendorf Bio Photometer plus) and in parallel by analysis of cell number and cell size distribution with a cell counter (Casy Counter TTC, Schärfe System). Yeast cells were incubated in a water bath at 30°C without shaking. In time steps of 15 min samples were removed from the water bath, vortexed, appropriately diluted, and analyzed in duplicate. To quantify the amount of haploid MATa and MATα cells, of diploid cells or of cells within the mating process, we measured fluorescence intensities for GFP and mCherry of 10. 000 living cells of each sample by FACS analysis taking advantage of the fluorescence of MATa Rpl9A-GFP and MATα mCherry in a BD FACS AriaII cell sorter (Becton Dickinson, Franklin Lakes, NJ), equipped with a 488 nm and a 561 nm laser with filter sets for GFP (525/50 BP, 505LP) and for mCherry (610/20BP, 600LB). Cultures were incubated in a water bath at 30°C without shaking. In 20 min time steps, duplicate samples were removed from the water bath, mixed vigorously, diluted in PBS and FACS analyzed. Gates for MATa, MATα and diploids were set by hand identifying the cell types as shown in Figure 6. As proof of the model prediction we performed a mating experiment with different cell densities. MATα TDH3pr-mCherry and MATa RPL9a-GFP BAR1 wild type cells were grown in SD medium to mid log phase. Cells were diluted in SD medium and cell numbers were adjusted to 10·106 cells/ml by measuring the cell number with a CasyTTC cell counter. MATα and MATa cells were mixed 1∶1 and diluted in SD medium in following concentrations: 10·106,5·106,2. 5·106,1·106,0. 5·106 cells/ml. 2 ml aliquots of the diluted cultures were incubated at 30°C in Petri dishes with a diameter of 36 mm (Falcon), in order to get an average monolayer of cells after sedimentation. In time steps of 15 min one Petri dish of each cell dilution was removed from the incubator, cells were re-suspended in the medium by intensive pipetting and 400 µl of the cultures were mixed with 400 µl PBS supplemented with CaliBRITE APC Beads (BD Biosciences #340487). The samples were analyzed by FACS. APC Beads were recorded with 640 nm excitation and 670/41BP filter, MATa Rpl9a-GFP and MATα mCherry reporter strains as mentioned above. APC beads were gated and used as internal standard. In each sample the number of cells corresponding to a fixed number of 90000 APC beads was analyzed, giving not only the relative amount of haploids and mating events but also the growth behavior of the components of the mixed culture (results of the experiment are shown in Figure 5). To analyze the influence of bar1Δ cheater cells in mating mixtures we used MATa Rpl9a-TagBFP2 bar1Δ reporter strain, together with the already introduced MATa and MATα reporter strains. The three strains were grown in SD media to mid logarithmic growth phase and diluted in SD media to 1·107 cells/ml. Several mating-with-cheaters-mixtures were prepared as shown in Table 4. The images were analyzed with CellID [36] to extract mating type, fluorescence activity, as well as position, size and shape of the cells. These data were transferred to a computational domain (see Figures S2 and S3 in Text S1 for details). From this computational domain a triangular mesh was generated using Gmsh [37] that can be used by various RD toolboxes. Here, we used the open source Toolbox DUNE [38] to solve the stationary as well as time-dependent equations by a finite element method with high accuracy [39], [40]. In Supporting Text S2 we systematically compare 2D and 3D simulations to account for the fact that images are taken in 2D, while diffusion and mating happen in 3D (see Figures S9, S10, S11 in Text S2). We found that the maximum difference between 2D and 3D in the simulated α-factor distribution was below 5%. Therefore, for the parameter fit the 2D solution was used, which resulted in a major speed-up.

Haploid budding yeast cells cannot actively move to find a mating partner, like some flagellated bacteria do. Instead they must grow a so-called shmoo – a mating projection – precisely into the direction of a potential partner. They communicate with each other by releasing pheromones into their environment, which are sensed by cells of the opposite mating type. This serves the localization of nearby cells and initiates growth arrest and mating. Paradoxically, yeast cells also secrete the protease Bar1 that destroys pheromones. To visualize the resulting pheromone distribution and understand the effect on mating efficiency, we combined fluorescence imaging and mathematical modeling. We observed that the controlled destruction of pheromones by the yeast cells is beneficial to communication since it causes relatively higher pheromone concentrations in areas where cells are dense and vanishing pheromone concentrations elsewhere. This allows the population to maintain two different cellular behaviors at the same time, i. e. mating and continued growth, a behavior which disappears when we genetically delete the gene for the pheromone-destroying protein.

Abstract Introduction Results Discussion Materials and Methods

systems biology fungi mycology microbial physiology biology and life sciences microbiology biophysics organisms yeast

2014

Yeast Mating and Image-Based Quantification of Spatial Pattern Formation

Varicella-zoster virus (VZV) is a human alphaherpesvirus that causes varicella (chickenpox) and herpes zoster (shingles). Like all herpesviruses, the VZV DNA genome is replicated in the nucleus and packaged into nucleocapsids that must egress across the nuclear membrane for incorporation into virus particles in the cytoplasm. Our recent work showed that VZV nucleocapsids are sequestered in nuclear cages formed from promyelocytic leukemia protein (PML) in vitro and in human dorsal root ganglia and skin xenografts in vivo. We sought a method to determine the three-dimensional (3D) distribution of nucleocapsids in the nuclei of herpesvirus-infected cells as well as the 3D shape, volume and ultrastructure of these unique PML subnuclear domains. Here we report the development of a novel 3D imaging and reconstruction strategy that we term Serial Section Array-Scanning Electron Microscopy (SSA-SEM) and its application to the analysis of VZV-infected cells and these nuclear PML cages. We show that SSA-SEM permits large volume imaging and 3D reconstruction at a resolution sufficient to localize, count and distinguish different types of VZV nucleocapsids and to visualize complete PML cages. This method allowed a quantitative determination of how many nucleocapsids can be sequestered within individual PML cages (sequestration capacity), what proportion of nucleocapsids are entrapped in single nuclei (sequestration efficiency) and revealed the ultrastructural detail of the PML cages. More than 98% of all nucleocapsids in reconstructed nuclear volumes were contained in PML cages and single PML cages sequestered up to 2,780 nucleocapsids, which were shown by electron tomography to be embedded and cross-linked by an filamentous electron-dense meshwork within these unique subnuclear domains. This SSA-SEM analysis extends our recent characterization of PML cages and provides a proof of concept for this new strategy to investigate events during virion assembly at the single cell level. Varicella-zoster virus (VZV) is an alphaherpesvirus that causes varicella (chickenpox) and herpes zoster (shingles) [1]. The host range of VZV is restricted to humans and its life cycle in the human host depends upon tropism for skin, lymphocytes and neurons in sensory ganglia, where it establishes latency [1], [2]. VZV pathogenesis can be investigated in vivo using xenografts of human dorsal root ganglia (DRG) and skin in a severe combined immunodeficiency (SCID) mouse model [3], [4]. Since VZV infectious particles are highly cell-associated, VZV spreads from cell to cell, accompanied by extensive cell-cell fusion and syncytia formation in vitro and polykaryocyte formation in DRG and skin in vivo [5]–[7]. All herpesviruses, and many other DNA viruses like adenoviruses, papillomaviruses or polyomaviruses, replicate in the host cell nucleus. During VZV infection, genome copies are synthesized in nuclear replication compartments and genomic DNA is packaged into icosahedral nucleocapsids formed by ORF40, the major capsid protein, and smaller capsid surface proteins, such as ORF23 protein. After assembly, nucleocapsids egress across the nuclear membrane for secondary envelopment in the cytoplasm and are then released as enveloped infectious virus particles [1], [8]. PML protein has many different isoforms and is a major organizing component of these nuclear domains, which vary in shape, size, and molecular composition. PML isoforms share a conserved N-terminus, which is involved in PML oligomerization and contains a characteristic RBCC/TRIM motif. Different PML isoforms have unique C-terminal domains, which may be important in isoform-dependent functions [9], [10]. PML-NBs have been implicated in controlling the replication of several alphaherpesviruses [11]–[21]. PML-NBs are targeted for disassembly in VZV-infected cells in vitro and in human epidermal cells of skin xenografts infected in vivo by a mechanism involving the interaction of SUMO-interacting domains (SIM) of the VZV immediate early protein ORF61 with sumoylated PML protein [21]. The targeting of PML-NBs for disassembly promotes VZV replication and spread in vivo in human skin xenografts and depletion of PML protein enhances VZV replication in cell culture, indicating a role for PML in the host cell defense [16], [21]. Whereas PML protein undergoes little degradation in VZV-infected cells, other alphaherpesviruses, including HSV-1, pseudorabies virus (PRV), bovine herpes virus type 1 (BHV-1) and equine herpesvirus type 1 (EHV-1) target PML for immediate proteosome-mediated degradation through functions of viral ICP0 ubiquitin ligase-like proteins, albeit with different degrees of efficiency [18]. HSV-1 appears to be most strongly regulated by PML isoforms I and II, based on their capacity to partially reverse the increase in plaque formation of an ICP0-null mutant observed in PML-depleted cells [17]. Recent work showed that HSV-1 ICP0 preferentially targets SUMO-modified isoforms of PML but also triggers PML I degradation independently of SUMO modification [19]. Interestingly, PML degradation also appears to be promoted by the US3 serine/threonine kinases of HSV-2 and PRV [20]. In the case of VZV, we found that if PML nuclear bodies are not dissociated effectively, these structures function to sequester VZV nucleocapsids in differentiated human cells within DRG and skin xenografts in vivo and in cultured cells [22]. Large ring-like PML-NBs created cages that contained nucleocapsids sequestered in the nuclei of neurons and satellite cells [22]. These PML cages in virus-infected cells resembled PML clastosomes, which sequester aberrant polyglutamine (polyQ) proteins, such as Huntingtin (Htt), in several neurodegenerative disorders [23], [24]. Thus, entrapment of VZV nuclecapsids may reflect a more basic cytoprotective function of PML in sensing and containing nuclear aggregates of aberrant proteins in a ‘nuclear safe house’, similar to the function of nuclear aggresomes [25], [26]. Further work demonstrated that of several PML isoforms tested, only PML IV promoted the sequestration of VZV nucleocapsids through an interaction with the ORF23 capsid surface protein, and that this process significantly inhibited VZV replication in vitro [22]. Quantitative immuno-electron microscopy analysis of ultrathin sections indicated that the majority (about 95%) of VZV nucleocapsids were found in PML cages, suggesting a surprisingly high efficiency of PML mediated capsid sequestration [22]. However, since ultrathin sections cannot reveal the shape and volume of PML cages, it was not possible to determine their sequestration capacity, that is, how many VZV nucleocapsids may be sequestered inside individual PML cages. Furthermore, because ultrathin (50–100 nm) cross-sections through a nucleus may represent <1% of the diameter of a typical mammalian cell nucleus the sequestration efficiency, defined as the proportion of all nucleocapsids present in a complete individual nucleus that are sequestered within PML cages, could not be determined. The goal of this study was to develop an EM imaging method with a high enough resolution to precisely identify, locate and count VZV nucleocapsids and at the same time, allow the efficient 3D reconstruction of large volumes of host cell nuclei, including complete PML cages. Here we describe a novel 3D imaging and reconstruction strategy that we term Serial Section Array-Scanning Electron Microscopy (SSA-SEM). Using this method together with electron tomography, we were able to create 3D reconstructions of complete nuclei of herpesvirus-infected cells and of PML cages with sequestered VZV nucleocapsids. Determining the shape and the volume of host cell nuclei and PML cages together with the precise 3D localization of several thousand VZV nucleocapsids enabled us for the first time to quantitatively estimate the sequestration capacity and efficiency of individual PML nuclear cages. The application of this strategy to resolve questions about PML-NB entrapment of VZV nucleocapsids is a proof of concept for its use to address other questions in virology and cell biology. As we have shown previously [22], PML cages in VZV-infected cells appear as ring-like structures that contain ORF23 capsid protein by confocal microscopy using antibodies to PML and ORF23, the small capsid protein (Figure 1A). At the higher resolution obtained by immunogold-TEM, mature (C-type capsids) and immature (A-and B-type capsids) can be identified that are embedded within and surrounded by densely immunogold-labeled PML positive material (Figure 1B). Next, we employed a high-contrast sample preparation protocol in order to be able to identify PML cages solely by their distinct morphology in samples not suitable for immunogold labeling. Similar to the densely labeled PML shell visible by immunoTEM (Figure 1B), a shell of amorphous electron dense material surrounding clusters of VZV nucleocapsids was visible in the high-contrast embedded samples (Figure 1C, green line). Sequestered mature C-type capsids and immature A-and B-type capsids could be distinguished clearly (Figure 1D). Importantly, the electron dense PML-positive shell surrounding sequestered VZV nucleocapsids was also visualized when the same sample was studied using a high-resolution scanning electron microscope (SEM) equipped with a field emission gun (FEG) and a back-scattered electron detector (BSE-detector) (Figure 1E) and the SEM resolution was sufficient to distinguish between mature and immature nucleocapsids (Figure 1F). Therefore, the distinctive morphological profiles of nuclear PML cages that contain sequestered nucleocapsids could be identified unequivocally by TEM as well as SEM. These results made it possible to perform the large volume and high-resolution 3D reconstruction of VZV-infected cell nuclei and PML cages aided by SEM imaging. VZV-infected cell nuclei with diameters of about 5–10 µm and PML cages of 0. 5-5 µm diameter [22] are too large to be fully reconstructed by conventional electron tomography approaches that usually use 100–300 nm sections. Although a serial ultrathin section approach in combination with TEM analysis could be used to reconstruct whole nuclei or cells, this approach has proven very time consuming and has several technical disadvantages. Since the area on one TEM grid is very small, several grids must used if imaging more than 10–20 serial sections is necessary to create a large volume reconstruction; TEM grids are easily damaged and damage to just one grid means that the whole series of sections before and after the missing grid cannot be used for the 3D reconstruction experiment. TEM sections are also prone to folding when placed on the grid, which causes distortions in the 3D reconstruction. In contrast, large samples fit into the microscope for SEM and long ribbons of ultrathin sections can be deposited on glass slides. We developed SSA-SEM as a method that provided both a high enough resolution to identify and precisely locate virion capsids and at the same time allowed the efficient 3D reconstruction of large volumes of host cell nuclei and complete PML cages (Figure 2). SSA-SEM combines principles and strategies of related methods such as immunofluorescence (IF) array tomography [27], [28], serial block face-SEM [29] and focus ion beam (FIB) or iron abrasion SEM [30], [31]. Ribbons of ultrathin serial sections were acquired by ultramicrotomy. We used 100 nm sections to avoid double counting of VZV nucleocapsids, which have a diameter of approximately 100 nm, in consecutive sections. Ribbons of serial sections were transferred onto gelatin-coated glass-slides (Figure 2A), followed by heavy metal counterstaining and a final carbon coating step to avoid charging during SEM imaging. The serial section array was then imaged with a high-resolution SEM using a BSE detector, which generates TEM-like images of cell structures with a contrast dependent mainly on the high atomic weight and differential adsorption of heavy metal stains to cellular proteins, membranes and nucleic acids (Figure 2B). Consecutive SEM imaging of serial sections created ordered stacks of unaligned digital images (Figure 2C). These stacks were then computationally aligned (Figure 2D). The aligned images were then segmented by manual or automatic (threshold) tracing of the morphology of structures of interest, e. g. nucleocapsids and PML cages (Figure 2E). From this data, a 3D model was generated that shows the shape of PML cages and the distribution of virion capsids within the reconstructed nuclear volume (Figure 2F). Using SSA-SEM we first analyzed a VZV-infected melanoma cell nucleus in which endogenous PML was expressed (Figure 3A–E and Video S1). The shape of the infected cell nucleus and the nuclear volume were determined by tracing the outer boundary of the nucleus in all 50 consecutive sections, encompassing a total thickness of about five microns and a nuclear volume of about 95 µm3. The 3D reconstruction revealed an irregular shape of the nucleus characterized by several indentations and deep invaginations (Figure 3B and Video S2). If visualization was limited to the original two-dimensional sections, these invaginations might be misinterpreted as ‘vesicles’ or ‘vacuoles’ within the nuclear matrix (Figure 3A and Video S1). Morphological tracing and 3D modeling revealed the location and distribution of the electron dense heterochromatin, which is located primarily at the periphery of the nucleus (Figure 3C–E, blue); also seen is the nucleolus in the lower center of the nucleus (Figure 3C–E, brown) and the mature and immature nucleocapsids (Figure 3C–E, red and yellow spheres, respectively). 3,467 (82%) immature capsids and 756 (18%) mature capsids were identified within the serial sections and their positions were precisely modeled in the reconstructed nuclear volume (Figure 3C–E and Video S2). This work revealed that mature and immature capsids were not segregated into different nuclear domains; instead, they were mixed randomly and were evenly distributed within the nuclear volume outside of the heterochromatin and the nucleolus and were excluded from the deep nuclear imaginations (Figure 3E and Video S2). These SSA-SEM results were confirmed by two more 3D models of large volumes of VZV-infected cell nuclei, which were derived by morphological segmentation of 18 consecutive TEM sections of 100 nm thickness (Figures 3F and 3G; Video S3). In the reconstructed nuclear volume in Figure 3F, which accounted for 46. 2 µm3,109 (25. 65%) mature and 316 (74. 4%) immature nucleocapsids were identified and 102 (7. 6%) mature and 1,238 (92. 4%) immature capsids were identified in the nuclear volume in Figure 3G (43. 4 µm3) (Figure 3G and Video S3). Similar to the nucleus in Figure 3A–E, most nucleocapsids were distributed evenly throughout the reconstructed nuclear volume; no extended clusters of aggregated nucleocapsids were visible. The quantifications of structures shown in Figure 3 are summarized in Table 1. We next used SSA-SEM to analyze VZV-infected melanoma cells that express PML IV when induced with doxycycline, together with endogenous PML [22]. Inducing PML IV creates conditions that allow enough PML cages to persist in VZV-infected cells for 3D ultrastructural analysis. Using SSA-SEM we first identified a typical VZV syncytium in which infected cells are fused into a polykaryon (Figure 4A, left panel). A nucleus that contained two distinct electron dense PML cages with numerous sequestered VZV nucleocapsids was identified within the syncytium (Figure 4A, middle and right panel). Next, 18 consecutive 100 nm serial sections through this nucleus were imaged by SSA-SEM and then traced and segmented as illustrated in Figure 2. Inspecting different sections in the series (Figure 4B and Video S4) suggested a spherical shape of the PML cages. Tracing and 3D modeling (Figure 4C–F and Video S5) of the electron dense heterochromatin (blue), all nucleocapsids (yellow spheres) and mature capsids (only in Figure 4E and F, orange), and the outer surface of the electron dense shell of the PML cages (green) was performed with all 18 sections (stack thickness approximately 1. 8 microns). The 3D model revealed that only six (0. 2%) of a total of 3,062 nucleocapsids located within the reconstructed nuclear volume (63 µm3) were not aggregated together with the other nucleocapsids (yellow spheres) (Figure 4C and D). 3,056 nucleocapsids (99. 8%) were in clusters enclosed by the two PML cages (green) present in this nuclear volume. Interestingly, each of the PML cages, whose reconstructed volumes were about 6. 2 µm3 (upper cage, Figure 4D) and 4. 6 µm3 (bottom cage in Figure 4D) contained more than a thousand nucleocapsids: 1,732 and 1,324, respectively. This information made it possible to estimate the packing density of capsids within the two PML cages, the mean of which was 284 nucleocapsids/µm3. A 3D model of the upper PML cage at higher magnification identified the position of both mature and immature capsids and revealed that both types were randomly packed within the PML cage (Figure 4E–F). Of note, both PML cages were associated with electron dense heterochromatin (blue) in the periphery of the nucleus (Figure 4D). These results encouraged us to attempt a 3D reconstruction of the complete volume of a VZV infected cell nucleus in order to visualize and quantify the shape, location, size and number of all PML cages and capsids present. We succeeded in imaging a ribbon of 82 consecutive serial sections (100 nm thickness) through a nucleus (Figure 5A and Video S6). Both the first and last sections contained large areas of heterochromatin, indicating that these sections were cut through the nuclear periphery at the top or bottom of the nucleus, respectively. Therefore, we estimate that an almost complete nuclear volume is represented in this stack of serial sections and in the 3D reconstruction. 3D modeling of the shape of the nucleus from tracing the outer boundary of the nucleus on each section revealed an irregular surface with a wide valley-like indentation (Figure 5B and E). Inspection of the original SSA-SEM images of the serial sections through this nucleus revealed that this wide indentation was directly adjacent to a very prominent ER network in the cytoplasm (Figure 5A, red arrow and Video S6). The total reconstructed nuclear volume was about 291 µm3 and contained four distinct PML cages (Figure 5A, black arrows 1–4) of irregularly globular shapes (solid green) (Figure 5C and F) of very different sizes and with volumes that ranged from about 0. 8–10 µm3 (Table 1). 5,597 nucleocapsids were traced and precisely localized within the reconstructed volume; 5,527 (98. 7%) were sequestered within the PML cages (yellow spheres) (Figure 5D and G) and only 70 (1. 3%) nucleocapsids (red spheres) were outside of PML cages (Figure 5C–I) (Table 1). Therefore, this comprehensive large volume nuclear reconstruction (Video S7) proved that PML cages are extremely efficient in reorganizing and sequestering thousands of VZV nucleocapsids. Depending on their size, individual PML domains were found to sequester from about 126 capsids to more than 2,700 nucleocapsids with an average packing density of 249±64 SD/µm3 (N = 4) (see also Table 1). Again, the four PML cages were found in the periphery of the nucleus associated with the electron dense heterochromatin (Figure 5A, H, I and Video S8). The electron density of the PML positive shell of nuclear PML cages allowed tracing and reconstruction of the shape (3D surface view) of this compartment by SSA-SEM; however the 3D distribution of PML protein within PML cages was not revealed using this approach. Therefore we used a serial section immunoTEM (ss-immunoTEM) approach to investigate quantitatively and in three dimensions how PML protein is distributed within the shell and in the core of the PML cages, where the nucleocapsids are entrapped. Seven consecutive sections (100 nm) through HPF/FS-treated and LRwhite embedded cells that contained PML cages with entrapped VZV capsids, were labeled with a PML specific antibody and Protein A conjugated to 15 nm gold particles, and then imaged by TEM (Figure 6A). The results of tracing and modeling of the PML labeling (small green spheres), mature capsids (red spheres) and immature capsids (yellow spheres) and the electron dense heterochromatin (blue) are shown in Figure 6B, C and Video S9. About 5,219 PML gold particles, 63 mature capsids and 403 immature capsids were identified; 272 of the entrapped nucleocapsids were directly associated with PML gold particles (half-green spheres). The 3D reconstruction clearly reveals a ring-shaped ‘cloud’ of dense PML-labeling that corresponds to the electron dense shell of PML cages as seen in the high-contrast embedded samples analyzed by SSA-SEM before. Significant amounts of PML gold labeling were also found in the core of the PML cage (Figure 6A, right panel and Figure 6C) where 58% of the entrapped nucleocapsids were directly associated with PML gold particles (half-green spheres). Therefore PML protein is not only a structural component of the electron dense shell of PML cages but also binds to nucleocapsids entrapped within the core of the PML cages. The observation by ss-immunoTEM that PML protein was present in the shell and in the center of PML cages, where it was found directly associated with many VZV capsids, suggested that PML protein is not only a structural component of the electron dense shell of PML cages, but may also be involved in the immobilization or cross-linking of sequestered VZV nucleocapsids. To address this hypothesis, we investigated the ultrastructure of PML cages and of sequestered nucleocapsids by electron tomography, which provided a higher resolution than SSA-SEM, albeit at the cost of allowing analysis of only a much smaller (thinner) sample volume. The samples for tomography consisted of HPF/FS-treated and epoxy resin-embedded VZV infected melanoma cells that expressed PML IV together with endogenous PML [22]. We first recorded dual-axis tomograms from 80 nm sections of VZV infected cell nuclei with PML cages (Figure 7A–E). The 3D models were generated by analyzing digital tomogram slices as was done for SSA-SEM, combining manual tracing and automatic threshold-based tracing. EM tomography revealed that all nucleocapsids within PML cages were embedded in an irregular electron dense meshwork with numerous fibrous structures emanating from the nucleocapsids and often cross-linking adjacent capsids (Figure 7A and Video S10). These irregular fibrils were even better visible when the contrast was inverted (Figure 7B, white arrows) and were then traced automatically by applying a threshold (green outline) (Figure 7C) in order to reconstruct a 3D model of the irregular meshwork (green) within PML domains (Figure 7D, E and Video S11). The 3D volume information of tomograms from 80 nm sections is very limited because of the small z-dimension of the section. In order to reveal the precise arrangement and packing of nucleocapsids within the center of the PML cages and to confirm the presence of an irregular electron dense meshwork entrapping VZV nucleocapsids, we next recorded dual-axis tomograms from 300 nm thick sections through PML cages. A volume view representation of a representative tomogram (Figure 7F and Video S12) and an ortho-slice view of the same volume (Figure 7G and Video S12) shows the packing of nucleocapsids in several layers and that, in contrast to paracrystalline inclusion bodies of nucleocapsids observed in some HSV-infected cells [22], those entrapped in PML cages were rather loosely configured, were usually not in direct contact, and the space between them was filled with an irregular electron dense meshwork and fibers. Threshold-aided tracing and 3D reconstructions of the irregular meshwork (green) (Figure 7J–K), and of mature (red) and immature (yellow) VZV nucleocapsids showed that all traced capsids were tightly associated with the irregular meshwork that also cross-linked adjacent capsids (Figure 7I–K and Videos S13 and S14). This cross-linking of adjacent capsids was also visible in the original digital tomogram slices (green arrows) (Figure 7L) and confirmed our observations from the 80 nm tomography reconstructions. In this work, we developed Serial Section Array-Scanning Electron Microscopy (SSA-SEM), a novel three-dimensional (3D) imaging and reconstruction strategy, and applied the technique to the analysis of VZV-infected cells. Using SSA-SEM and EM tomography, we have reconstructed the nuclei of host cells infected with this representative herpesvirus and, for the first time, revealed the numbers and precise location of thousands of VZV nucleocapsids, visualized the 3D shape and ultrastructure of nuclear PML cages that entrap nucleocapsids, and provided quantitative estimates of the volume, sequestration efficiency and sequestration capacity of these PML cages. The large volume reconstruction of nuclei in VZV-infected cells also provided basic information on how VZV infection affects the shape of the host cell nuclei and how subnuclear domains like electron dense heterochromatin or PML cages and nucleocapsids are spatially related. Of interest, our 3D analysis revealed that PML cages with entrapped capsids were consistently located at the periphery of the nucleus and associated with domains of electron dense heterochromatin, suggesting that the formation of PML cages and VZV capsid sequestration are initiated adjacent to these domains. Our experimental challenge, which has many similarities to obstacles encountered in addressing other virology and cell biology questions, consisted in how to combine an efficient approach for the large volume 3D reconstruction of infected cell nuclei and complete PML cages with the high ultrastructural resolution necessary to localize VZV nucleocapsids and differentiate mature from immature capsids. Infected cell nuclei have diameters of about 5–10 µm and PML cages are about 0. 5–5 µm [22]. These structures are about one order of magnitude too large to be readily reconstructed by conventional electron tomography approaches that usually use 100–300 nm sections. Recent technical and computational improvements have enabled some specialized laboratories to apply serial-sectioning tomography for the reconstruction of large organelles and even complete cells by merging individual tomograms from consecutive sections into a single large volume reconstruction [32]. However, this approach is very labor-intensive so that only a few 3D reconstructions can be generated and this limitation may raise questions about whether these models are fully representative of the structures of interest. SSA-SEM combines a sample preparation strategy (serial section arrays) similar to the method used in immunofluorescence (IF) array tomography with imaging and detection principles (high resolution SEM with back scattered electron detection) that have been used in serial block face (SBF) -SEM or focus ion beam (FIB) -SEM [27]–[29], [31], [33]. In principle, the latter two methods could also be used to analyze herpesvirus-infected cell nuclei or PML cages. In fact, Feierbach et al. used SBF-SEM to locate structures reminiscent of actin filaments and nucleocapsids in cells infected with HSV-1 and PRV, which have caspids that are similar to VZV capsids in size and shape [34]. Bennett et al. used FIB-SEM to locate human immunodeficiency virus (HIV) particles in surface-connected tubular conduits in HIV-infected macrophages [30]. However, these approaches require highly specialized equipment that may not be readily accessible. Most importantly, these techniques are destructive imaging methods that destroy the sample block during image stack acquisition by step wise FIB-milling or cutting the sample surface to allow successive surface imaging at different sample levels, while discarding the serially-cut sections. SBF-SEM and FIB-SEM may therefore not be ideal for valuable samples that are difficult to obtain or to prepare. In SSA-SEM, serial sections are secured on a glass slide, creating stable arrays that can be stored and imaged repeatedly, allowing the acquisition of several image series of the same sample at different magnification, resolutions, contrast modes or with different equipment. A major advantage of SSA-SEM is that the interior of cells and tissue become exposed at the section surface, enabling the use of immuno-histochemistry protocols to localize proteins or nuclei acids within the context of the 3D ultrastructure of cells or tissues. Our 3D reconstructions confirmed that most PML nuclear bodies in VZV infected cells expressing endogenous PML are disassembled efficiently during the course of infection. This process involves the interaction of SUMO-interacting domains (SIM) of the VZV ORF61 protein with sumoylated PML [21]. As a result, most of the several thousand VZV nucleocapsids that were produced in VZV-infected cells appeared randomly distributed in the reconstructed nuclear volume when examined by SSA-SM. As noted, other alphaherpesviruses disrupt PML nuclear bodies and in most cases, also eliminate PML protein by rapid ICP0-mediated degradation [18], [19]. However, when PML disassembly is incomplete, as it is in VZV-infected cells in skin and neural cells in vivo, nucleocapsids become sequestered in PML cages. Systematic random sampling analysis of hundreds of ultrathin sections through different PML cages suggested that >95% of all types of VZV nucleocapsids (A, B and C-type) were efficiently sequestered in PML cages [22]. Nevertheless, random ultrathin sections do not reveal the 3D shape and volume of single PML cages because these sections (50–100 nm) may encompass only 1–10% of the diameter of PML cages. Therefore, techniques used in the earlier study did not allow an assessment of the size, volume and shape of PML cages or how many VZV nucleocapsids may be sequestered within individual PML cages. Furthermore, since ultrathin cross-sections through a nucleus encompass only a very small fraction of the nuclear volume, the sequestration efficiency of PML cages could not be determined for single nuclei. These limitations were addressed by using SSA-SEM to reconstruct the shape and volume of individual PML cages, which demonstrated that up to several thousand (2,780) nucleocapsids can be sequestered by single PML cages. Furthermore, quantitative analysis of several thousand nucleocapsids in reconstructed volumes of single nuclei showed that more than 98% of all capsids could become entrapped in PML cages, proving their very high sequestration capacity and explaining the antiviral activity of PML IV [22]. Our method to estimate the sequestration capacity and efficiency of PML cages made it possible to provide information beyond just a morphological description and demonstrates that SSA-SEM can be used in quantitative analyses of virus interactions with nuclear structures. Given the high sequestration capacity of PML cages, now established by single 3D nuclear analysis and by quantitative random sampling analysis of hundreds of ultrathin cross-sections, it is somewhat surprising that infectious VZV titers were reduced only by about 50% in cell lines expressing PML IV [22]. These results indicate that only very few VZV infectious particles are needed to successfully enter and replicate in adjacent cells. This explanation is consistent with the observation that only very few PRV genomes are required to establish nuclear replication compartments and initiate productive replication, as shown using recombinant PRV, which is also an alphaherpesvirus, carrying a Brainbow cassette [35]. VZV does not release virus particles into the supernatant in cell culture and spreads only from cell to cell by a mechanism that may be facilitated by extensive syncytia formation [5], [7]; therefore, even the few nucleocapsids that may escape sequestration in PML cages should be sufficient to infect adjacent cells in vitro. In contrast, in the human host, VZV must infect complex tissues and overcome the barriers of intrinsic and adaptive immunity, which is likely to depend on production of larger numbers of infectious virus particles. Therefore the PML-mediated nuclear sequestration of many VZV capsids observed in human skin or DRG may be expected to have a more substantial antiviral effect [22]. The quantitative analysis of the different types of capsids present within infected cell nuclei revealed that the majority (70–90%) were immature (A and B-type capsids) while only a minority was in a mature stage (C-type, 10–30%). We speculate that large numbers of immature nucleocapsids help to outcompete the limited sequestration capacity of PML cages, giving mature capsids a better chance to egress from the nucleus. These observations also suggest that the relatively few mature virions observed in VZV infected cells in vitro is not just a tissue culture phenomenon. Using conventional EM tomography, we obtained the first insights about the 3D ultrastructure of PML cages, suggesting how VZV nucleocapsids may be kept entrapped in these nuclear domains. Tomographic 3D reconstructions revealed the presence of an electron dense meshwork surrounding sequestered nucleocapsids and fiber-like like structures, that often cross-linked adjacent nucleocapsids, suggesting that capsids were entrapped by restricting their mobility and ‘gluing’ them together. The 3D analysis of PML-labeled sections by serial section immunoTEM showed that PML protein was present both in the periphery of the cage (the ‘shell’) and associated with the capsids entrapped in the center of PML cages. PML protein which is the main structural component of PML nuclear bodies, forms homo-and heterooligomers [10]; therefore at least part of the electron dense meshwork is likely to consist of PML-oligomers that crosslink and embed capsids in a protein meshwork. The PML-positive meshwork and fibers were in general directly associated with the edges of VZV capsids, which is consistent with our previous biochemical data that demonstrated an interaction of PML with the small outer capsid protein ORF23 [22]. Many other proteins resident in PML-nuclear bodies, e. g. hDaxx or Sp100, may be part of this meshwork [9]. Cryo-tomography is an alternative that would enable a 3D reconstruction of PML cages at even higher resolution and more native conditions (avoiding resin embedding and heavy metal staining) but this approach can be predicted to encounter major experimental challenges. Frozen-hydrated sections through cell nuclei will be required, the identification of PML cages will be very demanding as these structures are not abundant, and identification would need to occur at a very low electron dose that creates noisy imaging conditions. Molecular docking of known crystal structures to electron densities will also be difficult, because of the many other proteins present in PML nuclear domain and currently only the PML RING domain has been crystallized [36], whereas the PML IV C-terminal domain is critical for nucleocapsid sequestration [22]. In summary, we were able to create complete reconstructions of herpesvirus-infected cell nuclei and PML nuclear domains in three dimensions for the first time using 3D SSA-SEM and EM tomography. This study supports and extends our recent discovery and characterization of PML cages that efficiently sequester VZV nucleocapsids in cell culture and in differentiated human skin and neural cells infected in vivo and represents a novel antiviral mechanism, distinct from the established role of PML in controlling several alphaherpesviruses shortly after virus entry by limiting early viral gene transcription. Visualization of the shape and measurements of the volumes of host cell nuclei and PML cages together with the 3D localization of VZV nucleocapsids with ultrastructural precision enabled us to determine the sequestration efficiency and capacity of PML nuclear cages. This work contributes not only to a more comprehensive understanding of the antiviral activity of PML cages against VZV, a pathogenic human herpesvirus, but also provides a novel method to undertake the 3D reconstruction and quantitative investigation of nuclear PML domains that have also been found to be associated with capsids of papillomaviruses and polyomaviruses [37]–[39]. The method has broad relevance for addressing other questions in virology and cell biology where large volume 3D reconstruction with high precision imaging of intracellular structures is needed. The human melanoma cell line (MeWo, ATCC number: HTB-65) was grown in Dulbecco' s modified Eagle' s medium supplemented with 10% fetal bovine serum, nonessential amino acids (100 µM) and antibiotics (penicillin at 100 U/ml and streptomycin at 100 µg/ml). Melanoma cells were passaged fewer than 25 times. Melanoma cells expressing doxycyline-inducible PML IV were constructed using the pRetro-X-Tet-On-Advanced vector system and pRetro-X-Tight-Pur plasmid (Clontech Laboratories) with the PML IV plasmid pcDNA3-PML IV as described recently [22]. The stable cells were induced with 5 µg/ml doxycyline for 24 hr before infection with VZV. The virus was recombinant Oka (rOka) derived from the wild type low passage parent Oka strain (pOka). Viral infection was done with cell-associated VZV at a ratio of 1/20 (infected cells/uninfected cells) for 48 hr. Cultured cells on glass coverslips were fixed in 4% paraformaldehyde in PBS for 20 min at room temperature. Cells were blocked and immunostained as described previously [5]. Antibodies used for confocal microscopy were: mouse monoclonal anti-PML (PG-M3) from Santa Cruz Biotech and rabbit polyclonal anti-VZV-ORF23 described previously [5], [22]. Secondary antibodies were Alexa Fluor 488 and Alexa Fluor 594 conjugated donkey anti-mouse or donkey anti-rabbit antibodies (Invitrogen). Infected cultured cells were imaged using a Leica TCSSP2 confocal laser scanning microscope (Heidelberg, Germany). Microscope objectives were 40×/1. 0 (Numerical Aperture, N. A.) or a 63×/1. 4 (N. A.) Plan Apochromat objectives. Images were scanned at 1024×1024 pixels with at least four times frame averaging and the pinhole adjusted to one airy unit. Brightness and contrast were adjusted using Photoshop CS3 (Adobe) or iPhoto (Apple). For standard TEM, samples were fixed in 4% paraformaldehyde and 2% glutaraldehyde in 0. 1 M phosphate buffer (ph 7. 2) and embedded in epoxy-resin. For immuno-TEM or EM tomography samples were fixed in 4% paraformaldehyde and 0. 1% glutaraldehyde in 0. 1 M phosphate buffer (ph 7. 2) and then high-pressure frozen (HPF) in a Leica EM PACT2 and freeze substituted (FS) in either LR-white or Epoxy resin (Embed812), respectively. Frozen specimen carriers with cells were placed into frozen cryovials containing acetone with 0. 1% glutaraldehyde and 0. 1% uranyl acetate (for LR White embedding) or in acetone with 1% osmium tetroxide and 0. 1% uranyl acetate (for Epon embedding). The frozen vials were then placed into a Leica AFS for the freeze-substitution procedure and then embedded in either or LR-White resin for immuno-TEM or epoxy resin Embed 812 for EM tomography. Sections (80–300 nm) were prepared with a diamond knife (Diatome) using an ultramicrotome (Ultracut, Leica). For immunogold-labeling LRwhite sections were pre-blocked in DIG-blocking solution (Roche) for 30 min. Primary antibodies and Protein A-gold particles (obtained from CMC, Utrecht, the Netherlands) were diluted in blocking solution and sections were incubated for 1 h or 30 min, respectively, at RT. Rabbit polyclonal anti-PML antibody (Santa Cruz Biotech) was used at 1∶10 dilution. Sections were stained with 3. 5% aqueous uranyl acetate for 15 minutes and with 0. 2% lead citrate for one minute and air-dried. Sections were analyzed using a JEOL 1230 transmission electron microscope (TEM) at 80 kV and digital photographs were captured with a GATAN Multiscan 701 digital camera. VZV-infected cells were fixed in 4% paraformaldehyde/2% glutaraldehyde in 0. 1 M phosphate buffer (pH 7. 2) for 24 hr and cell pellets were stabilized by embedding in 10% gelatine. The samples were washed with ultrapure water and then postfixed with 2% osmium tetroxide reduced with 1. 5% (w/v) potassium ferrocyanide for 2 hr at room temperature. The samples were then washed again, followed by one hour incubation with 1% (w/v) tannic acid, washing in ultrapure water and a final “en block” staining with 3. 5% (w/v) uranyl acetate over night. The samples were then dehydrated in a series of ascending ethanol concentrations (30%–100%), treated with propylene oxide and finally embedded in epoxy resin Embed 812 (Electron Microscopy Sciences, Inc.). The procedure for the preparation of Serial Section Arrays (SSA) was similar to the method described for fluorescence array tomography [27], [33], [40]: serial sections (100 nm thickness) were cut with a jumbo histo diamond knife (Diatome) and collected onto precleaned glass slides that had been coated with a solution of 0. 3% gelatine with 0. 1 g/l chromium potassium sulphate. To enable SEM imaging, the SSAs were finally counter stained with uranyl acetate and lead citrate and then heavily carbon coated (two cycles of 30 seconds until the surface color was light brown) using a Benchtop Turbo III apparatus (Denton Vacuum, LLC) and attached to SEM stubs using colloidal graphite or adhesive copper tape (both from EMS, Inc). The arrays were first pre-scanned with a Hitachi S-3400N VP-SEM to assess the quality of the ribbons of serial sections and to find regions of interest (ROIs). ROIs were mapped at a magnification from 100× (whole section image) to 10,000× (image of cell nucleus with resolved capsids) using an accelerating voltage of 10 kV, a working distance of 8. 5 mm and the back-scattered electron (BSE) detector. For the final acquisition of high-resolution digital image stacks from serial sections, the arrays were transferred to a Zeiss Sigma FE-SEM that is equipped with a field emission gun (FE). The mapped ROIs were identified and then imaged using the BSE detector at magnifications from 5,000–40,000× with an accelerating voltage from 6 kV–10 kV and a working distance from 6–7 mm; images were scanned at 2048×1536 pixels, with at least 2 times line averaging. SSA-SEM image stacks were automatically aligned (registration) in rigid mode using the ‘StackReg’ plugin in the Fiji/ImageJ software package (http: //fiji. sc/wiki/index. php/Fiji). Aligned images were saved as image sequence files and then imported into the 3D reconstruction program ‘Reconstruct’ (http: //synapses. clm. utexas. edu/tools/reconstruct/reconstruct. stm) [41]. Segmentation of cells and infected nuclei was accomplished by manual classification and tracing the boundary contours of ultrastructures of interest. The visualization of the 3D shape of cell nuclei (grey), heterochromatin (blue), protein aggregates (brown) and PML domains (green) was achieved by representing the traces of these objects as Boissonnat surfaces using “Reconstruct” [41]. Cross-sectioned nucleocapsids, which are rotation-symmetric icosahedral structures, were traced and modeled as 3D spheres with 100 nm diameters. A section thickness of 100 nm was deliberately chosen to avoid double counting of the same capsids (which have a diameter of about 100 nm) in consecutive sections. Because of the anisotropic resolution in the serial section reconstructions the position of VZV nucleocapsids is more precisely modeled in the x–y dimension than in the z-dimension (100–200 nm resolution). Volumes of reconstructed nuclei and of PML cages were calculated and the number of VZV capsids and PML gold particles were counted using the corresponding traces in the ‘Reconstruct” software. The 3D models were saved as 360° image series and then exported into Fiji/ImageJ, where the files were compressed (JPEG) and saved as movie files (. avi). VZV infected cells were fixed and high pressure frozen (HPF), freeze substituted (FS) and embedded in epoxy resin (Embed812) as described above (TEM sample preparation). 80 nm or 300 nm sections were cut with a diamond knife (Diatome) using an ultramicrotome (Ultracut, Leica) and placed on Formvar and carbon coated 75mesh TEM copper grids (TedPella). Sections were stained with 3. 5% aqueous uranyl acetate for 15 minutes and with 0. 2% lead citrate for one minute and air-dried. Finally, the sections were coated on both sides of the grid with 15 nm colloidal gold particles (Ted Pella) as fiducial markers by repeated dipping of the grids in the colloidal gold solution followed by air drying. The 80 nm thick sections were imaged on a JEOL 1400 TEM (JEOL USA, Inc.) at 120 kV equipped with a dual-axis tomography holder. The double-tilt series were recorded with the ‘SerialEM’ software package (http: //bio3d. colorado. edu/SerialEM/) using a tilt range of ±65° at 1. 5° angular increments [42]. The image pixel size ranged from 0. 43–1. 3 nm. The 300 nm thick sections were imaged on a Titan ETEM (FEI company, USA) operated at 300 kV using a dual-axis tomography holder. Double tilt series were recorded with a tilt range of ±65° at 1. 5° angular increments using the Xplore3D software (FEI). Image pixel size was 1. 4 nm at the specimen level. The series of dual-axis tilt images were aligned, reconstructed by weighted back-projection and then merged into dual-axis tomograms using the software package IMOD 4. 1 (http: //bio3d. colorado. edu/imod/) [43], [44]. The stack of digital tomogram slices (. rec file) was imported into Fiji/ImageJ and saved as image sequence file that was then imported into the ‘Reconstruct’ software for tomogram segmentation and 3D modeling. VZV capsids were traced manually and visualized using Boissonnat surfaces. The electron dense meshwork within the PML cages was traced automatically applying a threshold and the ‘wild fire’ tool in the ‘Reconstruct’ software; 3D visualization of the meshwork was also achieved using Boissonnat surfaces [41]. Volume views and ortho-slice views of the tomograms were generated by importing the digital tomogram slices (. rec file) into Fiji/ImageJ and applying the ‘volume viewer’ plug-in. Graph Pad Prism (version 5. 0) statistical software was used for quantification and statistical analysis.

Varicella-zoster virus (VZV), the cause of varicella and zoster, is a human herpesvirus that replicates in the host cell nucleus where viral genomes are packaged into virion nucleocapsids. We have recently identified antiviral PML (promyelocytic leukemia) nuclear cages that sequester VZV nucleocapsids and inhibit formation of infectious particles. Here we developed a novel three-dimensional (3D) imaging and reconstruction strategy, termed Serial Section Array-Scanning Electron Microscopy (SSA-SEM) that together with electron tomography made it possible to derive 3D reconstructions of complete herpesvirus infected host cell nuclei and of PML cages with ultrastructural precision for the first time. We determined the 3D distribution of several thousand nucleocapsids within reconstructed volumes of single host cell nuclei and in PML cages as well as their sequestration efficiency and sequestration capacity: more than 98% of nucleocapsids were entrapped within PML cages and individual PML cages could sequester nearly 3,000 nucleocapsids which were cross-linked by an irregular electron-dense meshwork within the PML cages. This 3D analysis provides a proof of concept for using SSA-SEM to investigate virion assembly at the whole cell level and further elucidates our observation that PML cages are antiviral nuclear domains which block VZV nucleocapsid egress from the infected cell nucleus.

Abstract Introduction Results Discussion Materials and Methods

virology biology microbiology host-pathogen interaction

2012

3D Reconstruction of VZV Infected Cell Nuclei and PML Nuclear Cages by Serial Section Array Scanning Electron Microscopy and Electron Tomography

Organogenesis relies on the spatiotemporal balancing of differentiation and proliferation driven by an expanding pool of progenitor cells. In the mouse pancreas, lineage tracing at the population level has shown that the expanding pancreas progenitors can initially give rise to all endocrine, ductal, and acinar cells but become bipotent by embryonic day 13. 5, giving rise to endocrine cells and ductal cells. However, the dynamics of individual progenitors balancing self-renewal and lineage-specific differentiation has never been described. Using three-dimensional live imaging and in vivo clonal analysis, we reveal the contribution of individual cells to the global behaviour and demonstrate three modes of progenitor divisions: symmetric renewing, symmetric endocrinogenic, and asymmetric generating a progenitor and an endocrine progenitor. Quantitative analysis shows that the endocrine differentiation process is consistent with a simple model of cell cycle–dependent stochastic priming of progenitors to endocrine fate. The findings provide insights to define control parameters to optimize the generation of β-cells in vitro. The pancreas is an organ performing vital exocrine and endocrine roles in nutrient metabolism and glucose homeostasis. In the mouse, multipotent pancreatic progenitor cells (MPCs) emerge from the endoderm around embryonic day 9. 0 (E9. 0) [1]. This population, characterized by the expression of transcription factors PDX1 (GenBank NP_032840), SOX9 (GenBank NP_035578), and HNF1B (GenBank AAH25189), eventually gives rise to all three major cell lineages of the pancreas: endocrine, acinar, and ductal [2–4]. Following early progenitor expansion, three-dimensional (3-D) organization of the pancreatic epithelium leads to the generation of an apico-basally polarized [5–7], branched tubular network. By E13. 5, it exhibits its final functional compartmentalization: the distal tip domains give rise to the acinar cells of the exocrine lineage [8], whereas the SOX9+/HNF1B+ proximal trunk domain is bipotent at the population level, giving rise to the ductal and endocrine cells [3]. The endocrine lineage arises from transient NEUROG3+ (GenBank AAI04328. 1) endocrine progenitors, as demonstrated by lineage tracing studies [2] and the absence of all pancreatic endocrine cells in Neurog3−/− mice [9]. NEUROG3+ endocrine progenitors originate from pancreatic progenitors expressing PDX1/SOX9/HNF1B during the early phases of MPC expansion and during the secondary transition spanning E12. 5–15. 5, with specific endocrine subtypes being specified during discrete time windows [10]. Whereas the majority of NEUROG3+ endocrine progenitors are post-mitotic [11] and unipotent, giving rise to only one endocrine subtype [12], we do not know whether individual PDX1/SOX9/HNF1B pancreatic progenitors give rise to both ductal and endocrine cells or are heterogeneous, encompassing cells with pre-specified lineage-restricted potentials. In this study, we ask how individual pancreas progenitors contribute to the population dynamics enabling organ expansion and endocrine differentiation. Over the last few years cell-labelling and tracing methods have brought forth quantitative descriptions of cell differentiation. In homeostatic systems, for instance, the maintenance of a hierarchy of stem and differentiating cells can be accounted for by populations of equipotent progenitors exhibiting probabilistic fate choices [13–15]. An attempt to extrapolate these notions to developing systems has encountered some difficulties because, in these instances, the growth of the tissue needs to be taken into consideration. Notwithstanding these complications, lineage analysis of progenitor cells in the vertebrate retina indicates that, similarly to the abovementioned homeostatic systems, the distribution of clone sizes is compatible with a model in which progenitors stochastically divide in three modes: (1) symmetric self-renewing, (2) asymmetric, and (3) symmetric differentiating divisions [16–20]. Contrary to homeostatic systems, however, the probabilities of each division mode are not assumed to be fixed but to vary over time, following phases of proliferation and differentiation. These models have proven successful in explaining the distributions of clone sizes but do not explain the observed frequencies of each division type. Alternative models have been put forward that invoke deterministic asymmetric inheritance of differentiative cues at the time of division [21–24]. In general, how decisions of single cells contribute to global organ growth and differentiation in developing organs remains an open question. Here we test some of these notions in the context of the emergence of endocrine progenitor cells from uncommitted pancreatic progenitors in the embryonic pancreas. This developmental model has a simple lineage configuration, with a reduced number of fates over well-characterized time windows, and thus provides an optimal testing framework. We use 3-D live imaging of pancreatic explants ex vivo and in vivo, together with lineage tracing at a clonal density, to address the dynamics of the progenitors of the endocrine lineage. In addition to monitoring their lineage, we determined measurements for cell cycle length, synchrony, and differentiation dynamics of these progenitors. This revealed three types of pancreatic progenitor behaviours: (1) symmetric progenitor self-renewal, (2) symmetric endocrinogenic divisions leading to two NEUROG3+ endocrine progenitors, and (3) asymmetric divisions generating a pancreatic progenitor and an endocrine progenitor. By live tracing individual cell fate specification events, we uncover the relationship between Neurog3 expression timing and mitosis. We identify major differences in the onset of Neurog3 transcription between cells stemming from symmetric and asymmetric divisions, and further show that this onset is highly synchronized between symmetrically generated sibling cells. Our analysis of such findings leads to a novel interpretation of the choice between symmetric and asymmetric cell divisions. We posit that asymmetric cell divisions are the result of the stochastic induction of endocrine fate in one of the progenitor daughters, rather than a decision made during cell division. Alternatively, if this progenitor divides a last time after induction, which is expected if the induction happens late in G1, the division will be seen as symmetric differentiative. These results argue against conventional views of asymmetric inheritance of differentiative cues at the time of division [21–24] and are instead consistent with a model of cell cycle–dependent stochastic specification of organ-specific progenitors. To study how individual pancreatic progenitors contribute to pancreas expansion and to monitor their differentiation into endocrine progenitors, we conducted live imaging of explants of dorsal pancreatic buds from E12. 5 Pdx1tTA/+; tetO-H2B-GFP embryos (Fig. 1A). The buds were dissected and laid on a fibronectin-coated coverslip bottom plate, where they grew (Fig. 1B) [25,26]. After 24 h of settling time, we initiated high-magnification time-lapse live imaging in 3-D with 6-min intervals for up to 24 h. Nuclear H2B-GFP fusion protein enabled us to observe cell divisions and to track individual cell nuclei. At the end of the experiment, the explants were fixed and immunostained for markers of pancreatic progenitors (SOX9) and endocrine progenitors (NEUROG3) (S1F–I Fig.), while preserving the native green fluorescent protein (GFP) signal (S1G Fig.), which enabled us to match to the cells from the last frame of the time-lapse movies. The SOX9+ cells constituted the majority of GFP+ epithelial cells (S1I Fig.), and NEUROG3+ cells were mainly observed in the middle trunk region of explant (S1H Fig.) [8]. In spite of the constant exposure to laser, explants grew, showed active cell movements, apico-basal polarization, branching, acini morphogenesis, and differentiation similarly to explants that were not subjected to imaging (S1A–E, J, N, O Fig. and S1 and S2 Movies). After 18–24 h of time-lapse imaging (42–48 h post-dissection), NEUROG3+ cells were detected by immunostaining, showing that the differentiation process occurred ex vivo, albeit less efficiently than in vivo (S1 Table). To determine how single progenitor cells contribute to balancing global pancreas expansion with endocrine progenitor generation (Fig. 1C), we systematically back-tracked NEUROG3+ endocrine progenitors in 3-D, as well as a random subset of SOX9+ pancreatic progenitors that were identified from immunostaining images and mapped onto the last frame of time-lapse movies. Once a cell division was observed through back-tracking, the sister cell was then forward-tracked, and its fate was determined by referring to the immunostaining. The tracking revealed that pancreatic progenitors divided in one of three different modes. The first type of division was symmetric, giving rise to two SOX9+ progenitor cells (P/P division; S3 Movie and Fig. 1D, F–I). The second type was asymmetric, giving rise to a SOX9+/NEUROG3− pancreatic progenitor and a NEUROG3+ endocrine progenitor (P/N division; S3 Movie and Fig. 1E, F–I). The last type was symmetric endocrinogenic, producing two NEUROG3+ cells (N/N division; S4 Movie and Fig. 1J–N). In order to quantitatively account for each division mode, we analysed 1,628 divisions comprising all observed division events of Pdx1tTA/+; tetO-H2B-GFP cells from randomly selected positions from four time-lapse movies. Thus, non-NEUROG3-producing divisions include both bi-potent progenitors and acinar cells, since Pdx1+ cells are multipotent at E13. 5. This quantification revealed 6. 6% ± 1. 6% of divisions producing endocrine progenitors, and 93. 4% ± 1. 6% generating either progenitors or exocrine cells (Fig. 1O and S2 Table). Of all the divisions producing NEUROG3 cells that could be tracked, 56. 3% ± 13. 8% produced a SOX9+ cell and a NEUROG3+ cell (P/N division), and 43. 7% ± 13. 8% produced two NEUROG3+ cells (N/N division; Fig. 1O). We could determine the origin of approximately half of NEUROG3+ cells through P/N or N/N divisions in the past 24 h, while some NEUROG3+ cells either did not exhibit prior division or were either lost or dead during tracking (Fig. 1P). Cell death might be a consequence of the explant culture since apoptosis is very rare in the pancreas epithelium in vivo [27]. Taken together, these results show that at E13. 5–14. 5 most progenitors undergo symmetric renewing divisions, accounting for pancreas size increase, while the remaining progenitors are approximately evenly split into those undergoing symmetric endocrinogenic division and asymmetric division. While ex vivo imaging enables constant monitoring of cell behaviours, it is performed in a somewhat artificial context. In order to determine whether pancreas progenitors undergo the same pattern of symmetric and asymmetric divisions in an in vivo context, we devised a clonal lineage tracing strategy (Fig. 2A) using Hnf1bCreER mice. Previously, this line was used to demonstrate that the E13. 5 HNF1B+ progenitor cells give rise to ductal and endocrine cells [3]. This could be accounted for either by individual cells giving rise to endocrine and ductal cells or by heterogeneity among HNF1B+ cells, some giving rise to endocrine cells and others to ductal cells. To investigate this question, we subjected pregnant mice carrying E13. 5 Hnf1bCreER; mT/mG embryos to a single low-dose intraperitoneal injection of 4-hydroxytamoxifen (Fig. 2B) to label pancreatic progenitors at a clonal density. We optimized conditions for clonal tracking leading to two-cell clones at E14. 5 (Fig. 2B–G, S3 Table, and S5 Movie). Since we know from the time-lapse experiments that the majority of daughters from the same cell do not move more than 30 μm apart, we called labelled cells within 30 μm of each other a clone (S8 Fig.). Reiterations with a 60 μm radius led to similar outcomes. Whole-mount immunostaining of 22 pancreata and detection of 244 two-cell clones revealed that the majority of progenitors in which the fate could be determined divided symmetrically (P/P; Fig. 2E) into two SOX9+ progenitors (59. 8%; Fig. 2H). This proportion is lower than the 93. 4% found in the explants, in part because the cells traced by HNF1B are a subgroup of PDX1+ and SOX9+ cells traced in the explants and some of the latter will give rise to acinar cells [2,4]. In vitro lineage tracing with Hnf1bCreER; mT/mG explants showed that 6. 3% of clones became endocrine (S2 Fig. and S4 Table). This shows that the in vivo differentiation process is more efficient than in vitro differentiation. After the 24-h tracing period, we could not yet observe any INSULIN+ clones in vivo, suggesting NEUROG3− or SOX9− clones might be in transition to endocrine differentiation. Of the NEUROG3-producing two-cell clones in which the fate of both daughters could be determined, 61. 8% were asymmetric, generating one NEUROG3+ daughter and a SOX9+ progenitor (P/N; Fig. 2F, H), and the remaining were symmetric with two NEUROG3+ daughters (N/N; Fig. 2G, H). As a consequence, more NEUROG3+ cells originated from symmetric divisions (Fig. 2I). These results thus provide in vivo evidence of asymmetric and symmetric endocrinogenic progenitor divisions, as well as of symmetric renewal of progenitors, confirming the modes of divisions detected by the explant tracking data. From the above data with regard to fate-determinable two-cell clones, we estimated expected ratios of P/P, P/N, and N/N divisions to be 69. 9%, 18. 6%, and 11. 5%, respectively, after excluding indeterminable clones. Progenitors undergoing symmetric differentiating divisions will contribute all of their progeny to the differentiated pool, whereas asymmetrically dividing progenitors will contribute only one half of their progeny to this pool. We can therefore directly estimate the probability of differentiation of new-born cells to be 20. 8% ([0. 5 × 18. 6] + 11. 5) %, which is consistent with a net expansion of developing pancreas (S1. 5 Text). If sibling cells adopted their fate independently of each other, the expected fractions for each division type would be 62. 7% for symmetric proliferative, 4. 3% for symmetric differentiative, and 33% for asymmetric. Notably, these last two fractions deviate from the experimentally reported ones (Fig. 2H), contradicting the hypothesis of independent sibling fate allocation. This is further supported by statistical tests indicating a significant divergence from the independence ratios (S1. 5 Text). Similar calculations can be made on the in vitro data leading to the same conclusion that a single conversion event leads to symmetric endocrine cell production (Fig. 1O and S1. 5 Text). To investigate the dynamics of differentiation, we generated transgenic Neurog3-RFP reporter lines that can be used for live imaging together with H2B-GFP (Figs. 3A, B and S3A, B). Immunostaining for NEUROG3 and comparison with red fluorescent protein (RFP) from E14. 5 Neurog3-RFP pancreas revealed that 40. 1% ± 4. 5% of NEUROG3+ cells were co-expressing RFP, while the remaining NEUROG3+ cells were RFP− (S3C, D Fig.). Some discrepancies may be expected because of the transient nature of Neurog3 expression and the different onset and decay kinetics of the RFP protein compared to the NEUROG3 protein (S1. 3 Text). Moreover, 77. 1% ± 2. 8% of RFP+ cells were NEUROG3− due to probable delay and perdurance of RFP as compared to that of NEUROG3 [28], as also seen for other reporters [29–31]. This maintenance was attested by the detection of hormones in 18. 5% ± 2. 3% of RFP+ cells. To further address the reliability of the reporter and assess its incidence in our analysis, we compared this line to the enhanced yellow fluorescent protein (EYFP) knock-add-on allele, which has been reported to show a greater overlap with NEUROG3 protein [29] and which is, in principle, less susceptible to exogenous chromatin environments, being in the endogenous locus. Our imaging of explants expressing one allele of EYFP and one of RFP (S4A, B Fig.) showed that all RFP+ cells were also EYFP+ (S4C Fig. and S5 Table), indicating no false positives due to genomic environment. Single cell tracing showed that RFP was turned on 4. 7 ± 1. 1 h after EYFP was detected (S4D Fig.); 11. 6% ± 3. 7% of EYFP+ cells never became RFP+, indicating the system was largely faithful. From time-lapse imaging extended to 48 h, we could observe a dynamic change of RFP signal in single cells from the onset of fluorescence: gradual increase and a subsequent decrease, which reflects the transient expression of NEUROG3 [32]. Our analysis predicts a short half-life of 5–6 h for RFP in a cell, most probably due to continuous laser exposure. We estimate a “perdurance” of detectable fluorescence of more than 20 h (see S5 Fig. and S1. 3 Text) and a minimum delay between cell priming and RFP onset of approximately 5 h. Monitoring all events of RFP onset (n = 323; Fig. 4A) initially suggested waves of cellular differentiation at the tissue level. However, statistical analysis of the timing of onset events showed that these are also compatible with a stochastic process of cell differentiation with homogeneous differentiation rate (i. e. , a Poisson process) throughout the imaging period (S6 Fig. and S1. 4 Text). While we cannot rule out a periodic process underpinning Neurog3 expression, confirmation of this would require more data points. Similar to earlier tracking, RFP+ cells were back-tracked from the last time point in time-lapse movies, their prior division was monitored, and sister cells were forward-tracked. Quantifications (S6 Table) revealed that among the RFP-producing divisions where the fate of the two sisters was determinable, as follows: 60. 2% ± 11. 9% were asymmetric divisions producing a progenitor and a RFP+ daughter (P/R; S6 Movie, and Fig. 3C–G, N), and 39. 8% ± 11. 9% were symmetric divisions producing two RFP+ daughters (R/R; S7 Movie, and Fig. 3H–N). In these long time-lapse movies, many RFP+ cells moved out of frame or were lost due to weak GFP signal before acquiring RFP expression (Fig. 3O and S6 Table). Excluding those indeterminable RFP cells, 18. 8% ± 6. 6% were generated through P/R division, 25. 0% ± 10. 0% through R/R division, and 3. 2% ± 2. 8% through RFP division, while no division was seen during the movie duration for 53. 0% ± 10. 3% of RFP cells (Fig. 3O). These RFP tracking results thus confirmed the calculated proportions of asymmetric versus symmetric divisions established from the previous live imaging and in vivo clonal analysis (Fig. 3P). The dynamic reporter revealed highly synchronized differentiation after divisions producing two NEUROG3 cells, the RFP signal being detected in both daughters within 0. 8 ± 0. 4 h of each other (Fig. 4B, correlation coefficient between lag times 0. 98 ± 0. 002). This outstanding synchrony confirms that it is unlikely that the two daughters are induced by independent events and suggests that mother cells have been primed to differentiate into NEUROG3 cells prior to their division. This observation further suggests a defined time between priming and NEUROG3 expression (or its RFP proxy). In addition, asymmetrically generated NEUROG3 cells exhibited a significantly longer lag time between the division and RFP onset, as shown in Fig. 4C, further supporting an interplay between cell cycle, the differentiation priming event, and the division mode. These results on the contrasting dynamics of differentiation between cells stemming from symmetric versus asymmetric division events are obtained with the RFP reporter, for which we have established a false negative rate of 11. 6% (S4C Fig.). This implies that, far from amplifying the differences between the dynamics of differentiation between the two groups, we might be underestimating them. Specifically, our reporter may miss a subset of Neurog3-expressing cells, thus leading to mis-allocation of around 11. 6% of symmetric events to asymmetric and therefore homogenizing the two categories and reducing the differences between them (See below). To try to understand the mechanism underlying the emergence of endocrine progenitors, we devised a simple mathematical model [33] of cell proliferation dynamics based on the lineage and differentiation dynamics data. The model is based on the premise that proliferating progenitor cells primed for Neurog3-dependent differentiation might either exit the cell cycle and become terminally differentiated or commit to complete the cell cycle and produce two terminally differentiated cells via mitosis (Fig. 5A–C). Thus in this model there are three, rather than two, cell types: (i) NEUROG3-primed (N) cells, which are post-mitotic; (ii) cells primed for differentiation but committed to cell cycle completion (L cells); and (iii) progenitor (P) cells, which will not differentiate (Fig. 5D). We assigned a probability q (the differentiation probability) for the differentiation of progenitors and a probability, θ, for primed cells to become N (1- θ to become L). Thus, the model describes the differentiation process in terms of two probabilities, which can be directly inferred from the lineage data (S1. 6 Text), i. e. it has no free parameters. From the in vivo clonal analysis data, we estimate θ = 56. 5% and, as we have already seen, q = 20. 8%. This means that approximately one-half of the progenitors primed for differentiation become post-mitotic (P→N; 56. 5%), while the other half (P→L[→N/N]; 43. 5%) will undergo one last division before differentiating. Because this latter group of cells (L) is transient and contributes terminally differentiated cells (N), its expected abundance in the tissue is residuary. The model predicts that, at any given time, only 9. 8% of cells in the developing tissue are primed progenitors committed to division-cycle completion (L), yet this small fraction accounts for 93. 8% of the symmetric differentiative divisions (L→N/N) and might therefore explain our observation that the fates of sibling cells are linked (S1. 6 Text). The remainder of symmetric differentiative divisions is interpreted to result from random, independent priming in two sister P cells. The model also allows multiple interpretations for the probability of becoming L versus N (e. g. , exposure to differentiation signals, gene expression noise, etc.). One such interpretation is the timing of the priming event after division (i. e. , θ can be construed as accounting for a cell cycle restriction point). For instance, if a cell is primed early after division it might differentiate and halt the cell cycle, whereas if the priming event occurs late in G1, the cell might have already committed to cell cycle completion. Such specific reading of the model, which we adopt hereafter, leads to a few qualitative predictions on the dynamics of differentiation. First and foremost, the vast majority of sibling cells (93. 8%) from symmetric divisions will have a perfectly synchronized differentiation program. According to the model, differentiated cells stemming from symmetric divisions shall turn on the differentiation program, on average, much earlier than those from asymmetric divisions (S1. 6 Text). To quantitatively account for these predictions and compare them to the experimental data, we performed computational simulations of the model (n = 10,000 clones, S11 Fig.) including the observed variability in the cell cycle length as well as the dynamics of the fluorescent reporter (Figs. 5E–G, S9, S11 and S1. 3, S1. 4, and S1. 6 Text; data deposited in the Dryad repository: http: //dx. doi. org/10. 5061/dryad. 4b58d [34]). The simulations reproduced the differences in the onset of the reporter in cells stemming from symmetric versus asymmetric divisions (Figs. 5E, F, S6E, F, and S9) and also for the high degree of synchronization between sibling cells (Figs. 5G, S6G, and S9). Furthermore, when we included the 11. 6% false negative rate in the reporter (S4C Fig.) of the model, the results were not significantly affected (S10 Fig. and S1. 6 Text). The results of the model led us to experimentally characterize the cell cycle. We used FACS-sorting of Pdx1tTA/+; tetO-H2B-GFP+; Ngn3-RFP− cells marking pancreatic progenitors at E14. 5 to establish their cell cycle partition and observed that 69% were in G0/G1,27% in S and 5% in G2/M, whereas 98% of Neurog3-RFP+ cells were in G0/G1 (S7 Fig.). The progenitors thus spend the majority of their time in G1. Our hypothesis is that priming in early G1 would lead to differentiation and exit of the cell cycle, and its mother would thus have an apparent asymmetric cell division. In contrast, priming in late G1 after the cell has committed to complete the cell cycle through DNA replication and mitosis would lead to an apparent symmetric differentiative cell division. Finally, simulations also predicted the existence of a residual fraction of primed cells that would turn on the reporter immediately before dividing. Noticeably, although most RFP+ cells did not divide, we observed 3 cases of RFP+ cell division producing six cells (Fig. 6, S8 Movie, and S6 Table) and thus accounting for 3. 2% ± 2. 8% of all tractable RFP cells. The average time between RFP signal acquisition and division was 1. 7 ± 0. 8 h. This result is in agreement with the previous estimations of the progression of NEUROG3 cells through S-phase (BrdU incorporation) [11] and further shows that the NEUROG3 cells can exceptionally progress through mitosis at an early stage of their life [29]. The longer movies enabled the observation of multiple rounds of division and the quantification of cell cycle parameters (Fig. 3B). They first confirmed that the daughter cells qualified as progenitors after asymmetric cell division based on SOX9 expression (Fig. 1H) were functionally behaving as progenitors. Indeed, among P/R divisions (n = 27), we observed 14 events where the RFP− daughter divided, producing second-generation progeny (S6 Movie and Fig. 3C). Of those 14 cases, we observed two events where one or both granddaughters divided again, producing third-generation progeny. For all of those, immunostaining at the end point revealed that the RFP− progeny were SOX9+ progenitors (Fig. 3E). In such cases, we could calculate the doubling time of daughter progenitor divisions, and we compared it between P/P and P/N or P/R divisions (Fig. 4D). The doubling time of self-renewing progenitors from P/P divisions was shorter than that from P/N or P/R divisions (p = 0. 04,21. 0 ± 2. 4 h and 26. 5 ± 7. 3 h, respectively). Moreover, the distribution of data points was greater in the asymmetric cell divisions. Finally, the time-lapse movies revealed that pancreatic epithelial cells were highly dynamic and that two daughters migrated to distances up to 64 μm apart from each other in the 24 h following division regardless of division mode (S8 Fig.). In this study, we elucidate the contribution of single cell decisions to the balance between expansion and differentiation in the pancreas. Our lineage analysis, combining in vivo genetic clonal tracing with dynamic imaging in explants, reveals the existence of three kinds of divisions: symmetric progenitor self-renewal, symmetric endocrinogenic divisions leading to two NEUROG3+ endocrine progenitors, and asymmetric divisions generating a pancreatic progenitor and an endocrine progenitor. Furthermore, we show that progenitors are stochastically primed for endocrine differentiation, and that timing of induction in NEUROG3+ cells within the cell cycle establishes the division mode. Whereas late-induced cells complete the cell cycle, resulting in a differentiative symmetric division, early induced cells exit the cell cycle, in which scenario their mother would have produced asymmetric progeny. The results can alternatively be interpreted as HNF1B+ cells being a mix of three pre-determined populations, amongst which only one is truly bipotent. However, the clonal analysis performed in vitro shows different proportions of P/P, P/N, and N/N divisions as compared to in vivo, which would not be expected if the three HNF1B+ subpopulations were predetermined (unless some would preferentially die, which was not observed). Our data is most consistent with a model in which all progenitors are similar, except for their cell cycle stage, and can be primed for endocrine specification with a differentiation probability of around 20% in vivo. Future studies should reveal how this probability changes with time. For example, how it evolves to the cessation of differentiation at the end of gestation, leading to homeostatic conditions that rely primarily on slow self-duplication of differentiated populations [35]. On the other hand, a first phase of symmetric progenitor expansion followed by an increase in the probability of differentiation minimizes the time to form mature organs [36] and may also be expected to occur in the pancreas. Analogous studies are also needed in the human pancreas, as the size of the organ and the length of the differentiation stage are much greater, and several parameters such as cell cycle length of progenitors, probability of differentiation, and ratio of symmetric and asymmetric differentiative divisions may differ. The high correlation between our in vivo and in vitro results (Fig. 3P) rules out erroneous interpretations due to in vitro artefacts and biases caused by subpopulations of progenitors marked by HNF1B at low tamoxifen doses. Spatially, the endocrinogenic divisions were observed in the centre of the pancreas where the HNF1B+ progenitors reside, but no areas of preferential symmetric or asymmetric division were observed. Our dynamic data, including the synchrony in differentiation of symmetrically produced endocrine progenitor cells and their shorter lag from division to differentiation, argue that the specification event can occur at different phases in the cell cycle conditioning the ability to execute a final division or not (Fig. 5A, B). This is further supported by our analysis of the cell cycle–dependent priming model, which displays a good fit to the experimental results and provides a causal understanding of the dynamics of the process. The model proposes parameters q and θ that can be measured in other systems to test its prevalence, and our analysis of previous data in other organs suggests that it may be more general rather than specific to the pancreas [33]. Although the molecular mechanisms of Neurog3 priming remain to be elucidated, especially whether it is under cell-intrinsic or extrinsic control, our data provide information on the general principles. Intrinsic control may be based on asymmetric inheritance of molecular components during division [21–24] or incremental or oscillatory expression of transcriptional determinants [37]. Our results strongly argue against the iterative asymmetric inheritance of differentiation cues at the time of division, as seen in Drosophila neurogenesis and also reported in the mouse brain [24]. Indeed, if the specification was determined at the time of division, the differentiation should occur after the same lag time in symmetric and asymmetric cell divisions. Moreover, the lag time between division and Neurog3-RFP onset is very heterogeneous ranging from 0 to 20 h (Fig. 4C), which is difficult to reconcile with a specification occurring at the time of division. If either cumulative increase or oscillations of an intrinsic determinant promoting endocrine fate lead to differentiation, the progeny of progenitor daughters arising from asymmetric division may exhibit an endocrinogenic bias. On the contrary, these progeny were all SOX9+ progenitors, which would rather suggest a negative bias. However, the movie duration might have been too short to observe differentiation after the second division. Moreover, we observe a slower doubling time of progenitor daughters from an asymmetric division, which may result from the inheritance of a factor that slows down the cell cycle [38–40]. Incremental specification could explain why the cell cycle time is also more heterogeneous in these progenitors. Our analyses are also compatible with extrinsic specification, for example, in the context of Notch-Delta-mediated lateral inhibition [41]. The apparent discrepancy with differentiation in the nervous system where uneven splitting of molecular cues at mitosis leads to asymmetric cell division requires further investigations in both systems. When quantified, the ratios of asymmetric and symmetric differentiation events are very similar in the pancreas and the nervous system [33], and our model would be compatible with the observation that lengthening of G1 impacts the cell division modes in the cortex [30]. Thus, an assessment of the differentiation dynamics in the nervous system similar to ours would be useful, and the possible existence of asymmetrically inherited of cues in mitotic cells in the pancreas can also be considered. Our results reveal that the balance between expansion of progenitors and endocrine differentiation can potentially be regulated by either controlling the probability of endocrine cell induction or its timing in the cell cycle to boost the generation of endocrine cells in vitro for a cell therapy of diabetes. Our approach paves the way to establish how the frequency of division and the ratio of the different types of divisions vary over time and how their balance is controlled by signalling pathways such as Notch and FGF. Genetically engineered mice used for this study were as follows: Pdx1tTA/+ [42], tetO-HIST1H2BJ/GFP (tetO-H2B-GFP) [43], Hnf1bCreER [3], Gt (ROSA) 26Sortm4 (ACTB-tdTomato, -EGFP) Luo/J (mT/mG) [44], Neurog3-EYFP [29], and Neurog3-RFP (S3 Fig.). For embryonic stage, noon of the day when vaginal plug appeared was referred as E0. 5. The Neurog3-RFP transgenic construct (S3A Fig.) was generated by fusing 7. 6 kb of the Neurog3-promoter [2] with a reporter construct composed of a chimeric intron; turbo RFP (Evrogen); a nuclear localization signal (NLS); a Myc-tagC; a bovine growth hormone polyadenylation signal (bGH-PolyA). Transgenic mouse lines were obtained by pronuclear injection of the construct (Transgenic Core Facility, EPFL, Switzerland). Two different lines were obtained initially, exhibiting similar levels of RFP signal detectable by a wide-field fluorescent microscope, and one of the lines was used for this study. All animals were handled humanely according to the authorized protocols of Switzerland and Denmark. Dorsal pancreata from E12. 5 Pdx1tTA/+; tetO-H2B-GFP or Pdx1tTA/+; tetO-H2B-GFP; Neurog3-RFP were cultured on a fibronectin (Sigma) -coated coverslip, adapted from the previously published protocol [25]. GFP and RFP were readily detectable under wide-field fluorescent microscopes. We used a culture medium composed of Medium 199,10% fetal calf serum, 1% Fungizone, and 1% penicillin/streptomycin (all from Gibco). After 24 h of culture that enabled stabilization of explant flattening to approximately 80 μm thick, pancreatic explants were imaged at a single-cell resolution using Leica SP5 or SP8 confocal microscopes with a 63X glycerol immersion objective in a humidified, heated, CO2-controlled chamber. Tiled positions (9 [3x3] or 12 [3x4] tiles) were scanned in 256x256–280x280 format with around 40 μm Z-stack (voxel size, 0. 506x0. 506x1. 3 μm3–0. 880x0. 880x1. 25 μm3) every 6 min for 18–48 h. The GaAsP hybrid detection system (Leica HyD™) enabled a substantial reduction of laser power by 62. 5% and increase in signal-to-noise ratio resulting in reduced scanning time, compared to conventional photomultiplier detectors. At each time point, it usually took approximately 5 min and 30 s to scan 9–12 tiled positions in 3-D. At the end point of image acquisition, the explants were fixed and prepared for whole-mount immunostaining. Tiled images were stitched using either Leica AF6000 software or a custom-built Massive Stitcher plugin (Bioimaging and Optics Platform, EPFL, Switzerland) in Fiji. Imaris (Bitplane, Switzerland) software was used to track cells and their divisions in 3-D maximum intensity projection. Once immunostaining was done, NEUROG3+ endocrine progenitor cells from staining images were manually identified on the last frame of time-lapse movies with Pdx1tTA/+; tetO-H2B-GFP explants by GFP superimposition. The identified endocrine progenitors were first back-tracked to monitor their prior divisions. Once a division was observed, the other sister was forward-tracked to the final frame, and its fate was determined from the immunostaining images. For time-lapse movies from explants with Ngn3-RFP in addition to Pdx1tTA/+; tetO-H2B-GFP, RFP+ cells were back-tracked, and the fate of each sister was determined by immunostaining. For the quantification of total cell divisions, due to the technical difficulties in tracking all Sox9+ cells from the immunostaining, we did not trace all the individual cells from those 1,628 divisions, but rather subtracted the tracked divisions that produced NEUROG3 cells from the total number of divisions. Pregnant mice carrying Hnf1bCreER; mT/mG embryos were injected intraperitoneally with 0. 175 mg 4-hydroxy (4-OH) tamoxifen (H6278, Sigma Aldrich) at E13. 5. 4-OH tamoxifen was prepared as a 10 mg/mL stock in 90% corn oil and 10% ethanol and diluted to obtain the desired dose. Embryos were harvested at E14. 5, and the dorsal pancreas was isolated and subjected to whole-mount immunostaining for GFP, SOX9, and NEUROG3, as indicated below. The fixation procedure eliminates native GFP and Tomato signals. After whole-mount immunostaining, dorsal pancreata were dehydrated through an ascending methanol series and subjected to clearing in a 1: 2 solution of benzyl alcohol to benzyl benzoate (BABB). Cleared samples submerged in BABB were mounted on glass depression slides and imaged whole-mount using a Leica SP8 confocal microscope with a 20X oil objective at a 1024x1024 format. 3-D reconstruction of whole-mount imaged pancreata was performed using Imaris (Bitplane), enabling detection of recombined clones while preserving the spatial organization of the pancreas, thereby ensuring detection of clonal progeny by allowing interclone distance measurements. Two-cell clones were identified in 3-D space, and categorized according to SOX9 and NEUROG3 status. Recombined cells were only considered to be of clonal origin if the distance between recombined cells was less than 30 μm after the tracing period, based on live imaging data (S8 Fig.). The results were not sensitive to this parameter as using 60 μm as a maximal distance to be considered as a clone led to the same proportion of the three types of division (S2 Data). Hnf1bCreER; mT/mG embryos were also used for in vitro clonal analysis by explanting pancreata at E13. 5 and growing these at the air–liquid interface on 0. 4 μm filters (Millipore). Explants were subjected to a 6 h pulse of 25 nM 4-OH tamoxifen in 100% ethanol to induce labelling at clonal density. Following tracing for 48 h, explants were fixed and subjected to whole-mount staining and imaging as indicated below. Whole-mount immunostaining was performed after live imaging or for pancreata harvested from the lineage tracing. After fixation with 4% paraformaldehyde (PFA) for 5 min on ice, samples were washed in phosphate buffered saline (PBS) for 5 min three times. Then, they were dehydrated through 50% and 100% methanol, and could be stored at −20°C until later use. When ready, samples were rehydrated through 50% methanol and washing solution, PBS+0. 5% Triton X-100 (Tx100). Throughout the procedure, all the solutions contained 0. 5% Tx100, and all the incubation was undergone in 4°C. After blocking overnight in blocking solution (1% Bovine serum albumin+0. 5% Tx100), samples were incubated with primary antibodies (S7 Table) in blocking solution for 24–48 h. After washing, secondary antibodies were applied overnight, followed by washing. Alexa fluorophore conjugated secondary antibodies (Invitrogen) were used. Stained explants were kept in PBS and imaged using a confocal microscope. For quantification from explants, NEUROG3+ cells were counted manually, and H2B-GFP+ cells were counted using a custom-made macro in Fiji. Immunostaining of frozen sections from E14. 5 Neurog3-RFP pancreata was performed as previously described [6], and images were taken with a Leica DM5500 microscope. Quantification was obtained by manually counting immunopositive cells on every sixth section. Statistical analyses were done by two-tailed Mann-Whitney U-test using GraphPad Prism software. Values were presented as the mean ± standard deviation.

In order to form organs of the right size and cell composition, progenitor cells must balance their proliferation and their differentiation into functional cell types. Here we study how individual progenitor cells in the developing pancreas execute their choices to either expand their pool or differentiate into hormone-producing endocrine cells. Using live microscopy to track the genetically marked progeny of single cells, we reveal that after they divide, individual cells generate either two progenitors, two cells on the endocrine path, or one progenitor and one cell on the endocrine path. Quantitative analysis shows that endocrine differentiation is largely stochastic and that the probability of progenitor cell differentiation by the end of mid-gestation is about 20%. We propose a model in which the production of a progenitor and a differentiated cell in the pancreas results from the stochastic induction of differentiation in one daughter after cell division, rather than the unequal partitioning of molecules between two daughters at the time of division, as observed in the nervous system. Furthermore, when two daughters become endocrine cells, this results from the induction of differentiation followed by cell division—rather than two independent induction events. This model may be applicable to other organs and provides insights to optimize the generation of β-cells in vitro for diabetes therapy.

Abstract Introduction Results Discussion Materials and Methods

2015

Cell Cycle–Dependent Differentiation Dynamics Balances Growth and Endocrine Differentiation in the Pancreas

There are two main classes of natural killer (NK) cell receptors in mammals, the killer cell immunoglobulin-like receptors (KIR) and the structurally unrelated killer cell lectin-like receptors (KLR). While KIR represent the most diverse group of NK receptors in all primates studied to date, including humans, apes, and Old and New World monkeys, KLR represent the functional equivalent in rodents. Here, we report a first digression from this rule in lemurs, where the KLR (CD94/NKG2) rather than KIR constitute the most diverse group of NK cell receptors. We demonstrate that natural selection contributed to such diversification in lemurs and particularly targeted KLR residues interacting with the peptide presented by MHC class I ligands. We further show that lemurs lack a strict ortholog or functional equivalent of MHC-E, the ligands of non-polymorphic KLR in “higher” primates. Our data support the existence of a hitherto unknown system of polymorphic and diverse NK cell receptors in primates and of combinatorial diversity as a novel mechanism to increase NK cell receptor repertoire. Natural killer (NK) cells are bone marrow-derived lymphocytes that form an essential part of the immune response against pathogens and are involved in the elimination of tumour cells. Equipped with a diverse array of germline-encoded receptors that are able to mediate inhibitory or activating signals [1], NK cells scan other cells for the presence of ligands of these receptors [2]. Activation of NK cells is typically achieved by discontinuation of inhibitory signalling and involvement of activating receptors, resulting in cytokine release or killing of target cells [2]–[4]. Most NK cell receptors interact with members of the MHC class I protein family and either belong to the killer cell lectin-like receptors (KLR) of the C-type lectin-like family such as CD94, NKG2, or Ly49, or the killer cell immunoglobulin-like (KIR) receptors, which are encoded in the natural killer complex (NKC) and leukocyte receptor complex (LRC), respectively [1]. Both NKC and LRC contain inhibitory and activating NK cell receptors. Inhibitory receptors are characterised by the presence of immunoreceptor tyrosine-based inhibitory motifs (ITIM) in the cytoplasmic tail, whereas activating receptors lack ITIMs and instead contain a positively charged amino acid (arginine or lysine) in their transmembrane region, thereby associating with signalling adaptor molecules DAP10, DAP12 or FcγR [2]. The polymorphic NK cell receptors are represented by KIR in humans, apes, Old World and New World monkeys [5]–[7] and KLR (Ly49 molecules) in rodents [8]. These two receptor systems are not structurally related, but have similar functions: interaction with MHC class I molecules and regulation of NK cell activity [2]. Due to these functional constraints, KIR and Ly49 genes have to keep pace with the rapidly evolving and polymorphic class I genes [9] and it has been shown that combinations of this highly complex and polymorphic genetic system of NK cell receptors and their MHC class I ligands can significantly influence susceptibility and resistance to infectious and malignant diseases, autoimmune disorders, and reproduction [10]. In contrast to KIR and Ly49, another genetic system of MHC class I-binding NK cell receptors has been kept conserved in primates (hominoids, Old World and New World monkeys) [11] and rodents [12], the CD94 and NKG2 molecules, which also belong to the KLR family. The CD94 molecule can either pair with the inhibitory NKG2A or the activating NKG2C and NKG2E molecules and these heterodimeric NK cell receptors specifically recognise conserved nonclassical class I molecules, HLA-E in humans [13] and H2-Qa1 in mice [14]. These nonclassical class I molecules bind peptides derived from signal sequences of certain class I molecules [15], thereby monitoring putative downregulation of the corresponding class I molecules mediated by pathogens as part of their immune evasion strategy. Various NK cell receptors (CD94/NKG2, KIR, Ly49L) have been identified in different nonhuman primates, and rapid evolution in particular of the KIR genes was reported [16]. Yet, two questions remain open: how far can we trace back KIR3DL diversification and what was the NK cell receptor content in the ancestor of all primates? We report here that strepsirrhine primates such as lemurs have evolved a ‘third way’ of a diverse NK cell receptor system and exhibit highly diversified, positively selected CD94/NKG2 receptors. This NK cell receptor system is further characterised by combinatorial diversity. Clones containing the LRC, NKC, and MHC genomic regions were isolated from a BAC library of the grey mouse lemur (Microcebus murinus). Nineteen BAC clones from the LRC region were identified and three clones covering the KIR-LILR subregion were sequenced: one clone contains the NCR1, FCAR, and LILR genes in addition to a KIR3DP pseudogene mapping between FCAR and the LILR gene cluster, and the other two clones both include LILR genes and a single copy of the KIR3DX1 [17] gene (Figure 1A). Similar to all other primate KIR3DX1 genes, the two lemur copies, of which only one is intact, also lack the MLTD1 repetitive elements in intron 3. The intact copy is transcribed in lemur PBMC and alternatively spliced (Figure S1); the other locus (KIR3DX1P) is a fragmented pseudogene. In contrast, KIR3DP is characterised by the presence of KIR3DL-typical repetitive elements such as MLT1D/LTR33A, MSTB1, MER70B [7] and is thus the only representative of the KIR3DL lineage in the grey mouse lemur. The repetitive elements flanking KIR3DP are found in the ‘outer’ human KIR genes KIR3DL3 and KIR3DL2 (data not shown), suggesting that the lemur KIR3DP is equivalent to all ‘higher’ primate KIR genes. Thus, the only functional KIR in the mouse lemur genome is represented by KIR3DX1, a KIR gene of unknown function [17]. Unlike the KIR region, the CD94-Ly49L genomic interval of the grey mouse lemur NKC showed amplifications of the CD94 and NKG2 genes (Figure 1B), which accounted for its 1. 5 times increased size compared to humans. Only two genes in this region represent one-to-one orthologs of ‘higher’ primate genes: NKG2D and Ly49L. Further investigations of the three mouse lemur CD94 genes indicate they encode typical CD94 receptor structures but their amino acid sequences show considerable diversity, differing by 23–24% among each other (Figure S2). Such level of divergence is significantly higher than in ‘higher’ primates, as human and common marmoset monkey CD94 sequences differ only by 13% for example [11]. Furthermore, eight NKG2-related genes were identified (Figure 1B), of which five are functional genes (Figure S3) and three are pseudogenes. The NKG2 molecules show different combinations of functionally relevant motifs: ITIM, positively charged residue in the transmembrane region, and the YxxM motif, which is a recognition site for the p85 subunit of phosphatidylinositol 3-kinase (PI3K). Allelic substitutions affect the presence of ITIM and YxxM motifs in NKG2-2 and NKG2-3 (Figure S3). These findings suggest that the mouse lemur NKG2 receptors are functionally complex, with inhibitory and activating properties. BAC clones from the MHC class I and class II regions could be mapped to the short arm of mouse lemur chromosome 6 by fluorescence in-situ hybridisation (FISH) (Figure 2). Whereas the class II region is conserved (Averdam et al. , manuscript in preparation), the linked class I gene-containing regions of the mouse lemur MHC lack any functional class I genes as the four identified class I genes and the single MIC gene all represent pseudogenes (Figure 1C). Remarkably, even MHC-E and MHC-F, the only two conserved MHC class I genes of ‘higher’ primates, are missing in the mouse lemur MHC. Interestingly, our screening of the mouse lemur BAC library for MHC class I genes also identified an unlinked genomic region that includes nine MHC class I genes. Complete sequencing of this region revealed six genes encoding functional MHC class I proteins, including putative alleles of the previously described classical class I genes of the mouse lemur, Mimu-W01 and Mimu-W04 [18] (Figure 1C). This class I gene cluster maps to another chromosome, the long arm of mouse lemur chromosome 26 (Figure 2), thus providing an additional example of a mammalian species where MHC class I and class II genes are not linked. A similar chromosomal splitting of class I genes was found for a further lemur species, Coquerel' s giant mouse lemur (Mirza coquereli) (Figure 2). The CD94 and NKG2 genes could be mapped to chromosome 7 in the giant mouse lemur, which corresponds to human chromosome 12, indicating conserved synteny of the NKC region. The KIR3DX1P gene-containing BAC was mapped to a small acrocentric chromosome not identical with chromosome 26 (Figure 2). Using the mouse lemur NKC sequences as reference we investigated corresponding cDNA sequences from the black-and-white ruffed lemur (Varecia variegata), a species that diverged from the grey mouse lemur about 43 million years ago (mya) [19]. A single NKG2D, five CD94, nine NKG2 and two Ly49L cDNA sequences were isolated from ruffed lemur PBMC by RT-PCR cloning. The deduced amino acid sequences exhibit similar functional features as described above for mouse lemur CD94 and NKG2 molecules (Figure S2, Figure S3). Two pairs of CD94 sequences (CD94-2*01, CD94-2*02, and CD94-3*01, CD94-3*02) only differ by a few nucleotide substitutions and are, therefore, regarded as alleles. Similarly, two allelic NKG2 (NKG2-6*01, NKG2-6*02) and two allelic Ly49L (Ly49L*01, Ly49L*02) sequences were isolated, so that the ruffed lemur NKC is estimated to possess three CD94, eight NKG2 genes and single NKG2D and Ly49L genes. Phylogenetic analyses of primate CD94 and NKG2 sequences encoding the C-type lectin-like domain indicate that for both gene families ‘lower’ and ‘higher’ primate sequences form their own groups, indicating that diversification of CD94 and NKG2 occurred in ‘lower’ primates after their separation with ‘higher’ primates (Figures 3A and 3B). ‘Lower’ primate NKG2 sequences form three groups, each with sequences from both lemurs, pointing to duplications that preceded speciation of both lemurs (Figure 3B). Similarly, in the largest of the three ‘lower’ primate NKG2 groups, some gene sequences show species-specific clustering indicating gene duplications following speciation of both lemurs. Such patterns demonstrate that NKG2 gene duplications in ‘lower’ primates occurred both before and after the separation of the two lemur species, a process similar to that seen for KIR in ‘higher’ primates and Ly49 in rodents [20]. In contrast to the C-type lectin-like domain, NKG2 sequences encoding the stem, cytoplasmic, and transmembrane part fall into two branches: one contains ‘lower’ and ‘higher’ primate inhibitory NKG2, the other one containing ‘lower’ and ‘higher’ primate activating NKG2, indicating that these two types of sequences separated before the speciation between ‘higher’ and ‘lower’ primates (Figure 3C). To determine whether amplification of CD94 genes is restricted to strepsirrhine primates living in Madagascar (lemurs) or is common to other ‘lower’ primates, we investigated the CD94 sequences of an African and an Asian primate, the potto (Perodicticus potto) and the tarsier (Tarsius syrichta). Whereas the potto belongs to the primate suborder Strepsirrhini, the tarsier is a member of the other primate suborder, the Haplorrhini (tarsiers, New and Old World monkeys, apes and human). Using all available primate sequences as reference, generic primers were constructed to amplify the region including exon 4, intron 4 and exon 5 of CD94 and we characterised seven and nine different CD94 sequences in potto and tarsier, respectively (Figure S4). These findings thus indicate that CD94 gene amplification is not restricted to lemurs, but can be found in other strepsirrhine primates and even in a primate more closely related to humans than to lemurs [21]. Insertions of repetitive elements of the Alu family in intron 4 revealed additional information: whereas none of the three mouse lemur CD94 genes contains any Alu element, four of the seven potto sequences include an AluJo and all nine tarsier sequences have an AluJb element (Figure S4). However, the integration sites of these Alu elements differ between potto and tarsier, indicating that integration events occurred after these lineages diverged from each other. Hence, the amplification of CD94 sequences occurred repeatedly and independently in ‘lower’ primates. The intron 4 sequences of human, rhesus macaque and common marmoset CD94 share an AluSx integration (data not shown), which is not present in ‘lower’ primates, further supporting the monophyletic origin of ‘higher’ primate CD94. While lemurs (and possibly other ‘lower’ primates) experienced a diversification of the CD94/NKG2 receptor system, they appear to lack MHC-E, the ligand of CD94/NKG2 in ‘higher’ primates, as their MHC region lacks a class I gene where the MHC-E gene of ‘higher’ primates is located (Figure 1C). To investigate if any of the grey mouse lemur MHC class I genes represents an ortholog or functional equivalent of ‘higher’ primate MHC-E, we performed phylogenetic analyses of primate MHC class I sequences. Analyses of the full-length sequences and of the peptide-binding-domain (PBD) alone, with or without the peptide binding residues (PBR) indicate that ‘lower’ and ‘higher’ primate sequences form their own groups (Figure 4A–4C, Figure S5). This demonstrates that MHC class I diversification in ‘lower’ and ‘higher’ primates took place in each taxonomic lineage after their separation and further indicates that the duplication that gave rise to the MHC class I genes on chromosome 26q occurred in the ‘lower’ primate lineage. Similarly, analysis of the PBR revealed that the ‘higher’ primate MHC-E sequences are more closely related to the rodent Qa1 group than to any of the ‘lower’ primate sequences (Figure 4D). Thus, these findings indicate that the grey mouse lemur genome neither encodes a strict orthologue (with a one-to-one relationship) nor a functional homologue of MHC-E and confirm earlier data of our group that ‘higher’ and ‘lower’ primate MHC class I genes lack strict orthology [18]. Our genomic analysis in one mouse lemur individual and cDNA sequencing in three individuals (two mouse lemurs and one black-and-white ruffed lemur) revealed that ‘lower’ primates experienced a diversification of both CD94 and NKG2. Such diversification is in contrast with the situation in ‘higher’ primates where CD94 is a non-polymorphic, single-copy gene and where the NKG2 gene family experienced limited diversification with moderate polymorphism. To investigate if such difference is the result of natural selection favouring functional diversity in ‘lower’ but not in ‘higher’ primates, we compared the non-synonymous to synonymous substitution ratio rate (dN/dS) of CD94 and NKG2 sequences of ‘lower’ and ‘higher’ primates (Table 1). This analysis shows that ‘lower’ primate CD94 sequences display highly significant evidence of positive diversifying selection (α = 0. 001) with 15 positions positively selected (posterior probability (PP) >0. 95). In contrast, their ‘higher’ primate counterparts do not show any evidence of positive selection (Table 1). A similar, yet not as marked, trend is seen for NKG2 as ‘lower’ primate sequences show highly significant evidence of positive diversifying selection (α = 0. 001), with six positions being positively selected (PP>0. 95) while in ‘higher’ primates the evidence was not as strong (α = 0. 01–0. 05) and limited to only one site (Table 1). Mapping of the 21 positively selected sites in ‘lower’ primates on the three-dimensional model of the human CD94/NKG2A heterodimer in contact with HLA-E [22], [23] suggests that many of these positions represent functionally relevant sites (Figure 5A). Indeed, closer inspection revealed that the distribution of these 21 positively selected sites in the CD94/NKG2 lectin-like domains shows a significant bias (α = 0. 05) toward the 54 sites involved in ligand-binding and/or dimer formation (Table 2, Figure S6). Further dissection of this bias shows it is particularly marked for the 28 ligand-binding sites (α = 0. 01), but not for the 30 sites involved in dimer formation. Within the ligand-binding sites a significant bias toward the 7 sites contacting the MHC class I peptide (α = 0. 003) was observed, but for the sites contacting the MHC-class I heavy chain the bias was marginal (α∼0. 06) (Table 2, Figure S6). Consistent with this, the four CD94/NKG2 positively selected positions that contact the MHC class I peptide account for 75% (9 out of 12) of all the contacts with the peptide (Figure 5B). This analysis thus shows that the genomic diversification of CD94/NKG2 in ‘lower’ primates was accompanied by positive diversifying selection that particularly targeted the sites contacting the MHC class I peptide. In ‘higher’ primates the limited (NKG2) or lack of (CD94) genomic diversification was accompanied by limited (NKG2) or lack of (CD94) sequence diversification through positive diversifying selection. To fully investigate the extent of allelic diversity of ‘lower’ primate NKC genes we specifically amplified the exons encoding the lectin-like domain of mouse lemur CD94, NKG2, and Ly49L genes in a cohort of 46 free-living unrelated animals derived from the Kirindy region in Madagascar [24]. Twelve animals were analysed for all genes and further 34 animals were additionally analysed for the three CD94 genes. No presence/absence polymorphisms of grey mouse lemur CD94 or NKG2 genes were observed in our cohort, indicating that copy number variation similar to what is known for human KIR and mouse/rat Ly49 haplotypes is not evident in grey mouse lemur CD94/NKG2 haplotypes. Single nucleotide polymorphisms (SNPs) were detected for all genes and with the exception of CD94-3 and NKG2D, all the NKC genes show an excess of non-synonymous substitutions (Table S1). While 30% (71/236) of all the codons forming the lectin-like domains of CD94 and NKG2 have non-synonymous allelic polymorphisms in at least one of the ‘lower’ primate CD94 or NKG2 genes, 66% (47/71) of these codons are neither sites involved in ligand binding or dimer formation nor positions we detected as positively selected in our gene analysis. Because the differences between the genes largely overshadowed allelic differences in our analysis for positive selection, such an observation suggests that following gene duplication natural selection first diversified functional sites and subsequently favoured in each species polymorphisms at sites often not directly involved in function, presumably to further ‘tweak’ functions. Unlike the ligand binding sites, the positions involved in dimer formation are not significantly enriched in positively selected sites (Table 2, Figure S6). Because of this apparent lack of major diversification of the dimer formation function, we hypothesised that the three CD94 and five NKG2 molecules of the grey mouse lemur can be freely combined to form various CD94/NKG2 heterodimers. To test this hypothesis, we checked all possible combinations with (extracellular) V5-tagged NKG2 and (intracellular) GFP-tagged CD94 molecules. Expression constructs were transiently transfected into 293T cells and CD94/NKG2 heterodimer formation was assayed by co-immunoprecipitation using tag-specific monoclonal antibodies. Heterodimers were found for all CD94/NKG2 combinations by co-immunoprecipitation (Figure 6). Cell surface expression of all CD94/NKG2 receptor combinations was examined with (both extracellular) FLAG- and V5-tagged CD94 and NKG2 molecules, respectively. All combinations of NKG2-1, NKG2-3, and NKG2-8 with the three CD94 molecules were found on the cell surface (Figure 7A). In contrast, cell surface expression of both NKG2-2 and NKG2-5 receptors in combination with CD94 molecules was either low (CD94-1/NKG2-5) or lacking (Figure 7A), suggesting that these receptors depend on the DAP12 adaptor molecule. An expression construct of mouse lemur DAP12 with a c-myc-tag was established and expression was controlled in parallel experiments by intracellular staining (data not shown). Co-transfection with c-myc-tagged mouse lemur DAP12 significantly increased cell surface expression of all combinations formed by CD94/NKG2-2 and CD94/NKG2-5 heterodimers (Figure 7B). This finding identifies NKG2-2 and NKG2-5 as classical activating receptors whose cell surface expression is dependent on interaction with DAP12. CD94 and NKG2 expression constructs were also transfected in 293T cells alone. Similar to human CD94, all three mouse lemur CD94 proteins can be found at the cell surface in the absence of any NKG2 molecule, but unlike human NKG2 molecules, also mouse lemur NKG2-2 (with DAP12) and NKG2-3, and to a lower extent NKG2-5 (with DAP12) and NKG2-8, are expressed at the cell surface in the absence of CD94 (Figure 7A and 7B). Taken together, our data show that three CD94 molecules can be combined with five NKG2 molecules to form 15 different NK cell receptors that display significant diversity at sites of interaction with MHC class I ligands and their bound peptide. This indicates that besides gene amplification and positive diversifying selection combinatorial diversity is also a mechanism that significantly contributes to the increase of NK cell receptor diversity in ‘lower’ primates. The polymorphic NK cell receptors of humans and mice are functionally similar but structurally unrelated. Therefore, we traced back NK cell receptors and MHC class I ligands to the base of the primate evolutionary tree by analysis of the LRC, NKC, and MHC genomic regions in primates distantly related to humans. We demonstrate that ‘lower’ primates deviate from ‘higher’ primates in the usage of polymorphic NK cell receptors. Except for the KIR3DX1 gene, lemurs and possibly other ‘lower’ primates have neither functional nor highly polymorphic KIR genes like their relatives, the catarrhine (Old World monkeys, apes, and humans) and platyrrhine (New World monkeys) primates, nor do they show an expansion of Ly49 genes as in rodents. Instead, lemurs have considerably amplified and diversified their C-type lectin-like CD94 and NKG2 genes. We conclude from our findings that the NK cell receptor repertoire of ‘lower’ primates is at least as diverse as in ‘higher’ primates or rodents. Thus, in addition to the KIR and Ly49 genes of ‘higher’ primates and rodents, the CD94/NKG2 heterodimers of lemurs represent a third system of polymorphic and diverse NK cell receptors. Compatible with such a system is that the duplicated lemur CD94 and NKG2 genes show sequence diversifications and strong signs of positive diversifying selection. In accordance with these characteristics, lemur NKG2 genes do not show signs of gene homogenisation as opposed to ‘higher’ primate NKG2 sequences [11]. Such homogenisation may serve to keep NKG2 amino acid sequences conserved for interaction with the invariable MHC-E ligand, a situation that is not observed for the polymorphic CD94/NKG2 receptors in ‘lower’ primates. CD94 gene duplications are not restricted to ‘lower’ primates from Madagascar as they were found in an African strepsirrhine primate (Perodicticus potto) and in the Asian tarsier (Tarsius syrichta). The latter is particularly interesting, as the tarsier is more closely related to ‘higher’ primates than to ‘lower’ primates [21]. Analysis of repetitive elements in CD94 intron 4 sequences revealed that duplications had occurred repeatedly and independently in ‘lower’ primates. Thus, the polymorphic CD94/NKG2 system is likely present in many if not all ‘lower’ primates. This is in sharp contrast to the situation in ‘higher’ primates where CD94 is a single copy, non-polymorphic and highly conserved gene. Three KIR genes were detected in the lemur LRC region, a functional and a pseudogene copy of KIR3DX1 and a KIR3DP pseudogene. According to its characteristics, the KIR3DP gene may represent the ‘Ur-KIR’ gene of all ‘higher’ primate KIRs. This gene already contains the repetitive elements MLTD1, MER70B, MSTB1 in its introns, which are assumed to be integrated about 60–100 mya [25], a time that is compatible with the splitting of lemurs and human of about 65–90 mya [26]. Thus, we postulate that KIR and CD94/NKG2 receptors evolved differently in primates: while in the lineage leading to ‘lower’ primates CD94 and NKG2 but not KIR genes expanded, the opposite happened in the lineage leading to ‘higher’ primates where KIR genes expanded and CD94 and NKG2 co-evolved with the non-classical MHC-E molecule to become a conserved receptor/ligand system (Figure 8). The finding that ‘lower’ primates did not amplify and diversify KIR or Ly49 genes but, instead, evolved a polymorphic CD94/NKG2 system, strengthens previous assumptions that mammals only utilise a single class of polymorphic NK cell receptors, despite their obvious ability to develop multiple classes [27], [28]. Most likely, this development is influenced by the pathogenic threat these organisms are subjected to and can involve different receptor types such as monomeric KIR, heterodimeric CD94/NKG2, and homodimeric Ly49. Nevertheless, mammalian species are equipped with all types of receptor genes [1], which gives some flexibility for adaptation according to NK cell receptor and ligand requirements. All class I genes in the MHC equivalent genomic region on grey mouse lemur chromosome 6 are pseudogenes and all functional MHC class I genes were translocated to chromosome 26. Additionally, no functional MIC gene could be identified in any of the class I gene-containing regions in the grey mouse lemur. However, this finding is not surprising, as NKG2D ligands are numerous and functionally redundant [29], absence of MIC is evident in rodents [30] and deletions of both MICA and MICB were reported in East-Asians [31]. In addition to this unusual organisation of the grey mouse lemur MHC, a striking difference to ‘higher’ primates is the apparent absence of a strict HLA-E orthologue or functional homologue. As the BAC library was screened exhaustively and in the light of our genome-wide approach published earlier [18], it appears rather unlikely that MHC-E-like class I genes were not detected. Thus, the putative CD94/NKG2 ligands are expected among the sequenced MHC class I genes. Nevertheless, we cannot completely rule out the possibility that a HLA-E-like gene was not detected by our approach or that a gene with HLA-E-like function is among the detected MHC class I genes. There has been some debate in the past on whether HLA-E and mouse H2-Qa1 have a common origin [32] or whether their similarity in the peptide-binding region is a consequence of convergent evolution [33]. In the light of our data, the second hypothesis appears more likely, as diversified KIR and conserved MHC-E/CD94/NKG2 emerged only in ‘higher’ primates and diversified Ly49 and conserved H2-Qa1/CD94/NKG2 evolved independently in rodents. The observed positive diversifying selection in ‘lower’ primates is more pronounced in the three CD94 than the NKG2 genes, suggesting that the CD94 molecules have significant impact on the binding of the polymorphic MHC class I ligands and their bound peptides. Indeed, the recently determined three-dimensional structure of human CD94/NKG2A in complex with HLA-E revealed that CD94 and NKG2A contribute about 80% and 20%, respectively, to the interaction with HLA-E and its bound peptide [22], [23]. Translated to the situation in ‘lower’ primates, duplication of CD94 genes and their strong sequence diversification by positive diversifying selection reflects the requirement to maintain binding to polymorphic (classical) MHC class I molecules and bound peptides. Finally, we demonstrate that all possible CD94/NKG2 combinations are able to form heterodimers at the cell surface, revealing important implications: exchange of the CD94 or the NKG2 subunit is likely to influence the binding specificity for MHC class I ligands and their bound peptide and the functional properties (inhibitory or stimulating) of the receptor. For example, the combination of the three CD94 and five NKG2 molecules in the grey mouse lemur or three CD94 and eight NKG2 molecules in the ruffed lemur gives rise to 15 or 24 different NK cell receptors, respectively. Thus, we conclude that the NK cell receptor repertoire in ‘lower’ primates is not mainly achieved by duplications, but rather by combinatorial diversity, a phenomenon that was so far unknown for any NK cell receptor. NK cells were recently shown to exhibit features of adaptive immunity, namely immunological memory [34]. Combinatorial diversity of immune receptors is a further hallmark of adaptive immunity. Although the combinatorial diversity of ‘lower’ primate CD94/NKG2 receptors is obviously much less than that usually seen for B or T cell receptors, our data additionally highlight the close relationship of two lymphocyte subsets: natural killer cells and cytotoxic T cells. In summary, we have uncovered a ‘third way’ of polymorphic and diversified NK cell receptors in mammals. The CD94/NKG2 receptor system (and not KIR) of ‘lower’ primates is characterised by duplication, sequence diversification by means of positive diversifying selection and allelic diversity. Consistent with this highly dynamic CD94/NKG2 system, the MHC class I molecules as putative ligands of these receptors show strong signs of co-evolution and an unusual chromosomal organisation. CD94 and NKG2 subunits constitute the main NK cell receptor repertoire in ‘lower’ primates and are freely combinable. This finding discloses a so far unknown mechanism of generating the NK cell receptor diversity: combinatorial diversity. All experiments were carried out in accordance with the French Rural Code Directive (articles R21-87-90), the German Animal Welfare Law, guidelines of the German Research Foundation, and the European Communities Council Directive (86/609/EEC). Field sampling of Microcebus murinus tissue samples in Madagascar was conducted under the autorisation of the Ministère de l' ' Environnement, des Forêts et du Tourisme of the Republic of Madagascar. Samples were exported under a CITES permit by the Bundesamt für Naturschutz, Bonn, Germany. Grey mouse lemurs (Microcebus murinus) and black-and-white ruffed lemurs (Varecia variegata) are housed in the facilities of the UMR CNRS/MNHN and the German Primate Center, respectively. Blood was obtained during regular veterinary inspections. DNA samples from potto and tarsier were kindly obtained from Helga Schulze (University of Bochum, Germany) and Jürgen Schmitz (University of Münster, Germany). Filters of BAC library CHORI-257 derived from a female grey mouse lemur (Microcebus murinus) were obtained from BACPAC Resources at the Children' s Hospital Oakland Research Institute (http: //bacpac. chori. org/home. htm). Filters were screened with gene probes for human KIR2DL4, LILRA2, NCR1, and rhesus macaque KIR3DX1 (LRC region), human CD94, NKG2A, and Ly49L (NKC region), catta BAT1, POU5F1, TCF19, and rat Gnl1, Cat56, Trim39, Trim26, Trim10, Trim15, Ppp1r11, Mog, and mouse lemur MHC class I (MHC class I region). Screening with radioactively labelled probes was done according to the supplier' s recommendations. BAC clone DNA was purified by CsCl density centrifugation and isolated DNA was sheared by sonification. DNA fragments of 1. 5–3. 5 kb were selected and cloned into plasmid pUC19. Inserts were amplified by PCR using insert-flanking M13 forward and reverse primers and amplificates were sequenced with Applied Biosystems BigDye terminator chemistry and analyzed in ABI3730xl sequencers (Applied Biosystems). Raw sequences were processed by Phred (www. phrap. org) and assembled into a contiguous sequence by Phrap (www. phrap. org). Both programs are available from Phil Green, University of Washington. Exons and introns of genes were identified in finished BAC clone sequences manually or by BLAST and FGENESH-2 algorithms (http: //www. ncbi. nlm. nih. gov/BLAST/; http: //www. softberry. com/all. htm). All BAC clone sequences have ‘finished sequencing’ quality (1 putative mistake/100,000 bases) and the data have Phred values of 60. Exons 4 to 6 of the grey mouse lemur CD94, NKG2, and Ly49L genes were amplified from genomic DNA of 12 (all genes) and additional 34 (only CD94 genes) unrelated free-living grey mouse lemur individuals from the Kirindy region in Madagascar. Specific primers (Table S2) were designed to be located in exon-flanking introns. PCR products were completely sequenced on both strands and SNPs were identified. Primers were designed for amplification of conserved regions of potto and tarsier CD94 exon 4, intron 4 and exon 5 (Table S2). Genomic DNA was used of a single potto (Perodicticus potto) and a single Philippine tarsier (Tarsius syrichta) individual. PCR products were cloned and sequenced. For every CD94 sequence at least two identical clones were identified. DDBJ/EMBL/GenBank database accession numbers for BAC clones of MHC class I regions (AB480748, FP236831, FP236832, FP236833, FP236839), NKC region (haplotype 1: FP236838, haplotype 2: FP236834), LRC region (CR974412, CR974436, CR974413), and cDNA of grey mouse lemur and ruffed lemur CD94, NKG2 and Ly49L sequences as well as potto and tarsier CD94 sequences were assigned accession numbers FJ869057-FJ869114. Grey mouse lemur KIR3DX1 and DAP12 cDNA sequences are found under FJ882074-FJ882079. Total RNA was extracted from a liver sample of a grey mouse lemur and a blood sample from a further grey mouse lemur housed at UMR CNRS/MNHN in Brunoy (France) and a blood sample from a black-and-white ruffed lemur (Varecia variegata variegata) housed at the German Primate Center. Reverse transcription was performed with M-MLV reverse transcriptase (Promega). Based on the mouse lemur BAC sequences primers were designed to obtain the complete open reading frames of mouse lemur and ruffed lemur NKC genes. cDNA sequences of the ruffed lemur were completed by 5′- and 3′-RACE PCR with the GeneRacer Kit (Invitrogen). All cDNA sequences were cloned and sequenced. For interaction studies, grey mouse lemur CD94 and NKG2 cDNA sequences were inserted into the pcDNA3. 1/NT-GFP-TOPO and the pcDNA3. 1/V5-His-TOPO vector (Invitrogen), respectively. For cell surface expression, the three CD94 cDNA sequences were expressed without GFP at the NH2 terminus, but with a FLAG tag (DYKDDDDK) at the COOH terminus for extracellular detection. The grey mouse lemur DAP12 cDNA was isolated by RT-PCR. The encoded DAP12 was fused at the COOH terminus with the c-myc peptide epitope (EQKLISEEDL). Primer sequences and performed PCRs are listed in Table S2. CD94 and NKG2 molecules were tagged with GFP and with V5 for co-immunoprecipitation experiments. To avoid any potential unwanted interactions between the tags, GFP was fused to the amino terminus of CD94 (= intracellular localisation) and V5 to the carboxy terminus of NKG2 molecules (= extracellular localisation). Expression constructs were transiently transfected in the human 293T cell line using Metafectene (Biontex GmbH). 24 h after transfection, cells (1×107 cells/ml) were lysed in NP-40 lysis buffer containing 0. 1% NP-40,50 mM Tris, pH 7. 6,150 mM NaCl, 4 mM EDTA and protease inhibitors (Roche). Lysates were incubated overnight at 4°C with monoclonal mouse anti-GFP antibody (Clontech) and then incubated with protein G Sepharose 4 Fast Flow beads (GE Healthcare) for 3 h. Sepharose beads were washed five times with lysis buffer and bound proteins were eluted with loading buffer at 95°C for 10 minutes. Samples were electrophoresed in 10% polyacrylacrylamide gels and transferred to nitrocellulose membrane. Western blotting was performed with HRP-coupled monoclonal mouse anti-V5 antibody (Invitrogen) to test for co-immunoprecipitated V5-tagged NKG2 molecules. Deglycosylation was performed with PNGase F according to supplier' s recommendations (New England Biolabs). For analysis of CD94/NKG2 receptors on the cell surface, CD94 was FLAG-tagged at the carboxy terminus (= extracellular localisation). Respective expression constructs of CD94 together with the V5-tag NKG2 constructs (see above) were transiently transfected in 293T cells. Cell surface expression of CD94/NKG2-2 and CD94/NKG2-5 receptors were tested by additional transient transfection with c-myc-tagged mouse lemur DAP12 expression constructs. As control, single transfections with either CD94 or NKG2 expression constructs were performed. Potential formation of heterodimers of grey mouse lemur NKG2 and human CD94 (from 293T cells) was excluded by testing with an anti-human CD94 antibody (AbD Serotec). 24 hours post transfection cells were detached from cell culture dishes and washed twice with 1× PBS. Cells were stained with APC-labelled anti-V5 (Abcam) and FITC-labelled anti-FLAG (Sigma-Aldrich) monoclonal antibodies for surface expression of NKG2 and CD94. In parallel experiments, expression of DAP12 was monitored with an anti-c-myc mouse monoclonal antibody (Sigma-Aldrich) by intracellular staining of cells, which were previously fixed with 1. 5% paraformaldehyde (Merck) and permeabilized with 0. 25% saponin (Roth), and binding was detected by a PE-Cy5-labelled goat anti-mouse IgG polyclonal antibody (Santa Cruz Biotechnology). After washing twice with 1× PBS cells were resuspended in 200 µl of 1× PBS and 50,000 events were measured. Living cells were gated based on forward and side scatter characteristics and analysed for APC and FITC staining. All samples were analysed in a LSRII flow cytometer (BD). Data were acquired with BD FACS Diva 5. 1 software (BD) before analysis with FlowJo 7. 2. 9 software (TreeStar). CD94, NKG2 and MHC class I nucleotide sequences were aligned with MAFFT [35] and corrected manually. Phylogenetic analyses were conducted using three methods: neighbor-joining (NJ), parsimony and Bayesian phylogenetics. NJ analyses were performed with MEGA3. 1 [36] using the Tamura-Nei method with 1,000 replicates. PAUP*4. 0b10 [37] and the tree bisection-reconnection branch swapping algorithm were used for parsimony analyses with 1,000 replicates and a heuristic search. For the Bayesian analysis, the model of DNA substitution was selected using MODELTEST3. 7 [38] and the Akaike information criterion. Bayesian phylogenetic analyses were conducted with MRBAYES3. 1. 2 [39]; sampling was performed with one cold chain and three heated chains, which were run for 106 generations or until average standard deviation of split frequencies was <0. 01. Trees were sampled every 200 generations and the first 2,500 trees were discarded before a consensus tree was generated; three simultaneous runs were conducted and average standard deviation of split frequencies was always <0. 01. The tree topologies obtained with the three methods were compared with PAUP*4. 0b10 using the Shimodaira-Hasegawa test of alternative phylogenetic hypotheses with re-sampling estimated log-likelihood optimization, and 10,000 bootstrap replicates; in all analyses the test failed to reject any of the alternative tree topologies (α = 0. 05). This comparison was made with the maximum likelihood model defined by MODELTEST. A maximum-likelihood analysis was also performed for the study of the NKG2 cytoplasmic and transmembrane sequences using RAxML7 [40] under the GTR+CAT model with 1,000 replicates (rapid bootstrapping). dN/dS (ω) ratios for CD94 and NKG2 lectin-like domains were estimated using PAML v3. 15 [41] with the F3 X 4 model of codon frequencies. Bayesian tree topologies were used for these analyses and three sets of likelihood ratio tests were conducted to compare null models that do not allow ω>1 (M1a, M7 and M8a) with models that do (M2a and M8). Significance was assessed by comparing twice the difference in likelihood between the models (2ΔL) to a χ2 distribution with one (M8a/M8) or two (M1a/M2 and M7/M8) degrees of freedom. Codons with ω>1 were identified using the Bayes Empirical Bayes approach [42]. The distribution of the selected sites in the CD94/NKG2 lectin-like domains was studied using a binomial distribution: considering Ω = (0,1, 2, …, n), ∀k ∈ Ω, p = (X = k) = nCk * pk * qn−k. This indicates for example that under a random distribution it is unlikely to have more than 5 of the 21 positively selected sites in the ligand binding region that represents 28 of the 239 total sites (α = 0. 05); so the fact that we observe 7 of the 21 positively selected sites in this region indicates a distribution biased toward this region (α = 0. 01).

Most receptors of natural killer (NK) cells interact with highly polymorphic major histocompatibility complex (MHC) class I molecules and thereby regulate the activity of NK cells against infected or malignant target cells. Whereas humans, apes, and Old and New World monkeys use the family of killer cell immunoglobulin-like receptors (KIR) as highly diverse NK cell receptors, this function is performed in rodents by the diverse family of lectin-like receptors Ly49. When did this functional separation occur in evolution? We followed this by investigating lemurs, primates that are distantly related to humans. We show here that lemurs employ the CD94/NKG2 family as their highly diversified NK cell receptors. The CD94/NKG2 receptors also belong to the lectin-like receptor family, but are rather conserved in “higher” primates and rodents. We could further demonstrate that lemurs have a single Ly49 gene like other primates but lack functional KIR genes of the KIR3DL lineage and show major deviations in their MHC class I genomic organisation. Thus, lemurs have evolved a “third way” of polymorphic and diverse NK cell receptors. In addition, the multiplied lemur CD94/NKG2 receptors can be freely combined, thereby forming diverse receptors. This is, therefore, the first description of some combinatorial diversity of NK cell receptors.

Abstract Introduction Results Discussion Materials and Methods

genetics and genomics/comparative genomics genetics and genomics/genetics of the immune system

2009

A Novel System of Polymorphic and Diverse NK Cell Receptors in Primates

Gammaherpesviruses such as KSHV and EBV establish lifelong persistent infections through latency in lymphocytes. These viruses have evolved several strategies to counteract the various components of the innate and adaptive immune systems. We conducted an unbiased screen using the genetically and biologically related virus, MHV-68, to find viral ORFs involved in the inhibition of type I interferon signaling and identified a conserved viral dUTPase, ORF54. Here we define the contribution of ORF54 in type I interferon inhibition by ectopic expression and through the use of genetically modified MHV-68. ORF54 and an ORF54 lacking dUTPase enzymatic activity efficiently inhibit type I interferon signaling by inducing the degradation of the type I interferon receptor protein IFNAR1. Subsequently, we show in vitro that the lack of ORF54 causes a reduction in lytic replication in the presence of type I interferon signaling. Investigation of the physiological consequence of IFNAR1 degradation and importance of ORF54 during MHV-68 in vivo infection demonstrates that ORF54 has an even greater impact on persistent infection than on lytic replication. MHV-68 lacking ORF54 expression is unable to efficiently establish latent infection in lymphocytes, although it replicates relatively normally in lung tissues. However, infection of IFNAR−/− mice alleviates this phenotype, emphasizing the specific role of ORF54 in type I interferon inhibition. Infection of mice and cells by a recombinant MHV-68 virus harboring a site specific mutation in ORF54 rendering the dUTPase inactive demonstrates that dUTPase enzymatic activity is not required for anti-interferon function of ORF54. Moreover, we find that dUTPase activity is dispensable at all stages of MHV-68 infection analyzed. Overall, our data suggest that ORF54 has evolved anti-interferon activity in addition to its dUTPase enzymatic activity, and that it is actually the anti-interferon role that renders ORF54 critical for establishing an effective persistent infection of MHV-68. Virus infection induces numerous immune responses in the host, the earliest of which is the innate immune response [1], [2]. The innate immune response is comprised of many layers of non-specific defense, including anatomical barriers, such as skin and mucosa, the complement system, inflammation, and various cells, such as natural killer cells, phagocytes, mast cells, macrophages, dendritic cells, neutrophils, and basophils [3]–[5]. The innate immune response plays a crucial role in shaping the ensuing adaptive immune response, in part by the production of cytokines in response to infection [2], [6]. Interferons (IFN) are cytokines secreted upon virus infection that induce the expression of a variety of antiviral gene products, reducing virus replication and further infection [1], [7]–[9]. Interferons are classified as type I and II, as defined by the cell types able to produce them and the receptors they bind to [1]. Unlike the type II IFN-γ that is produced by specific cells of the immune system, IFN-α and IFN-β are type I IFNs that can be produced in most cell types [10]. Mammals encode a single IFN-β and several IFN-α species. All type I IFN species bind to the same ubiquitously expressed receptor, called the type I interferon receptor, or IFNAR [11]. This receptor is a heterodimer comprised of IFNAR1 and IFNAR2 [12]. Although normally unassociated, IFNAR1 and IFNAR2 dimerize upon the binding of IFN-α or IFN-β first to IFNAR2, and then to both receptors in the dimer [13]. IFNAR1 and IFNAR2 are each pre-associated with one of the members of the Janus protein tyrosine kinase family, where TYK2 is associated with IFNAR1 and JAK1 with IFNAR2. IFN binding and formation of the complete IFNAR dimer leads to cross-phosphorylation of TYK2 and JAK1, and the phosphorylation of the IFNAR chains they are permanently associated with. These phosphorylation events set up a platform for the recruitment of STAT1 and STAT2, which in turn are also phosphorylated. The phosphorylated STAT proteins dimerize prior to joining with IFN-regulatory factor 9 (IRF9) to form the Interferon-Stimulated Gene Factor 3 (ISGF3γ) transcription factor, which translocates to the nucleus where it induces the expression of hundreds of interferon stimulatory genes (ISG) (reviewed in [1]). Herpesviruses are large, double-stranded DNA viruses defined by their ability to persist for the lifetime of the host by establishing latent infections and by evading the host immune response [14]. Both lytic and latent infections of herpesviruses are able to directly cause disease [15]–[23]. Gaining understanding of the mechanisms by which herpesviruses maintain persistent infections and evade immune surveillance is a key step in controlling the diseases they are associated with. Herpesviruses are subdivided into alpha, beta, and gamma herpesviruses [14]. The human gammaherpesviruses are Epstein-Barr virus (EBV) and Kaposi' s sarcoma-associated herpesvirus (KSHV). Both EBV and KSHV are associated with malignancies; EBV with Burkitt' s lymphoma, nasopharyngeal carcinoma, and oral hairy leukoplakia and KSHV with Kaposi' s sarcoma, primary effusion lymphoma, and multicentric Castleman' s disease [23]–[28]. We use the biologically and genetically related virus, murine gammaherpesvirus-68 (MHV-68) as a model to study the human gammaherpesviruses [29]–[32]. Like KSHV, MHV-68 is a gamma-2-herpesvirus. MHV-68 establishes lytic and latent infections in mice [33], replicates readily in in vitro cell culture systems, and has a genome that can be genetically manipulated by utilizing a bacterial artificial chromosome (BAC) system [34]. In KSHV and MHV-68 several studies have identified viral proteins involved in the inhibition of the host innate and adaptive immune responses. In particular, KSHV open reading frame (ORF) 10 binds to JAK and STAT proteins to block IFN mediated signaling [35]. KSHV and MHV-68 ORF36 bind to phosphorylated IRF3, thus inhibiting the production of IFN-β [36]. KSHV ORF45 interacts with and inhibits IRF7 from entering the cell nucleus [37]. The KSHV immediate early transcription factor and E3 ligase, RTA, targets IRF7 for protein degradation [38]. Other viral ORFs also contribute to immune evasion by inducing the downregulation of surface molecules critical for immune activation. K5 of KSHV directs the downregulation of Tetherin/BST2, ICAM, and MHC class I [39]–[42]. Furthermore, K3 of both KSHV and MHV-68 inhibit the surface expression of MHC class I [39], [43]. 2′-deoxyuridine 5′-triphosphate pyrophosphatase (dUTPase) reduces the misincorporation of uracil in the DNA genome by controlling the level of dUTP through conversion of dUTP into dUMP, ultimately leading to an increased amount of dTTP and a lower dUTP∶dTTP ratio [44]. This enzyme can be found in all classes of organisms and in many RNA and DNA viruses as well [45], [46]. In this study, we define the role of viral ORF54, a functional dUTPase, in evading the host innate immune response to virus infection. In an effort to understand these two separate functions of ORF54, we analyzed the signaling pathways altered by wild-type ORF54 and a dUTPase-null mutant. Through in vitro and in vivo infection, in wild-type and transgenic mice, with recombinant MHV-68 harboring mutations in ORF54, we found that ORF54 interferes with-type I interferon signaling, which further affects persistent infection and the establishment of latency. To systematically identify MHV-68 viral ORFs that inhibit type I IFN signaling, we conducted a screen where 293T cells were transiently transfected with a reporter construct containing firefly luciferase driven by the interferon-stimulated response element (ISRE_firefly-luciferase). Cells were also co-transfected with a reporter construct containing renilla luciferase driven by the constitutively active PGK promoter and each MHV-68 viral ORF or a vector control. Transfected cells were treated with human IFN-α and the induction of the ISRE reporter was measured by dual luciferase assay. Since this screen examines cellular responses after IFN-α treatment, the viral proteins previously identified to inhibit IRF3 or IRF7 signaling, thus preventing the induction of type I IFN production, would not necessarily be identified. Of all the MHV-68 viral ORFs that were screened, we found 8 ORFs that were potentially able to inhibit type I IFN signaling to a level that is 50% of the activation seen with vector control. Two of them are M2 and M8, ORFs specific to MHV-68. M2 has been previously shown to inhibit IFN signaling [47], thus validating our screen. Among the other 6 ORFs that also have homologues in KSHV and EBV, ORFs 10,11, and 54 are particularly interesting because a previous sequence analysis study identified a shared dUTPase-related domain, although only ORF54 contains catalytic active sites [48]. Since ORF54 is a viral dUTPase and is one of the top three strongest inhibitors, our following study focuses on its potential anti-IFN function and the biological significance of this function during viral infection. Cells transfected with ORF54 demonstrated a diminished activation of ISRE following treatment with IFN-α (Figure 1), with only 20% of the activation seen in control transfections. KSHV ORF54 also demonstrated diminished activation of ISRE, at 23% of control (Figure 1), suggesting that the ability to inhibit type I IFN signaling is a gammaherpesvirus conserved function of ORF54. To test whether the dUTPase function is required for the anti-IFN activity of ORF54, we constructed a catalytic domain mutant of MHV-68 ORF54 by replacing the amino acid histidine at position 80 with alanine and the amino acid aspartic acid at position 85 with asparagine (ORF54 H80A/D85N). These two amino acids were chosen for mutagenesis due to their proximal location in the putative active site of ORF54 dUTPase and their predicted importance for enzymatic reaction. Viral and non-viral dUTPases typically share five highly conserved motifs, although the arrangement found in herpesviruses is different compared to human dUTPase [44], [48]. Several studies have identified the presence of dUTPase motif III, which is critical for catalytic activity, in the herpesvirus dUTPases [44]. Aspartic acid at position 85, located in motif III, was altered because aspartic acids at positions 84 and 86 in human endogenous retrovirus (HERV-K) were found to be critical for catalytic activity, but not for dUTP binding [44]. Histidine at position 80 was chosen because it is conserved in gammaherpesviruses, and in EBV a histidine at position 71 contains a necessary imidazole group [49]. The ORF54 H80A/D85N mutant demonstrates a complete loss in enzymatic activity, although the protein expression level remained the same (Figure 2). When co-transfected with ISRE_firefly-luciferase, ORF54 H80A/D85N maintains the ability to diminish IFN-α induced activation of the ISRE to 32% of the activation seen in control samples (Figure 1). This result suggests ORF54 inhibition of the type I IFN signaling cascade is independent of its dUTPase enzymatic activity. To further clarify if the dUTPase function was sufficient to inhibit type I IFN signaling, we tested the ability of murine cellular dUTPase to inhibit IFN-α induced activation of ISRE in our reporter assay. The cellular dUTPase was unable to block activation of ISRE_firefly-luciferase (Figure 1), further indicating that dUTPase enzymatic activity does not necessarily correlate with anti-IFN activity. The type I IFN signaling pathway begins with the binding of IFN-α to the surface IFNAR and results in the production of ISGs [7], [8]. As each step in the JAK/STAT pathway that ensues is well defined [1], we aimed to identify the step where ORF54 exerts its function. We first assayed a central event in type I IFN signaling, the phosphorylation of STAT1 protein. 293T cells ectopically overexpressing MHV-68 ORF54 or ORF54 H80A/D85N both demonstrated a reduced level of phosphorylation of STAT1 following treatment by IFN-α in comparison to cells transfected with vector control or an unrelated MHV-68 ORF (Figure 3A). By assay of steps upstream of the phosphorylation of STAT1, we show that cells expressing MHV-68 ORF54 or ORF54 H80A/D85N both demonstrated a reduced level of total IFNAR1 (Figure 3B). As a control for specificity, we also found that ORF54 and ORF54 H80A/D85N do not alter levels of the surface protein type I insulin-like growth factor receptor-β (IGF1β) or IFNAR2 (Figure 3B). These results suggest that ORF54 induces the degradation of IFNAR1 independently of its dUTPase enzyme activity, and that this degradation results in a reduction of the type I IFN response, including the phosphorylation of STAT1. The transcript level of IFNAR1 remains comparable between the vector control and viral ORF transfected cells (Figure 3C), suggesting that the ORF54 induced reduction of IFNAR1 is at the protein level. We generated three recombinant MHV-68 to study the importance of ORF54 (Figure 4A). Because ORF54 is not required for the virus to replicate in cultured cells [50], the first mutant is an ORF54-null virus that has triple translational stop codons inserted near the N-terminus of MHV-68 ORF54 (54Stop). The second virus is a revertant for 54Stop, where the translational stop codons have been removed and reverted back to wild type (54R). This virus ensures that any phenotype demonstrated with 54Stop is due to the ORF54-null mutation and not any additional recombination or unintentional mutagenesis of the MHV-68 viral genome. The third virus (54DM) has the same two amino acid mutations that abolish dUTPase activity as in the ORF54 H80A/D85N expression construct (Figure 4A). NIH3T3 cells were infected with an MOI of 2 to initiate single-step growth kinetics. 24 hours post infection, cells were either treated for 15 minutes with 500 units/mL of mouse IFN-α or left untreated. Uninfected cells were also treated as a control. Immunoblots probing for the phosphorylation of STAT1 protein demonstrate that while wild-type (WT) MHV-68,54DM, and 54R are able to block STAT1 phosphorylation, higher levels of phosphorylated STAT1 are found in cells infected with 54Stop (Figure 4B). Total STAT1 is comparable in all samples and all cells are infected to a similar level, as demonstrated by equal production of the MHV-68 late protein ORF65. Therefore, this result supports the role of ORF54 in the inhibition of type I IFN signaling during viral infection, as lacking ORF54 partially alleviates the block. However, we also noted that the phosphorylated STAT1 seen in cells infected with 54Stop is still not at the level observed in uninfected and treated cells. This is perhaps due to the other viral ORFs that are present and still able to inhibit the type I IFN signaling pathway, as evidenced by the multiple ORFs identified in our original screen. As an additional control to demonstrate that infected cells are still capable of responding to stimuli, NIH3T3 cells were identically infected with an MOI of 2 and treated at 24 hours post infection with 160 µM prostratin (12-deoxyphorbol 13-acetate) for 30 minutes (Figure 4B). Prostratin initiates a signaling cascade that induces the reactivation of latent viruses [51]. In this control assay, immunoblots probing for the phosphorylation of p44/p42 (Erk1/2) demonstrate that infected cells are viable enough to respond to stimuli even 24 hours post infection. These studies demonstrate that although at this MOI the 54Stop virus can establish a robust viral infection, the lack of functional ORF54 makes it unable to effectively block the resulting induction of the type I IFN signaling cascade. Since we showed the degradation of IFNAR1 in the presence of overexpressed ORF54 (Figure 3B), we assayed this phenotype during virus infection to determine the biological relevance. By infecting cells with WT MHV-68 and 54R we found that IFNAR1 is degraded (Figure 4C). This degradation is reduced in cells infected with 54Stop virus, suggesting that ORF54 is required for the virus to induce degradation of IFNAR1. Furthermore, because cells infected with 54DM virus show a similar level of IFNAR1 reduction as with WT or 54R infection, this indicates that ORF54 dUTPase enzymatic function is not required for the degradation of IFNAR1 and the inhibition of the type I IFN signaling cascade. Interestingly, unlike with STAT1 phosphorylation, we observed a comparable level of IFNAR1 between 54Stop-infected cells and uninfected cells, indicating that ORF54 is the sole viral protein responsible for IFNAR1 degradation during viral infection. As a control for specificity, we also found that virus infection does not alter the levels of IFNAR2 or the surface protein IGF1-β (Figure 4C). Therefore, ORF54 mediated degradation is not a general phenomenon for all surface proteins and IFNAR1 is a specific target for such ORF54 function. The transcript level of IFNAR1 remains comparable amongst cells infected with WT, 54Stop, 54DM, and 54R viruses (Figure S1), suggesting that ORF54 does not alter the transcription of IFNAR1 and the ORF54 induced degradation of IFNAR1 is at the protein level. Because we found 54Stop defective in inhibiting type I IFN signaling due to its inability to induce IFNAR1 degradation, we assayed the downstream induction of ISGs following infection of bone marrow derived macrophages. Macrophages were chosen for infection due to their high endogenous induction of anti-viral genes following virus infection. 24 hours post infection at an MOI of 2, cells were harvested for immunoblot against the ISG IFIT2. All infected cells show induced expression of IFIT2 compared to uninfected cells. IFIT2 protein expression was highest in 54Stop infected cells compared to in cells infected with WT, 54DM, and 54R (Figure 5A), suggesting a virus lacking functional ORF54 is not as effective in blocking ISG induction as WT MHV-68. Additionally, total RNA was harvested from infected macrophages to measure the transcript level of several ISGs, including MX1, IFIT1, and IFIT3. In all cases, the ISG induction in cells infected with 54Stop is higher than with infection by WT, 54DM, and 54R, suggesting ORF54 has a role in inhibition of type I IFN responses (Figure 5B). Although ORF54 is not essential for MHV-68 replication, the 54Stop virus exhibits a defect in STAT1 activation and downregulating IFNAR1 expression. Hence, we speculated that lack of ORF54 might have some effect on viral replication in cells that are capable of producing and responding to type I IFN, such as NIH3T3 cells. Indeed, as shown in Figure 6A, we found that the 54Stop virus has moderately attenuated multiple-step growth in NIH3T3 cells (approximately 3. 8-fold on day 4 and 4. 1-fold attenuation on day 5 post infection compared to WT). Moreover, this attenuation of the 54Stop virus was not observed in Vero cells (Figure 6A), which are unable to produce type I IFN in response to infection [52]. To ensure this difference in phenotype of the 54Stop virus is due to its inability to block IFN signaling, we infected bone marrow derived macrophages from wild-type and IFN α/β receptor knockout (IFNAR-/-) mice at an MOI of 4. By immunoblot of infected wild-type macrophages, we see a lower production of the MHV-68 capsid protein, ORF65, with 54Stop infection compared to WT, 54DM, and 54R infection. However, the ORF65 production with 54Stop infection is rescued in the IFNAR−/− macrophages (Figure 6B). Similarly, the infectious titer produced from wild-type macrophages infected with 54Stop is between 6. 3- to 7. 5-fold lower than that of WT, 54DM, and 54R, while the infectious titers produced from IFNAR−/− macrophages are similar amongst all four virus types (Figure 6C). To further demonstrate the role of ORF54 in antagonizing IFN signaling during viral replication, virus production in NIH3T3 cells with or without the treatment of 100 units/mL of IFN-α was compared (Figure 6D). The peak viral titers of WT, 54DM, and 54R viruses were relatively unaffected at this low dose, while the average drop for 54Stop was 3-fold, p-value = 0. 022. The growth defects observed with 54Stop in the presence of type I IFN response demonstrate the role of ORF54 in antagonizing the signaling pathway. The consistent, but modest, defects seen with 54Stop infection are also meaningful as it is known the virus has multiple anti-IFN genes [34]–[38], [53], [54]. Additionally, all of above in vitro viral replication results demonstrate that 54DM behaved similarly to WT and 54R viruses but not to 54Stop, indicating that the anti-IFN function of ORF54 is independent of its dUTPase enzymatic function. We examined the role of ORF54 in vivo by infecting Balb/C mice with WT MHV-68,54Stop, 54DM, or 54R and measuring lytic replication in the lungs on 5 and 7 days post infection (dpi) and the establishment of latency at its peak on 14 dpi. Lytic infection does not critically require ORF54. With intranasal infection we found that lytic replication in the lungs at 5 dpi appears to be only slightly affected by the lack of ORF54. The average infectious titer was approximately 3. 5- to 3. 8-fold lower with 54Stop infection than with WT MHV-68,54DM, or 54R, with p-values 0. 047,0. 002, and 0. 018, respectively (Figure 7A). At 7 dpi the effects from the lack of ORF54 on the lung viral titers were even smaller (1. 2- to 1. 8-fold lower than others), and possibly insignificant with p-values of 0. 048 for WT, 0. 396 for 54DM, and 0. 022 for 54R (Figure 7B). Similar results and trends were obtained with analysis of the viral genome copy number in the lung lysates (Figure S2). Infection of 54DM was comparable to WT and 54R viruses, indicating that function (s) other than dUTPase activity of ORF54 play a role, however minor, during productive infection in the lung. At 14 dpi, we analyzed the establishment of splenic latency by performing an infectious center assay and quantifying viral genome copy numbers of the spleens of mice infected with WT MHV-68,54Stop, 54DM, or 54R. The infectious center assay quantifies the amount of latent virus that reactivates from a population of B-cells upon ex vivo culturing. We found that the 54Stop virus has approximately 53- and 45-fold less reactivated virus per 107 lymphocytes compared to WT and 54R viruses, respectively (Figure 7C). Quantitation of the viral genome copies is an unbiased measurement that does not rely on virus activity in the assay itself, and it showed an even more pronounced reduction. The 54Stop genome is almost undetectable, with 140- to 164-fold lower genome copies than WT MHV-68,54DM, and 54R (Figure 7D). Lower levels of viral genome copies and infectious centers suggest the defect of the 54Stop virus is in the establishment of latency and not in the ability to reactivate. If this drastic reduction in 54Stop latency is due to the inability of the virus to block type I IFN signaling, we would expect to see latency of 54Stop rescued in IFNAR−/− mice, which lack the type I IFN receptor. Indeed, in IFNAR−/− mice splenic latency of 54Stop at 14 dpi is greatly increased to a level similar to 54R virus; the number of infectious centers with 54Stop infection is 2. 8-fold lower than with 54R, but with a statistically insignificant p-value of 0. 248 (Figure 7E). Viral genome copies of 54Stop and 54R viruses from IFNAR−/− mice splenocytes are nearly identical (Figure 7F). Our results indicate that ORF54 is required for establishing a latent infection, and this requirement is based on its anti-IFN activity as the deficiency of 54Stop is rescued in mice unresponsive to type I interferon. This result concludes that the major role of ORF54 for establishing and/or maintaining latency is to inhibit type I interferon responses. Viral pathogens have adapted several avenues of immune evasion, including inhibition of the innate immune response. The importance of the interferon system is highlighted by the multiple evasion strategies employed by viruses, such as herpesviruses [14], [55]. Here we present a report on ORF54, a conserved viral dUTPase, identified by our unbiased screen designed to isolate gammaherpesvirus ORFs involved in the inhibition of the type I IFN induced signaling pathway. Interestingly, ORF54 enzymatic activity proves dispensable for its anti-IFN function. Using KSHV ORF54 we were further able to demonstrate the anti-IFN role of ORF54 is conserved amongst KSHV and MHV-68. We have found that when high levels for ORF54 are present, either by ectopic expression or by infection with MHV-68, the total amount of cellular type I interferon receptor 1 is reduced, causing depression of the type I IFN response. Finally, through manipulation of the viral genome and murine host, we uncovered the biologically relevant function that ORF54 plays in not only blocking a critical component of the innate immune response, but also in persistent infection of gammaherpesviruses. The most surprising and significant finding of our study is that the primary role of ORF54 during MHV-68 infection of cells and mice appears to be its anti-IFN function rather than its dUTPase activity. In three separate cell culture systems where type I IFN signaling is functioning, we demonstrate attenuation of the 54Stop virus, but not 54DM, a dUTPase-null virus (Figure 6). None of the defects seen with the 54Stop virus are observed with 54DM infection, indicating that the dUTPase activity is dispensable for the role of ORF54 during MHV-68 replication in cultured cells. In mice, while the 54Stop virus has a relatively normal productive infection in the lung, it shows a very strong deficit in spleen latency that is not found in the dUTPase-null virus (Figure 7C, 7D). However, this deficit is largely rescued in IFNAR−/− mice (Figure 7E), supporting the conclusion that it is the anti-IFN but not the dUTPase activity of ORF54 that plays a major role for MHV-68 establishment of latent infection of mice. Across eukaryotes, prokaryotes, and viruses, dUTPases are highly conserved proteins, especially at the structural level [56]. Most eukaryotic dUTPases have 5 conserved motifs ordered 1–5 from the N- to C- terminal [45], [48], [49]. Functional dUTPases are formed as homotrimers containing three complete active sites formed from motifs 1–5 of each protein [45], [49]. However, the herpesvirus monomer is an active dUTPase with one active site formed by motifs 1–5 [48], [57]. The herpesvirus dUTPase protein sequence is twice as large as the cellular, with the C-terminal half containing domains 1,2, 4, and 5 [48] and a herpesvirus unique domain called motif 6 between motifs 2 and 4. Towards the center of the protein, there is an actual motif 3, that was maintained after the loss of motifs 1,2, 4, and 5 from the N-terminal portion [45], [48]. By examining the ability of murine cellular dUTPase to inhibit type I IFN signaling in our reporter assay, we found that dUTPase activity alone does not diminish IFN-α induced activation of ISRE, while an ORF54 dUTPase-null mutant still does (Figure 1). Therefore, it is possible that the N-terminal half of ORF54 may have evolved the additional function of type I IFN inhibition. For the gammaherpesviruses MHV-68, KSHV, and EBV, the ORF54 motif 3 starts at amino acid 82,78, and 73, respectively. Since the motifs required for active site formation are found by the C-terminal halves of each protein, the N-terminal portion upstream of motif 3 in herpesviral dUTPases contain no recognizable or conserved sequences from cellular dUTPases. The sequences in the N-terminal half of ORF54 demonstrate some conservation between the gammaherpesviruses examined, but further studies designed to functionally map ORF54 anti-IFN activity are required to identify such a domain. ORF54 is an early protein [58] that, like other viral dUTPases, is expressed in both the cytoplasm and nucleus of infected cells, while cellular dUTPases are only found in the nucleus and mitochondria of cells [59]. Using a recombinant MHV-68 with a FLAG epitope tag on ORF54 to study its expression kinetics, we found that during early stages of infection ORF54 appears evenly distributed, but has preference for the nucleus at late stages of virus replication when cytopathic effect is obvious (Figure S3). As the MHV-68 genome is replicated in the nucleus, perhaps ORF54 functions as a dUTPase in the nucleus to help maintain genomic integrity and as an inhibitor of type I IFN signaling in the cytoplasm. It would be informative to identify ORF54 protein interaction partners in both compartments and at the various stages of virus infection. Interestingly, beta- and gammaherpesviruses contain several ORFs with dUTPase-related domains that demonstrate strong divergence, where besides their catalytic motifs they exhibit different functions [48], some involving innate immune responses. Betaherpesviruses are unique in that many do not have a single ORF with dUTPase catalytic activity [60], [61]. However, several betaherpesvirus ORFs exist with dUTPase-related domains, such as UL72, UL82, UL83, UL84, and UL31 [48]. Human cytomegalovirus (HCMV) encodes pp65, also called UL83, that contains the dUTPase-related domain motif 6 [48]. HCMV pp65 expression led to the inhibition of phosphorylation of IRF3 and to its sequestration in the cytoplasm following induction of the IFN pathway [62], [63]. In gammaherpesviruses, although ORF54 is the only functional dUTPase, ORFs 10 and 11 both contain the dUTPase-related domain motif 6 [45], [48]. Here we showed in a transient transfection reporter assay that KSHV ORF54 is capable of reducing IFN-α responses (Figure 1). However, EBV ORF54 was found to induce expression of several pro- and anti-inflammatory cytokines when cells were treated with purified EBV ORF54 [64]–[67]. The biological significance of immune modulation by EBV ORF54 in the context of virus infection remains to be determined. KSHV ORF10 is a viral lytic protein that blocks IFN signaling by forming inhibitory complexes with JAK1 and TYK2 [35]. EBV ORF11 (LF2) was found to bind to IRF7 to inhibit dimerization and IRF7-mediated activation of type I IFN production, in a manner unrelated to its dUTPase domain [54]. Previously, we also found that MHV-68 ORF11 has a similar effect to ORF54 in inhibiting ISRE reporter induction by IFN-α [34]. Therefore, it is possible for large DNA viruses, such as herpesviruses, to utilize multiple proteins to inhibit type I interferon responses. This is further supported by the observation that STAT1 activation upon IFN-α treatment is not fully recovered in cells infected with the 54Stop virus (Figure 4B), indicating the presence of other viral proteins with overlapping anti-IFN functions. Our results have revealed a previously unrecognized and critical anti-IFN function of ORF54, but they also raise a question about the biological role of its dUTPase activity. Numerous RNA and DNA viruses, such as retroviruses, poxviruses, and herpesviruses, encode a functional dUTPase in their genomes, suggesting its importance for the viral life cycle [45], [68], [69]. However, viral dUTPases are generally dispensable for virus replication [50], [70], [71], likely because cellular dUTPase is readily available and active in most dividing cells. Many cellular enzymes involved in DNA replication and nucleotide metabolism are strongly cell-cycle dependent. Therefore, it is presumably advantageous for the virus to encode its own enzymes for genome replication in terminally differentiated and non-dividing cells where the cellular counterparts may not be available. For example, herpesviruses encode several other enzymes in addition to dUTPase, such as thymidine kinase (TK), exonuclease, Uracil-DNA glycosylase (UNG), and ribonucleotide reductase (RR). For MHV-68, these viral enzymes are not required for in vitro virus replication in actively dividing cultured cells, but disruption of TK or RR leads to severe attenuation in acute productive infection in the lung as well as in splenic latency [50], [72], [73]. However, because site-specific mutations targeting the enzymatic activity of these proteins were not used in these studies, it cannot be completely excluded that other possible mechanisms unrelated to their known TK or RR functions account for the observed severe attenuation. In this study, we constructed a dUTPase-null virus, 54DM, to study the specific contribution of the enzyme activity of ORF54. Interestingly, unlike with TK and RR, lack of dUTPase activity or even the entire ORF54 protein does not detrimentally retard virus replication in the lungs of infected mice. Previous studies in our lab have found that ORF54-null viruses constructed with a large transposon insertion cause a more significant defect on lytic replication in the lungs of infected mice [50]. This discrepancy may be due to the large disturbance each transposon causes in the highly compact MHV-68 viral genome, where most promoter regions overlap coding regions. The ORF54-null virus, 54Stop, shows a very strong defect in the infectious center assay that is not found with the dUTPase-null virus, implying that the lack of ORF54 enzymatic activity does not dramatically hinder the establishment of the MHV-68 latent load, while lack of the entire protein does. However, mice infected with 54DM do have a slight, but statistically insignificant, reduction in the amount of reactivated virus from infected splenocytes, at 2. 8- and 2. 4-fold defect compared to WT and 54R viruses, respectively (Figure 7C). This marginal defect of 54DM was not observed when the viral genome copy number was analyzed (Figure 7D), thus it is possible that the phenotype seen in the infectious center assay is due to the lack of dUTPase during reactivation. The 54Stop virus lacks any expression of ORF54, including its dUTPase activity, and the 2. 8-fold reduced level of infectious centers compared to 54R in IFNAR−/− mice mirrors that seen with 54DM virus during reactivation in Balb/C mice. Indeed, as with 54DM in wild-type mice, 54Stop has nearly identical viral genome copies as 54R in the splenocytes of IFNAR−/− mice, again suggesting that the minor defect in the infectious center assay is due to reactivation. Although it was expected that dUTPase-null viruses grow normally in actively dividing cell culture, it is surprising that the only phenotype we observed in mouse infection is a moderate defect in reactivation from latency. However, our observation is not unique. Following in vivo infection, HSV-1 dUTPase mutants replicate like wild type in the footpad, sciatic nerve, and dorsal root ganglia; a defect of about 10- to 100-fold is only visible when the virus moves to the CNS spinal cord. These dUTPase mutants were able to establish latent infections but demonstrated a defect in reactivation from latency [74]. Taken together, our results indicate that the dUTPase activity of ORF54 does not have a significant role during MHV-68 productive infection in the lung or latency in the spleen. However, our interpretation for the maintenance of dUTPase in the viral genome is that dUTPase activity is likely required for genome stability over many generations. Furthermore, reactivation is considered much less effective than lytic replication in permissive cells, thus a less optimal replication environment may have a more profound effect on the reactivation process over time. Gammaherpesviruses are characterized by their ability to establish latent infection in lymphocytes. Establishment of viral latency first requires the efficient infection of lymphocytes. By affecting persistent infection of gammaherpesviruses, the IFN response has far reaching effects on the viral life cycle and the establishment of latency, instead of only acting primarily on initial infection. Furthermore, the type I IFN response is critical in shaping the adaptive immune response [1], [2], [6], [9], [36]. Altering the balance between the host immune response and viral immune evasion genes could drastically affect the overall outcome of viral infection. The anti-IFN function of ORF54 appears to be relatively dispensable for lytic replication in vitro and in vivo, but absolutely required for the establishment of latency. As a whole, our data suggests that evading the type I IFN response is critical for the establishment of latent infection in lymphocytes and that ORF54 plays a major role in this evasion. The reduced latency observed in the infection of 54Stop may be due to the inability of the virus to replicate well in a particular cell type, such as lymphocytes, that is more sensitive or responsive to the anti-viral effects of IFN. Although several genes are likely required and may have some overlapping functions, lack of one of these genes, such as ORF54, retards establishment of latent infection. Our results in IFNAR−/− mice not only emphasize the importance of ORF54 in the type I IFN pathway by demonstrating a rescue in an infectious center assay, but also hint at the necessity of this host pathway in blocking the establishment of latent infection in lymphocytes. However, the mechanisms by which the virus spreads from the inoculation and lytic replication sites to the latency compartment remain largely elusive. Thus, although our data suggests its importance, we are currently unable to isolate at which stage during this viral spread the anti-IFN function of ORF54 is required. Analysis with a detailed time course and different cell types is required to further understand the interaction between type I interferon responses and latency establishment. Defining immune evasion strategies employed by MHV-68 will allow better design of a live attenuated vaccine to human gammaherpesviruses KSHV and EBV. One strategy to increase the success of a vaccine is to limit the establishment of latency and increase its immunogenicity by removal of viral immune evasion genes [75]. Because ORF54 is not required for virus replication, plays a role in type I IFN inhibition, and is necessary for the efficient establishment of latency, a virus lacking ORF54, as well as other immune evasion genes and genes required for latency, is a promising vaccination strategy. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Public Health Service (National Institutes of Health). The protocol was approved by the Institutional Review Board and Animal Research Committee of the University of California Los Angeles (assurance of compliance number: A3196-01). All procedures were performed under ketamine and xylazine anesthesia and all efforts were made to minimize suffering. 293T and Vero cells were cultured in complete Dulbecco' s modified Eagle medium (DMEM) containing 10% FBS. NIH3T3 cells were cultured in DMEM containing 10% BCS. Bone-marrow derived macrophages (BMDM) from both wild-type and IFNAR−/− mice were immortalized by v-raf and c-myc oncogenes, cultured in RPMI containing 10 mM HEPES, 10% FBS, and MCSF, and were a kind gift from Shankar Iyer (Genhong Cheng Lab, UCLA). All culture medium was also supplemented with penicillin and streptomycin and all cells were cultured at 37°C with 5% CO2. dUTPase assay was adapted and modified from previous methods used to measure enzymatic activity [44], [59], [76]. Briefly, each potential FLAG-tagged dUTPase construct was transfected into 293T cells using Lipofectamine 2000 (Invitrogen). 48 hours post transfection, cells were lysed with buffer containing 100 mM Tris pH 7. 5,50 mM NaCl, and 1% NP-40, supplemented with 1 mM PMSF, 1 µg/mL Aprotinin, 1 µg/mL Leupeptin, 1 µg/mL Pepstatin A, 1 mM Na3VO4, and 1 mM NaF. Lysates were kept on ice for 10 minutes and clarified with a high speed spin at 4°C for 15 minutes. FLAG-tagged proteins were immunoprecipitated by incubating lysates with Protein G Sepharose (GE Healthcare 17061801) and 4 µg of anti-FLAG antibody (Sigma F3165), followed by elution from the antibody by incubation with 20 µg of 3×FLAG peptide (Sigma F4799). 5 ul of purified FLAG-tagged proteins were incubated at 37°C with 5 ul of 5 mM dUTP (Promega U1191) for 0,2, 5,10,15,20, and 25 hours in 10 ul of 2× reaction buffer comprised of 100 mM Tris pH 7. 5,20 mM MgCl2,20 mM DTT, and 0. 2 mg/mL BSA. Reactions were terminated by freezing. Potentially digested dUTP are at a maximum final concentration of 1. 25 mM, lower if digested. PCR was conducted using MHV-68 BAC DNA as a template, with primers to amplify a region in ORF57 (forward primer: 5′-ACTGAAACCTCGCAGAGGTCC-3′ and reverse primer 5′-GCACGGTGCAATGTGTCACAG-3′) using the potentially digested dUTP alongside dATP, dCTP, and dGTP. Cycle conditions were 95°C 5 min; 95°C 30 seconds, 60°C 30 seconds, 72°C 30 seconds for 35 cycles; 72°C 10 minutes; hold at 4°C. PCR products were run on a 3% agarose gel. The recA+ Escherichia coli (E. coli) strain GS500 harboring a BAC containing the wild-type MHV-68 genome was used to construct recombinant MHV-68 by allelic exchange with the conjugation competent E. coli strain GS111 containing a suicide shuttle plasmid pGS284, as previously described [34], [77], [78]. For each recombinant MHV-68, overlap extension PCR was used to construct a unique shuttle plasmid pGS284 harboring the desired mutation with a 500 base pair flanking region. To screen for the correct mutation, restriction enzyme digests were performed on PCR products amplified with the outer primer sets on the BAC MHV-68 clones. WT MHV-68 and 54R contain a BamHI restriction enzyme site that is altered to a HindIII site in 54Stop. The base pair changes in 54DM allow the introduction of an AseI site that is absent in WT MHV-68. Following selection of the desired recombinant clone, the MHV-68 BAC DNA was purified and transiently transfected with Lipofectamine 2000 into 293T cells with an equal amount of a plasmid expressing Cre recombinase to remove the BAC sequence. Three days post transfection, a single viral clone was isolated by limiting dilution and propagated for future studies. Produced viruses were quantified by plaque assay and limiting dilution. Genomic integrity of the final recombinant viruses was analyzed by SmaI and EcoRI restriction enzyme digestion and Southern blot with DIG-labeled DNA probes to the whole genome or 6 kb of the left end (Roche 11585614910). Primers used for the construction of each shuttle plasmid are listed 5′ to 3′ as follows: 54R and 54Stop, outer primers: CAGGACAGATCTCACTAGACACTGTGACTATAGAC and CTGTCCCCCGGGTAGCAGACACAGGTCCTCAG; 54Stop, inner primers: CACATAACTGAGTAAGCTTACCAGACTAAAATTTCAAAACATCAC and TGGTAAGCTTACTCAGTTATGTGCCTTGTGTAGTTAGGGCC; 54DM, outer primers: GAAGAGATCTCACTCCCCCACTGAGGGACGTATGTGTCAGC and GTTGGCTAGCCAATTTCAGACTTGTCTGCAGCTTCGTGCGGAACCCTAATAAAC; 54DM, inner primers: ATTAATCAGTCCAGTTGCGCCGGTGACGAGACTGTT and GCAACTGGACTGATTAATCCCGGCTACCGGGGTGAAAT. MHV-68 concentrated virus was titered using plaque assay and limiting dilution. For plaque assay, 10-fold serial dilutions of each virus were incubated on a monolayer of Vero cells for 1 hour. The infected cells were overlaid with 5% methylcellulose DMEM. At 6 days post infection, cells were fixed with 2% crystal violet in 20% ethanol. Plaques were counted at the optimum dilution to calculate virus titer. To determine the 50% Tissue Culture Infectious Dose (TCID50), 12 wells seeded with Vero cells in a 96-well plate were infected by 100 ul of each 10-fold serial dilution. 7 to 10 days post infection, every well was scored for cytopathic effect (CPE). TCID50 was determined by the calculation TCID50/mL = 10×10∧[ (highest dilution with 100% CPE) −[0. 5+ ( (# of wells with CPE/total wells) in next highest dilution) + (each following dilution' s CPE fraction) ]]. Viral titers of multiple step growth curves were quantified by plaque assay. Wild-type MHV-68 BAC was constructed as previously described [34]. The firefly luciferase reporter driven by the interferon-stimulated response element (ISRE_firefly-luciferase) and the renilla luciferase reporter driven by the promoter of housekeeping gene phosphoglycerate kinase (PGK_renilla-luciferase) were kind gifts from Dr. Genhong Cheng and Dr. Lily Wu, respectively (UCLA). All MHV-68 open reading frames were cloned into pENTR (Invitrogen) by PCR amplification from MHV-68 DNA, then transferred to a modified destination vector resulting in a 3×FLAG epitope tag at the N-terminus. Specifically, for MHV-68 ORF54, primers used are 5′-CCGAGCGAATTCAATGAAAGTGGAATACTCCTTTGTG-3′ and 5′-GCTCGGGGTACCTTAATTCACCCCACTTGACCCAAAC-3′. To construct ORF54 MHV-68 H80A/D85N by overlap extension PCR, the same inner primers as for 54DM virus were used, the outer primers used are 5′-CCGAGCGAATTCAATGAAAGTGGAATACTCCTTTGTG-3′ and 5′- GCTCGGGGTACCATTCACCCCACTTGACCCAAAC -3′. Wild-type ORF54 and ORF54 H80A/D85N have slightly different molecular weights due to the cloning process of FLAG-ORF54 H80A/D85N. A modified Gateway cloning system (Invitrogen) was utilized to generate entry clones of the target protein, then recombined by an LR reaction into the same destination vector as FLAG-ORF54. This process inserted 6 additional amino acids to the N-terminus of the protein (from the multiple cloning site) after the 3×FLAG tag and before the ORF54 H80A/D85N ATG start site. Also, the H80A/D85N clone utilizes a STOP codon in the destination vector while the FLAG-ORF54 clone includes the endogenous STOP codon. Therefore, 10 additional amino acids (from the multiple cloning sites) are added to the C-terminus of FLAG-ORF54 H80A/D85N. In total, the FLAG-ORF54 H80A/D85N protein has an additional 16 amino acids, or approximately 1. 7 kDa additional molecular weight, compared to the wild type FLAG-ORF54. These changes demonstrated no effect on ORF54 or ORF54 H80A/D85N function. KSHV ORF54 was cloned from the a KSHV BAC construct, a kind gift from Dr. Jae Jung (University of Southern California). The primers used are 5′-TAATGGATCCATGAACAACCGCCGAGGCTC-3′ and 5′-TAATGTCGACCTAAAACCCAGACGACCCCAG-3′. 293T cells were seeded at 5×104 cells per well in a 48-well plate 16 hours prior to transfection. Cells in each well were transfected using Lipofectamine 2000 (Invitrogen), with 20 ng of ISRE_firefly-luciferase, 5 ng of PGK_renilla-luciferase, 200 ng of viral ORF or vector control, and 175 ng of filler DNA. 24 hours post transfection, identically transfected wells were left untreated or were treated with 3×104 units of human interferon-α (both human and mouse IFN-α are purchased from pbl Interferon Source). 24 hours post treatment, both firefly and renilla luciferase activity was measured (Promega Dual-Luciferase Assay Kit). Fold activation was calculated by first normalizing all values to their internal renilla luciferase control, and then by dividing luciferase activity in the treated samples by that of the untreated samples. Expression of viral ORFs was determined by immunoblot against the FLAG epitope expressed at the N-terminus of each of the ORFs in our MHV-68 expression library. NIH3T3 cells were seeded in 24-well plates and infected with FLAG-54 MHV-68 at an MOI of 1. Cells were fixed and permeabilized with treatment for 30 minutes at room temperature with 100% methanol. Cells were then washed in three changes of PBS and stored in PBS until staining. Mouse anti-FLAG M2 (Sigma F3165) was incubated with cells overnight for 16 hours at a dilution of 1∶750, and Alexa Fluor 594 goat-anti mouse IgG (Invitrogen A11005) at a dilution of 1∶1000 was incubated with cells for 1 hour. Hoechst dye was added for 10 minutes prior to analysis. Cells were lysed for 10 minutes on ice in lysis buffer (50 mM Tris pH 7. 5,1% NP-40,0. 25% sodium deoxycholate, 150 mM NaCl, 1 mM EDTA) supplemented with 1 mM PMSF, 1 mM Na3VO4, and 1 mM NaF. Lysates were then combined with 4× protein sample buffer (0. 25 M Tris pH 6. 8,8% SDS, 40% glycerin, 20% β-mercaptoethanol, 0. 008% Bromophenol blue), sonicated, and boiled for 10 minutes prior to loading on a 10% polyacrylamide gel. Membranes were stripped with Multi-Western Stripping buffer (Bioland Scientific), prior to re-probing with each subsequent antibody. The antibodies used in this study were rabbit anti-human phosphoSTAT1 (Cell Signaling 9167), rabbit anti-mouse phosphoSTAT1 (Millipore 07307), rabbit anti-total STAT1 (Cell Signaling, 9175S), mouse anti-FLAG M2 (Sigma F3165), rabbit anti-human IFNAR1 (Abcam, ab45172), mouse anti-β-actin (Sigma A5316), rabbit anti-IGF-1 Receptor β (Cell Signaling 3027S), rabbit anti-IFNAR2 (Novus Biologicals 31665), rabbit anti-phosphoErk1/2 (Cell Signaling 4376), and rabbit anti-IFIT2 (Abcam 55837). Rabbit serum against viral lytic protein ORF65 was derived in our lab. Secondary antibodies conjugated to HRP were donkey anti-rabbit IgG (GE Healthcare NA934V) and sheep anti-mouse IgG (GE Healthcare NXA931). Total RNA was extracted from cells by RNeasy Mini Kit (Qiagen) and reverse transcribed into cDNA by qScript cDNA Synthesis Kit (Quantas). Primers used in RT-PCR to quantify cellular transcripts are as follows: actin: 5′-GTATCCTGACCCTGAAGTACC-3′ and 5′-TGAAGGTCTCAAACATGATCT-3′; human IFNAR1: 5′- AACAGGAGCGATGAGTCTGTC-3′ and 5′- TGCGAAATGGTGTAAATGAGTCA-3′; murine IFNAR1: 5′- AGACGAGGCGAAGTGGTTAAA-3′ and 5′- GCTCTGACACGAAACTGTGTTTT-3′; murine MX1: 5′-GAATAATCTGTGCAGGCACTATGA-3′ and 5′-CTCTCCACTCCTCTCCTTCTTTC-3′; murine IFIT1: 5′-TGCTTTGCGAAGGCTCTGAAA-3′ and 5′-TTCTGGATTTAACCGGACAGC-3′; murine IFIT3: 5′-AGTGAGGTCAACCGGGAATCT-3′ and 5′-TCTAGGTGCTTTATGTAGGCCA-3′. All in vivo procedures were performed according to protocols approved by the University of California, Los Angeles, the Animal Research Committee, and the Institutional Review Board. Balb/C mice were purchased from Charles River Laboratories. IFNAR−/− mice were a kind gift from Dr. Genhong Cheng at UCLA. Mice were intranasally infected with 500 pfu under sedation by I. P. injection of 2 mg ketamine and 0. 04 mg xylazine. At 5 and 7 days post infection, mice were sacrificed and lung tissue was harvested in 1 mL of complete DMEM. Lung tissue was homogenized to measure the viral titer by plaque assay. At 14 days post infection, mice were sacrificed for infectious center assay of splenocytes. Briefly, a single cell suspension was isolated from the spleen of each infected animal. The splenocytes were co-cultured for one day with a monolayer of Vero cells, then overlaid with 5% methylcellulose DMEM for 6 additional days prior to fixing cells with 2% crystal violet in 20% ethanol. Each viral plaque reflects MHV-68 reactivated from splenocytes. Plaques were counted at the optimum dilution and number of infectious centers calculated per 1×107 splenocytes. To determine viral genome copies, total genomic DNA for quantitative-PCR was harvested from lung lysates and splenocytes using QiaAmp DNA Mini Kit (Qiagen). The PCR reaction was comprised of 100 ng of total genomic DNA as a template and the primers used were 5′-ACCTTGAAACCCGTGAAGG-3′ and 5′-CATCTGCCACGACCTGAGAT-3′. Swiss-Prot accession numbers for the proteins/genes used in this study are as follows: MHV-68 ORF54, P88991; KSHV ORF54, P88942; MHV-68 ORF48, P88986; and murine cellular dUTPase (DUT), Q9CQ43. The MHV-68 viral genome GenBank accession number is U97553. 2.

Human gammaherpesviruses, Kaposi' s sarcoma-associated herpesvirus and Epstein-Barr virus, are the cause of several malignancies, especially in patients immunocompromised due to HIV infection. The study of these human gammaherpesviruses is difficult due to their inability to replicate in cell culture and the lack of a small-animal model. Murine gammaherpesvirus-68 is a genetically and biologically similar virus that is utilized as a mouse model because it offers such advantages as the ability to replicate in cell culture, a manipulatable genome, and infection of mice. In this study, we have identified viral open reading frame 54 (ORF54) as an inhibitor of innate immunity, specifically of the type I interferon response. Although ORF54 is a conserved viral dUTPase, we found that its anti-interferon activity does not require its enzymatic activity. Through infection of cells and mice, we define the critical role of ORF54 in establishing persistent latent infection of MHV-68 by inducing the degradation of the type I interferon receptor. Our studies provide new insights into the far reaching effects of type I interferon signaling and the dual role of ORF54. This work could aid in the development of vaccine strategies to gammaherpesvirus infection.

Abstract Introduction Results Discussion Materials and Methods

viral latency genetic mutation cytokines viruses and cancer immune activation viral enzymes immunology microbiology host-pathogen interaction membrane receptor signaling immunologic techniques animal models of infection stat signaling family signaling in cellular processes viral immune evasion biology viral replication molecular biology immune response immune system mutagenesis signal transduction viral persistence and latency immunity virology innate immunity genetics molecular cell biology genetics and genomics immunologic receptor signaling

2011

The Anti-interferon Activity of Conserved Viral dUTPase ORF54 is Essential for an Effective MHV-68 Infection

Genome rearrangements are associated with eukaryotic evolutionary processes ranging from tumorigenesis to speciation. Rearrangements are especially common following interspecific hybridization, and some of these could be expected to have strong selective value. To test this expectation we created de novo interspecific yeast hybrids between two diverged but largely syntenic Saccharomyces species, S. cerevisiae and S. uvarum, then experimentally evolved them under continuous ammonium limitation. We discovered that a characteristic interspecific genome rearrangement arose multiple times in independently evolved populations. We uncovered nine different breakpoints, all occurring in a narrow ∼1-kb region of chromosome 14, and all producing an “interspecific fusion junction” within the MEP2 gene coding sequence, such that the 5′ portion derives from S. cerevisiae and the 3′ portion derives from S. uvarum. In most cases the rearrangements altered both chromosomes, resulting in what can be considered to be an introgression of a several-kb region of S. uvarum into an otherwise intact S. cerevisiae chromosome 14, while the homeologous S. uvarum chromosome 14 experienced an interspecific reciprocal translocation at the same breakpoint within MEP2, yielding a chimaeric chromosome; these events result in the presence in the cell of two MEP2 fusion genes having identical breakpoints. Given that MEP2 encodes for a high-affinity ammonium permease, that MEP2 fusion genes arise repeatedly under ammonium-limitation, and that three independent evolved isolates carrying MEP2 fusion genes are each more fit than their common ancestor, the novel MEP2 fusion genes are very likely adaptive under ammonium limitation. Our results suggest that, when homoploid hybrids form, the admixture of two genomes enables swift and otherwise unavailable evolutionary innovations. Furthermore, the architecture of the MEP2 rearrangement suggests a model for rapid introgression, a phenomenon seen in numerous eukaryotic phyla, that does not require repeated backcrossing to one of the parental species. Eukaryotic genome content and architecture can vary dramatically as populations of organisms evolve, or as populations of cells evolve during disease processes like cancer [1], [2]. Chromosome number may change, resulting in polyploidy and/or aneuploidy, or chromosomes may be restructured by translocations, inversions, deletions and amplifications. A striking example of genomic change, homoploid hybrid speciation, occurs when gametes of closely related species fuse to form viable hybrids. If both parental species have the same number of chromosomes, the homoploid hybrid will contain a “diploid” genome that has the same chromosome number as its ancestors; such hybrids can also be called “F1 hybrids, ” as they arise in the first filial generation following hybridization. By contrast, allopolyploid hybrid speciation typically results in a doubling (or more) of the ancestral chromosome number. Although homoploid hybrid speciation has been most commonly observed in plants [3], it has been documented in every eukaryotic Kingdom (e. g. , [4]–[7]). In the wild, as well as in brewing and wine-making, both homoploid and allopolyploid hybrid yeast have been isolated whose genomes are wholly or partly derived from two or more different members of the Saccharomyces “sensu stricto” group [8]–[10]. These Saccharomyces species can also be mated in the lab to create de novo interspecific hybrids [11]–[13]. In addition to homoploid hybrids that bear one copy of each of their parental species' chromosomes, “introgressive hybridization, ” also known as introgression, has been observed among the sensu stricto group of Saccharomyces. This term was first coined by Anderson and Hubricht in 1938 [14] to denote the infiltration of the “germplasm” of one species into that of another following hybridization and repeated backcrossing. If this region is not selected against, in time it can become established as an “island” of the minor species' genome encompassed within the major species' genome. Introgressive hybridization is thought to be a long-term process, requiring an initial interspecific hybridization event, followed by the repeated backcrossing with only one of its parent species [15]. Since its first description, introgressive hybridization has been identified in numerous eukaryotic phyla (see [16], [17] for reviews). Introgression events have been documented among many of the Saccharomyces sensu stricto species, in yeasts isolated from natural environments [11], [18], clinical and animal sources [19], [20], and from wine, beer, and other industrial environments ([8], [21]–[28]; also see [29] for review). Like homoploid and polyploid hybridization, introgression is also considered to be important as a mechanism leading to speciation [30], [31]. Indeed, hybridization and introgression have been suggested as sources of unexpected, extreme ‘transgressive’ phenotypic traits upon which natural selection can act [32], facilitating rapid within-lineage evolution [30]. Although the evolutionary implications of interspecific hybridization in general, and introgressive hybridization in particular, have been appreciated for some time, their molecular bases and relative importance as evolutionary mechanisms among various Kingdoms are incompletely understood [33]. Moreover, while genomic technologies have greatly expanded our understanding of genome content and stability during adaptive evolution, we have limited knowledge of how genomes stabilize following the initial ‘shock’ of interspecific hybridization [34]. Significantly, the actual process of introgressive hybridization has never been captured in action. Budding yeasts of the genus Saccharomyces provide an ideal eukaryotic system in which to close these knowledge gaps. Not only do Saccharomyces yeasts readily form interspecific hybrids, they also have a relatively simple life cycle, reproduce quickly, tolerate aneuploidy [35] and can be propagated as stable haploids or diploids. Environmental variables and the size and structure of yeast populations can also be controlled experimentally, and because yeasts can be preserved cryogenically, it is possible to compare evolved to ancestral strains or to replicate any stage of an experiment [36]. Saccharomyces cerevisiae and S. uvarum (previously called S. bayanus) are distantly related members of the Saccharomyces sensu stricto group, having diverged ∼20 million years ago [37]. Despite having only 80% sequence identity in coding regions and 62% in intergenic regions, the S. cerevisiae and S. uvarum genomes are largely syntenic, with the exception of 3 large reciprocal translocations and within some regions of their telomeres, where rapid structural evolution has occurred [37], [38]. Because of their synteny and their sequence divergence, which allows their genomes to be distinguished, we experimentally investigated the evolution of an F1 homoploid interspecific hybrid formed between S. cerevisiae and S. uvarum. We evolved three independent replicate populations of this hybrid under continuous nitrogen limitation, a selective pressure often encountered in wine-making as well as in other ecological settings where S. uvarum and S. cerevisiae likely occur [39]–[41]. We determined each parental species' contribution to the evolving genomes by array Comparative Genomic Hybridization (aCGH) as well as by whole genome sequencing of select ancestral and evolved hybrid clones. We discovered a recurrent genomic rearrangement in all three independently evolved hybrid populations. This rearrangement ultimately produces two copies of an interspecific MEP2 fusion gene, which in both S. cerevisiae and S. uvarum encodes for a high-affinity ammonium permease. In all cases the fusion gene is structured such that the 5′ end of the gene is derived from S. cerevisiae sequences and the 3′ end is derived from S. uvarum sequences, an evolutionary innovation that could only arise in a hybrid genome. Repeated evolution of this novel fusion gene in independent populations suggests that it is adaptive under nitrogen-poor environments where ammonium is the sole nitrogen source. The architecture of the rearrangement suggests a model for rapid introgression without the need for repeated backcrossing to one of the parental species. At steady state, the interspecific hybrid performed better than either of its ancestral species under aerobic ammonium limitation at 25°C. At the beginning of steady state growth (t-initial), residual ammonium was near or below detection limit (<0. 01 ppm or 0. 01 µg L−1) for both parental species and for the interspecific hybrid. Residual glucose at t-initial in diploid S. cerevisiae, diploid S. uvarum and in the interspecific hybrid was 5. 0±0. 34,5. 7±0. 23 and 2. 3±1. 20, respectively (mean ± Std. Error, g L−1; P = 0. 04, one-way ANOVA followed by Student-Newman-Kuels (SNK) test), while optical density at A600 was 0. 90±0. 04,1. 11±0. 02 and 1. 55±0. 17, (mean ± Std. Error, P<0. 01 one-way ANOVA followed by SNK test); thus for both parameters the hybrid showed superior growth performance compared to either of its parents. In three independent hybrid populations evolved for 200 generations (t-final) we detected no significant change in either residual glucose or optical density relative to the ancestral unevolved hybrid. The extent to which uptake of the limiting nutrient was enhanced could not be assessed, as ammonium concentration in the experimental populations was close to our assay detection limit at t-initial and below this limit at t-final. Based on these observations, we concluded that in a nitrogen-limited, glucose-sufficient environment, S. cerevisiae×S. uvarum interspecific hybrids had limited scope for measurable improvement in the physiological parameters we measured. To directly test whether individual clones from the evolved populations were more fit than their common ancestor, we performed short-term (15 generation) competitive chemostat experiments. We competed the founder S. cerevisiae×S. uvarum hybrid (GSY86) and each of three individual 200-generation evolved clones (GSY2532, GSY2533 and GSY2535, representing one isolate from each vessel; Table 1) against a fluorescently marked unevolved S. cerevisiae×S. uvarum interspecific hybrid strain (GSY2590; Table 1), under the same ammonium-limited conditions used for our long-term evolution experiments. GSY2590 is identical to the ancestral founder strain except for the presence of an integrated Green Fluorescent Protein (GFP) gene, as described in Materials and Methods. Because we observed only modest fitness differences between the founder hybrid (GSY86) and GSY2590 (competition coefficient = 1. 04±0. 009), we concluded that the latter could serve as a surrogate “founder” in the competitive chemostat experiments, with a 0. 04 correction to the competition coefficient. We found that each of the three evolved F1 hybrids consistently outcompeted GSY2590 under continuous ammonium limitation: corrected selection coefficients for GSY2532, GSY2533 and GSY2535 were 1. 15 (±0. 005), 1. 14 (±0. 003), and 1. 11 (±0. 008), respectively (mean ± Std. Error). These fitness gains are statistically significant (P<0. 001) and similar in magnitude to values reported for S. cerevisiae evolving under ammonium limitation (1. 09 in [43]), but less than fitness gains reported for S. cerevisiae evolved under aerobic glucose limitation (1. 16 to 1. 60 in [44]). The scope for fitness improvement in yeast evolving at low growth rates is likely greater under aerobic glucose limitation because cells can switch from respiro-fermentative to respiratory metabolism, which greatly increases the efficiency of converting substrate to biomass [45]. Furthermore, as fungi in nature face chronic nitrogen limitation [39], [40], natural selection has likely fine-tuned mechanisms to scavenge inorganic nitrogen. For time points corresponding to generations ∼50, ∼100, ∼150, and ∼200, archived population samples from vessels A, B and C were revived from cryogenic storage and plated on YPD; for each time point two clones were selected at random for karyotype analysis using CHEF (Clamped Homogeneous Electric Fields) gel-electrophoresis; one clone from the founding (t-initial) population was also included (Figure 1). Although most isolates exhibited the parental karyotype, several variants exhibited size changes in one or two chromosomes. For example, both isolates from generation 200 of vessel A exhibited an increase in size of one of the chromosomes corresponding to the S. cerevisiae chromosome 7+15 doublet at 1200 Kb (Figure 1, yellow arrow). Interestingly, in both vessels B and C (inoculated independently with different starter cultures), clones isolated at generation 100 and generation 200, respectively, demonstrated absence of the ∼650 Kb band, apparently corresponding to chromosome 11 of S. uvarum (Figure 1, red ovals). Other karyotypes transiently appeared in the populations, such as that observed in vessel C at 100 generations involving a size increase in S. uvarum chromosome 2–4 at 1500 Kb (Figure 1, blue arrow), as well as multiple instances of size variation in S. cerevisiae and S. uvarum chromosome 12 (1640/1900 Kb; topmost chromosomal band seen in Figure 1). This last observation may reflect variation in copy number of tandemly-arrayed ribosomal DNA repeats on chromosome 12, as this region of the yeast genome is known to be labile [46]. CHEF analyses clearly demonstrated genome malleability in interspecific hybrids evolving under continuous nitrogen limitation. However, while CHEF analysis reveals the phenomenon of malleability, and serves as a screen to identify interesting karyotypes, it tell us little about the underlying architectural changes, and nothing at all about the molecular mechanisms that might be at play. To further investigate the evolved clones, we used aCGH to assay whole genome copy number variation arising from non-copy-neutral changes such as deletions, amplifications and non-reciprocal translocations. aCGH profiles of evolved hybrids revealed that a small number of chromosomes had undergone rearrangement; however, the rearrangements detected using aCGH were not those detected by CHEF, indicating that the rearrangements detected by CHEF were most likely copy-neutral events. aCGH analysis showed that clones isolated from each of two independent t-initial founding populations had the expected, non-rearranged F1 hybrid genome configuration, i. e. , they contained one complete non-rearranged chromosomal set from each of the input genomes, within the limits of detection of aCGH (Figure S2). Strikingly, however, aCGH revealed that 9 out of the 10 evolved hybrid clones we examined—four of four 150-generation clones (one from Vessel A, two from Vessel B, and one from Vessel C), and five of six 200-generation clones (2 each from Vessels A, B, and C) —contained a distinctive and apparently identical, or extremely similar, rearrangement on chromosome 14, whereby fully half of the S. uvarum chromosome 14 (much of the distal portion of the left arm) was replaced with the corresponding region of the S. cerevisiae chromosome 14 (Figure S2). This appeared in all cases to be a “non-reciprocal” translocation event, resulting in increased copy number of the distal left portion of the S. cerevisiae chromosome 14, with the concomitant deletion of the corresponding S. uvarum chromosome 14 region (detailed aCGH results for the chromosome 14 region are shown for three 200-generation clones, one from each vessel, in Figure 2). Because the lengths of the translocated regions of the two chromosome 14 s are roughly equivalent between these species, we would not expect to see in these clones any change in chromosome 14 mobility by CHEF, and in fact, none was seen (see Figure 1). Because this rearrangement is seen in clones from all three vessels, it must have arisen independently. The fact that the chromosome 14 rearrangement occurred independently and is seen in a large majority of evolved clones examined suggests that it is adaptive under inorganic nitrogen limitation; indeed, as shown below, the rearrangement always occurs precisely within the MEP2 gene (YNL142W), which encodes the high-affinity, low capacity ammonium permease in Saccharomyces [47], [48]. The MEP2 gene is found in both the S. uvarum and S. cerevisiae genomes, in the same (syntenic) position on each genome' s chromosome 14, sharing 85% DNA sequence identity. In addition to the MEP2 rearrangement, a few additional rearrangements resulting in copy number variation—including deletions of ∼15 to ∼50 kb occurring on chromosomes 5,12, and 15 of S. cerevisiae and chromosome 9 of S. uvarum, plus a probable extra copy of S. cerevisiae chromosome 12 in Vessel B clones—were evident among some of the evolved clones, but none of these were shared across vessels (Figure S2). We designed primers well outside the chromosome 14 fusion junctions detected by aCGH in the evolved clones (Table S1) to PCR-amplify the junction-containing regions of the three 200-generation evolved clones whose aCGH results are shown in Figure 2; these are clones GSY2532, GSY2533, and GSY2535, coming from Vessels A, B, and C, respectively (Table 1). Sanger sequencing of these PCR products revealed that the junction breakpoints of the rearrangement differed among clones (Figure S3A), indicating that despite appearing almost identical by aCGH, the rearrangements were indeed independent, as expected since the clones arose in three separate vessels. The junction sites for these three clones were all located within the coding sequence of the MEP2 gene and in all three cases the gene remained in-frame. For GSY2532 and GSY2535 the junctions result in a predicted fusion protein with the N-terminal one-third (approximately) of the protein coming from S. cerevisiae and the C-terminal two-thirds from S. uvarum; for GSY2533, these proportions are swapped (Figure S3B). The S. cerevisiae and S. uvarum Mep2 proteins are each 499 amino acids long, with 17 amino acid differences between them; each of the three predicted Mep2 fusion proteins has a novel predicted protein sequence derived from the combination of the S. cerevisiae and S. uvarum MEP2 genes. To further elucidate genomic changes that occurred during evolution of the interspecific F1 hybrids, we performed Illumina whole genome sequencing on the three independent 200-generation evolved clones containing MEP2 fusion genes, whose junctions we had sequenced as described above (GSY2532,2533,2535), and also on the ancestral clone used to found the three replicate vessels (GSY86). Read depths across the MEP2 regions indicated that as expected for the ancestral GSY86 clone, there were no rearrangements resulting in copy number changes in either genome' s MEP2 region (Figure 3A). In contrast, and confirming our aCGH results, we detected large-scale copy number changes in the MEP2 region for each of the three evolved clones (Figure 3A). Surprisingly, however, we observed from our whole genome sequencing that the architecture of the genome rearrangement was more complex than we had predicted by aCGH. Instead of a simple translocation (and/or “breakage-induced replication” event) to yield one S. cerevisiae – S. uvarum chimaeric chromosome (with the junction located within the MEP2 gene) and one intact S. cerevisiae chromosome, in all three cases the expected S. cerevisiae – S. uvarum chimaeric chromosome was present, but there was an additional rearrangement on the S. cerevisiae chromosome (Figure 3A, B). This additional event resulted in a complete deletion of 5 to 15 kb of the S. cerevisiae chromosome, the region instead being precisely replaced with the corresponding S. uvarum chromosomal region within an otherwise intact S. cerevisiae chromosome; this is an event that can be considered to be the equivalent of an “introgression” of the S. uvarum genome into the S. cerevisiae chromosome (Figure 3B). In each case the distal junction on the S. cerevisiae chromosome occurred within the MEP2 gene, with exactly the same junction as that found in the partner S. cerevisiae – S. uvarum chimaeric chromosome (note: this is why our Sanger sequencing of PCR products described above gave readable sequences). In all cases the junction found by whole genome sequencing matched exactly the junction we had found by Sanger sequencing. The proximal junction was always well “downstream” from the MEP2 gene and varied for each clone, occurring anywhere from ∼5 Kb (GSY2532 and GSY2533; within THO2) to 15 kb (GSY2535; near FPR2) toward the centromere (Figure 3A). The most interesting outcome of this additional rearrangement within the S. cerevisiae chromosome is that each of the evolved clones contains two copies of identical MEP2 fusion genes (with junctions as shown in Figure S3A and S3B), and no copies of either the S. cerevisiae or the S. uvarum endogenous (“wild-type”) MEP2 genes. Analysis of the whole genome sequences for shared SNPs (and/or shared SNP-containing genes) revealed that no such shared mutations existed among the evolved clones (Table S2). Because we saw MEP2 rearrangement events occurring on both the S. uvarum and S. cerevisiae chromosomes of the evolved clones, we wished to know if the rearrangements occurred in a single concerted step, or whether a sequential multi-step process led to the final configuration. We therefore performed diagnostic PCRs on 12 single colony isolates from evolved populations corresponding to 0, ∼50, ∼100, ∼150 and ∼200 generations, from Vessels A and B, for a total of 120 isolates (12 per time point, 60 per vessel). We used 4 PCR primer combinations for each clone, using primer combinations (Table S1) specific for the S. cerevisiae MEP2 gene, the S. uvarum MEP2 gene, the S. cerevisiae-S. uvarum fusion MEP2 gene (found in evolved clones), or the S. uvarum-S. cerevisiae “reverse-fusion” MEP2 gene (not found in the evolved clones described above that were examined by aCGH and/or sequencing). Almost all clones from generations 0 and 50, as expected for an un-rearranged (“ancestral”) hybrid, showed the coexistence of the S. cerevisiae MEP2 gene and the S. uvarum MEP2 gene, with no evidence of a MEP2 fusion gene (Figure S4A). We further found that the MEP2 S. cerevisiae-S. uvarum fusion gene appeared in both vessels starting at 100 generations and persisted through to the 200-generation time point (Figure S4A). At 100 generations, in both vessels, less than 20% of the clones contained the ancestral un-rearranged MEP2 genes; instead most clones contained the MEP2 fusion gene either alone (presumably in two copies as seen in GSY2532,2533, and 2535), or the fusion gene in conjunction with the S. uvarum-only MEP2 gene. By the 150 and 200-generation time points, the MEP2 fusion gene alone was predominant. In these later time points, there also appeared clones containing only the S. cerevisiae MEP2 gene or only the S. uvarum MEP2 gene, without the presence of the MEP2 fusion gene (Figure S4A). Interestingly, although we observed the MEP2 fusion gene in conjunction with the S. uvarum MEP2 gene, we never observed the S. cerevisiae MEP2 gene occurring with the MEP2 fusion gene. Finally, the “reverse-fusion” MEP2 gene was not found in any of the 120 clones. We Sanger-sequenced the PCR products corresponding to the MEP2 fusion gene from all clones yielding such PCR products. We found that there were several additional MEP2 fusion junctions present in the evolved clones of both vessels, with junctions differing from those found in the three clones (GSY2532,2533, and 2535) we had previously characterized by Sanger and whole genome sequencing. As seen in Figure 3C and Figure S4B, in addition to the junctions found for GSY2532 (Vessel A) and GSY2533 (Vessel B), four additional distinct and separate novel MEP2 fusion junctions were found in Vessel A clones, and a further three distinct and separate novel MEP2 fusion junctions were found in Vessel B, for an observed total of nine different MEP2 gene fusion junctions (including that of GSY2535); in all cases, the junctions occurred within the MEP2 coding sequence and were in-frame. Because the evolved fusion genes have the S. cerevisiae MEP2 promoter, we hypothesized that the MEP2 gene fusion events may have been selected because that promoter might result in higher transcript levels. We thus performed qRT-PCR reactions for each genome' s version of the MEP2 gene on the founding ancestor GSY86, assaying (in triplicate) two independent biological replicates of GSY86 that had been grown to steady state in the same nitrogen-limited media and chemostats used for the original evolutions. We determined that the S. cerevisiae genome' s copy of the MEP2 gene is indeed expressed at a somewhat higher level than the S. uvarum copy, by almost 2-fold, supporting our hypothesis (Figure S5; raw and normalized data given in Table S4). However, when we determined the expression of the fusion gene in an evolved clone (GSY2532), it appeared to produce less transcript per locus than either the S. cerevisiae or S. uvarum genes did in the founding hybrid (Figure S5, Table S4; note, in the evolved clone, the transcript quantified by qPCR is produced from 2 fusion loci, so the amount per locus is less). The mechanistic basis for MEP2 fusion genes' adaptive advantage is therefore more complex than increased expression, and may relate instead to changes in protein structure that increase the novel permeases' catalytic efficiency, decrease their Km for ammonium, and/or alter their activity as nutrient signaling molecules. Laboratory strains of S. cerevisiae are the best-studied group of fungi in terms of their genome structure. However, even within this relatively homogeneous group, strains differ widely in rates of mitotic chromosome loss and levels of chromosome-length polymorphism [49]–[51]. Furthermore, mitotic genome instability in S. cerevisiae has been shown to be evolutionarily significant in the laboratory [52]–[54], in wine fermentation [24], [55] and in biomass conversion to fuel ethanol [56]. A large amount of standing genomic variation (e. g. , ploidy differences, transposon copy number, and chromosome length polymorphism) is found among Saccharomyces isolates collected from natural and industrial settings (e. g. [57]–[59]), reinforcing the view that genomic plasticity may be evolutionarily important in diverse settings (see [10], [24] for reviews). Of special relevance to our study is the discovery that this variation very often takes the form of mosaic genomes that result from natural interspecific hybridization events [20], [28], [60]–[63]. Mosaic genomes arising from interspecific hybridization have been discovered in other yeasts. For example, Pichia sorbitophila appears to have arisen in recent centuries via allopolyploidization between two species affiliated with the genus Millerozyma [64]. Resolution of the initial hybridization event has produced 7 chromosome pairs that are either completely homozygous, completely heterozygous or mosaics. In mosaic chromosomes, breakpoints between homozygous and heterozygous regions can occur in protein coding genes [64], though with unknown phenotypic consequences. While the foregoing example provides an interesting snapshot of a recent hybridization event, no published study to date has explored genome dynamic changes that occur as experimentally-created interspecific hybrids evolve. Using CHEF and aCGH analysis we were able to detect chromosomal loss and/or size changes, large indels, and non-reciprocal translocations in evolving interspecific F1 hybrids. Overall, however, the frequency with which we observed genomic rearrangements in hybrids evolving under nitrogen limitation was considerably less than that reported for S. cerevisiae evolving under glucose limitation [53]. Further, very few large-scale genomic changes were observed by CHEF and aCGH analysis when the diploid parental species themselves, S. cerevisiae and S. uvarum, were evolved under nitrogen limitation (Dunn, Piotrowski et al. in prep.). However, for a large number of evolved interspecific hybrid clones we observed a distinctive recurrent rearrangement, involving both parental genomes at the locus encoding high affinity ammonium permease, which we first observed by aCGH and then confirmed by Sanger and whole genome sequencing. These recurrent MEP2 rearrangements in S. cerevisiae×S. uvarum hybrids provide an interesting contrast with the results of experimentally evolving haploid S. cerevisiae under different types of limiting nitrogen. There, recurrent rearrangements were observed at the GAP1 locus, which encodes for the general amino acid permease [43]. A single homologous recombination event was seen to produce two different alleles: GAP1extrachromosomal circle or gap1Δ; the former being associated with higher fitness in clones adapted to L-glutamine and L-glutamate, the latter with higher fitness in clones adapted to urea, allantoin, and ammonium. Owing to differences in genome content, hybrid interspecific diploids are able to explore adaptive possibilities not open to haploid S. cerevisiae. While we failed to detect any mutations or rearrangements that inactivate GAP1 in any of the ammonium-adapted clones that we sequenced, our observations are remarkably similar to what occurs when either haploid or diploid S. cerevisiae is evolved under sulfur limitation, where a recurrent rearrangement, resulting in gene amplification, has been observed at the SUL1 locus which encodes a high affinity sulfate permease of the SulP anion transporter family [65]. In our experiments, the recurrent event is a complex rearrangement at the MEP2 locus, which encodes a high affinity ammonium permease. This rearrangement yields a genome containing two copies of a fusion MEP2 gene with the 5′ portion derived from S. cerevisiae and the 3′ portion from S. uvarum. It is likely that this rearrangement is adaptive under N-limitation, due to both its high allele frequency and the fact that it was selected for independently multiple times. As expected, when tested in direct head-to-head competition experiments, each of three independently-evolved evolved clones having the characteristic MEP2 rearrangement showed significant fitness increases relative to an unevolved ancestral clone. We have not yet shown that the presence of only the two MEP2 fusion genes is necessary and sufficient to confer this selective advantage within the context of an otherwise unevolved hybrid. Nevertheless, whole genome sequencing of these three evolved clones revealed no shared SNPs or rearrangements in other genes, suggesting that the recurrent MEP2 rearrangement is a key shared adaptive innovation in these evolutions, an innovation unavailable to either parent alone. A number of recent studies suggest that gene fusions may contribute to the evolution of novel functions (reviewed in [66]). Because new folding structures could quickly produce traits unattainable by point mutation alone, fusion genes could be potent drivers of adaptive change [67], and indeed, in vitro generation of fusion genes has been directly shown to create novel enzymatic functions [68]. Fusion genes have been discovered in many organisms, and even play an important role in the initial steps of tumorigenesis [69]. Examining hybrid lager yeast, Usher and Bond recently described a fusion gene formed by recombination between homoeologous chromosomes of S. cerevisiae and S. eubayanus [21]. The result, a chimaeric gene for GPH1, which encodes for glycogen phosphorylase, fails to produce mature mRNA because of a frameshift in its coding sequence; loss-of-function at GPH1 leads to a glycogen phenotype typical of haploid cells. In contrast, the chimaeric genes that we discovered in the course of evolving yeast hybrids are always formed by in-frame fusions of the 5′ end of the S. cerevisiae MEP2 coding sequence to the 3′ end of the S. uvarum MEP2 coding sequence. Chemostat theory [70], [71] predicts that when cells evolve under nutrient limitation, adaptive genotypes arise as a result of either increased efficiency of nutrient use or increased capacity for assimilating the limiting nutrient. Previous studies have shown that in S. cerevisiae evolving under glucose limitation both mechanisms come into play [45], [53], [72], resulting in increased yield biomass and diminished concentrations of residual substrate at steady state. To test whether this was also the case in our experiments, we grew to steady state under ammonium limitation single-colony isolates of the ancestral hybrid and the three evolved clones that we sequenced. We found that in all cases residual ammonium was near or below detection limits (0. 01 ppm), which follows from the very low Km of ammonium permease (1–2 µmolar; [73]) and the high velocity of ammonium uptake (<0. 25 µmolar per second at D = 0. 1 h−1), under glucose-sufficient conditions. We found that culture density and residual glucose concentrations for ancestral and evolved strains were also not statistically different (though interestingly, GSY2532 and GSY2535 produced a small but significantly greater amount of dry weight biomass than their ancestor, see Table S3, P = 0. 002). These biochemical results, are, as for the genomic results, again reminiscent of those obtained when S. cerevisiae is experimentally evolved under inorganic sulfur limitation [65]. Evolved sulfur-limited populations show very modest increases in cell biomass, compared to evolved glucose and phosphate-limited populations, even though SUL1, which encodes a high-affinity sulfate transporter, is amplified in multiple independent evolutions, and even though this mutation demonstrably increases fitness when crossed into an unevolved wild-type background. Gresham et al. [65] conclude that the scope for metabolic innovation in inorganic sulfur metabolism is constrained, in this case by the small contribution of sulfur to cell biomass, relative to that of glucose or phosphate, and that this constraint results in the repeated evolution of a rearrangement resulting in SUL1 amplification. Our nitrogen-limitation results suggest metabolic constraints of a similar nature, which may be driven in part by the fact that fungi as a group are chronically nitrogen limited and have likely been under strong selection to acquire the capacity to scavenge this element to extremely low levels. Our qRT-PCR results from the unevolved hybrid show a relatively modest two-fold difference in expression of the MEP2 gene from each of the genomes present in the unevolved hybrid, with S. cerevisiae the higher expressed gene of the two. This would seem to indicate that the unidirectional nature of the fusion gene rearrangement, whereby we always observe the S. cerevisiae promoter and 5′ end of the gene fused to the 3′ end of the S. uvarum gene, arises simply to increase overall transcript levels of the MEP2 gene. However, in an evolved clone, we surprisingly discovered that the fusion gene produces slightly lower transcript levels per locus than does either the wild-type S. cerevisiae or S. uvarum locus in the progenitor hybrid. It is unclear exactly what this finding means – possibly that transcription from the MEP2 gene is governed by a feedback mechanism that reduces its transcription. It may be that the actual fusion proteins themselves, despite the few amino acid differences they show relative to the two parental genes (Figure S3B), provide an adaptive advantage. Possibly the novel chimaeric ammonium permeases differ from their ancestors in having a lower Km, which would lead to lower residual ammonium levels, and/or a higher kcat, which would result in greater overall uptake velocity. Alternatively, because in S. cerevisiae the Mep2 protein forms multimeric complexes in the plasma membrane [74], it may be of adaptive benefit to hybrids to produce Mep2 proteins that contain only S. cerevisiae N-termini and/or only S. uvarum C-termini; this could possibly result in better oligomerization for improved transport function and/or prevent dominant negative interactions between the two species' proteins. Indeed, dominant negative interactions have previously been noted between different alleles of the closely-related Mep1 and Mep3 proteins in yeast [75]. In this regard it is also provocative that we did not observe the coexistence of the MEP2 fusion gene with the S. cerevisiae MEP2 gene in any of the 120 clones that we genotyped, although we did see coexistence of the MEP2 fusion gene with the S. uvarum MEP2 gene; this may be evidence for dominant negative interactions between the MEP2 fusion gene and the S. cerevisiae MEP2 gene. Based on our genotyping results shown in Figure S4A, we believe that a two-step recombination event such as that shown in Figure 4 occurred between S. cerevisiae and S. uvarum chromosomes to generate the rearrangements seen in the evolved clones. We presume the event began with a double-strand break (DSB) in or near the S. cerevisiae MEP2 gene, followed by some amount of resection of the sequences surrounding the break (as described in [76], [77]). Strand invasion into a homologous region of the S. uvarum chromosome would have then been followed by repair of the resected sequences using the S. uvarum chromosome as a template, creating a gene conversion event with resultant loss of heterozygosity (LOH) (top portion of Figure 4). At the end of the first event, the S. uvarum chromosome would have been intact, while the other resultant chromosome would still be almost completely composed of S. cerevisiae sequences, aside from several Kb of S. uvarum genome precisely substituted at the MEP2 region (the exact Kb of S. uvarum sequences would depend on the amount of resection on either side of the DSB). Subsequently, either a DSB in the S. uvarum chromosome within the shared MEP2 gene region, followed by break-induced replication (BIR), or alternatively, a mitotic crossover event in G2 (left and right lower portions, respectively, of Figure 4) would have led to the final evolved genome configuration of two fusion MEP2 genes, sharing an identical fusion junction. Such gene conversion and BIR mechanisms have been previously well-documented and described in detail for yeast and many other organisms (see [78], [79] for reviews). We believe that a two-step process brought about the final evolved clone configuration, because some of the isolates from the ∼100 generation time-points in two independent populations (Figure 3A) showed the coexistence of the fusion MEP2 gene with the S. uvarum MEP2 gene (as for the intermediate shown in the first step of Figure 4). We further believe that genetic information was always transferred unidirectionally by a gene conversion event from S. uvarum to S. cerevisiae, because we never observed coexistence of the fusion MEP2 gene with the intact S. cerevisiae MEP2 gene (as depicted in Figure S6A and S6B), and we never detected the S. uvarum - S. cerevisiae “reverse” fusion (as depicted in Figure S6B). These findings suggest that the alternative models shown in Figure S6A and S6B are unlikely. Interestingly, this same type of event, leading to a virtually identical rearrangement configuration, has been seen in chromosomes of mouse cell lines lacking the Bloom syndrome helicase [80]; other similar, but not identical, patterns of rearrangement have been seen in yeast using plasmid-based “chromosome fragmentation vectors” and are thought to arise from template switching [81]. Introgression—infiltration of the “germplasm” of one species into that of another—occurs widely [16] and may produce extreme transgressive traits [32], which can drive rapid evolution and even speciation [30], [31]. Horizontal gene transfer, long known to be an engine of biodiversity in prokaryotes, has also been observed between eukaryotes [82], and in yeasts has recently been shown to be a mechanism by which “germplasm infiltration” can rapidly occur. Galeote et al. reported variable integration of a 17 kb ARS-containing Zygosaccharomyces bailii genome segment into dozens of S. cerevisiae wine strains [83]; this Z. bailii insertion was first discovered by whole genome sequencing of a wine strain [84]. The organization of Z. bailii insertions and the conspicuous absence of sequence similarity at breakpoints suggest they replicate via an extrachromosomal circular intermediate and insert via nonhomologous recombination. By contrast, introgression that arises via interspecific hybridization is currently thought to occur slowly, requiring repeated backcrossing with one of the parental species [15]. However, the structure of the MEP2 rearrangements we have discovered suggests a mechanism by which introgressive hybridization can occur rapidly. In each case, one of the rearranged chromosomes consists almost exclusively of S. cerevisiae sequences except for a precise replacement of several Kb with S. uvarum sequences. Diploid MEP2 fusion hybrids that undergo meiosis may produce a small number of spores that contain a haploid complement of only S. cerevisiae chromosomes including the one S. cerevisiae chromosome with the S. uvarum MEP2 region. Alternatively, loss of the S. uvarum chromosomes from the MEP2 fusion hybrid, similar to what has been described before for interspecific hybrids undergoing selection [42], could result in an “S. cerevisiae” strain containing just the introgressed S. uvarum MEP2 region. These scenarios do not require repeated backcrosses to one of the parents, and open up the possibility for rapidly evolving adaptive innovations forbidden to either parental species. The S. cerevisiae parental strain is a derivative of laboratory strain S288C (strain “CC230”; MATα; ura3-52; ho), while the S. uvarum parental strain is derived from strain CBS7001 (strain “CC6”; MATa/MATα; lys2-5/lys2-5; HO/HO; see Table 1 for complete list of strains used in this study). Their F1 interspecific hybrid, GSY86, was obtained by mass-mating CC230 with mass-sporulated CC6 and selecting for prototrophy. Individual evolved clones that were further studied are also shown in Table 1. The “Delft” nitrogen-limiting medium used for batch and chemostat cultures was based on that described by Boer et al. [85] as follows: the basal nitrogen-limiting medium (“basal salts”) consisted of the following components per liter: 0. 15 g (NH4) 2SO4,5. 3 g K2SO4,3. 0 g KH2PO4, and 0. 5 g MgSO4. 7H2O, to which was added 1× vitamins and 1× trace metals (both as in [86]), as well as 0. 02 g uracil, 0. 03 g lysine, 0. 06 g leucine and 9 g glucose. Strains were grown at 25°C in 500 mL fermenters (INFORS AG) with a working volume of 300 mL. Impeller speed was set to 300 rpm; airflow to 10 L h−1; the target dilution rate was 0. 16 h−1. The founder F1 hybrid was first grown overnight in 2 mL of YPD-1% glucose, whereupon 500 µL of this culture was transferred into 25 mL Delft nitrogen-limiting media and grown overnight at 25°C. Three mL of this culture were sterilely transferred into 27 mL of sterile glucose, vitamins, metals, uracil, leucine, and lysine at the prescribed concentrations, and the suspension added by positive pressure to an INFORS vessel containing 270 mL of autoclaved basal salts. Three separate fermenters (Vessels A, B, and C) were inoculated in this manner. Populations were sampled every 48 h (∼10 generations). Two and one-half mL of cell suspension were withdrawn from the chemostat vessels and apportioned as follows: (i) 500 µL were added to 500 µL sterile 30% glycerol, then archived in duplicate at −80°C; (ii) 1 mL of cells were sterile-filtered through a 0. 45 µm in-line filter and retained for assay of extracellular metabolites, (iii) 100 µL were diluted 9∶1 in glass-distilled water, and optical density measured at 600 nm using a Spectronic Biomate 3 spectrophotometer. Every 50 generations, archived populations were streaked onto YPD agar, and a random subset karyotyped by CHEF gel electrophoresis, as described below. A subset of these clones was further analyzed for changes in genome architecture by aCGH and by Illumina whole genome sequencing, as described below. Residual glucose was assayed spectrophotometrically on cell-free filtrate using R-Biopharm assay Kit #716251 (R-Biopharm, Darmstadt, Germany). Residual ammonium was determined by a modified version of the Berthelot reaction [87], scaled down for 96-well format. Biomass was estimated by filtering 50 mL of chemostat culture onto tared 0. 45 µm nylon filters, and drying filters in a desiccator at 37°C for 48 hrs. To test for statistically significant differences in growth and residual metabolites between experimental populations from the first steady-state (t-initial, 10 generations) and the last (t-final at ∼200 generations) time points, and to test physiological data obtained by growing single clones to steady state, we used ANOVA followed by a Student-Newman-Kuels multiple comparison test. All statistics were calculated with Sigma Plot 11 (Systat Software, San Jose, CA). CHEF analysis was performed on two randomly-chosen single colonies isolated on YPD agar from frozen glycerol stocks of ∼50, ∼100, ∼150, and ∼200 generation population samples, from each of the three independent experimental populations. Analyses were performed essentially as described [88], [89]. A complete description of the 2-species (S. cerevisiae and S. uvarum) array design is given in [62]; arrays were manufactured by Agilent and contain ∼5500 oligonucleotide species-specific probes per species, approximately evenly spaced across each genome. Genomic DNA from ancestral and evolved clones was prepared using Zymo Research YeaStar columns according to the manufacturer' s recommendations, and then digested with HaeIII. We then labeled 350 ng of this DNA with Cy5 (red). We similarly labeled, but instead with Cy3 (green), the same amount of reference DNA, which consisted of an equimolar mix of sheared genomic DNA from S. uvarum (CBS7001) and from S. cerevisiae (S288c). The labeled experimental and reference DNAs were then mixed together and hybridized to the 2-species microarrays as described [62]. Microarray data have been deposited in the Gene Expression Omnibus (GEO) repository [http: //www. ncbi. nlm. nih. gov/geo/] under accession GSE18060. The Caryoscope program [90] was used to view microarray data in a genomic context. Single-colony isolates were obtained by plating onto YPD agar small aliquots of the populations corresponding to 0, ∼50, ∼100, ∼150 and ∼200 generations, from Vessels A and B. Twelve isolates per time point, per vessel (60 isolates total per vessel) were subjected to colony lysis and subsequent PCR using primer combinations (Table S1) specific for the: (1) S. cerevisiae MEP2 gene, (2) S. uvarum MEP2 gene, (3) S. cerevisiae-S. uvarum fusion MEP2 gene found in evolved clones, and (4) S. uvarum-S. cerevisiae “reverse-fusion” MEP2 gene. PCR products that arose from the S. cerevisiae-S. uvarum MEP2 fusion gene-specific PCRs were Sanger-sequenced using the sequencing primers shown in Table S1. DNA isolation was performed, using Qiagen G-100 genomic-tip columns as described by the manufacturer, from strains GSY86, GSY2532, GSY2533, and GSY2535. The DNA was then used to prepare libraries for Illumina sequencing as described [91], using barcoded adaptors for multiplexed paired-end sequencing. Flow cells for the Illumina HiSeq 2000 platform were prepared according to manufacturer' s instructions and sequencing was performed for 100 cycles for each of paired-end reads. Read data has been deposited at the NCBI under BioProject PRJNA172024. Reads were mapped using BWA [92] to a combined S. cerevisiae – S. uvarum reference genome (S. cerevisiae genome downloaded from http: //www. yeastgenome. org on Feb 24,2011, plus S. uvarum genome assembly downloaded from http: //saccharomycessensustricto. org May 26,2011). SNPs were identified using the GATK [93], [94]. No subsequent hard-filtering of identified SNPs was performed; instead, SNPs present in the unevolved GSY86 ancestor were discarded from the analysis, and the remaining SNPs present in the evolved clones were manually inspected using Samtools tview [95], with only those showing sufficient coverage and quality given further consideration; these are shown in Table S2. The founder interspecific hybrid GSY86 and evolved strain GSY2532 were cultured in monoculture to steady state (∼15 generations) in two independent NH4+-limited chemostats each; cultures were harvested by fast-filtration on 0. 45 µm Nylon filters, frozen in liquid nitrogen, then stored at −80°C until RNA purification. Hot phenol RNA preparation was performed as described previously [96], followed by treatment with Ambion TURBO-DNAfree DNAse using manufacturer' s recommendations (Life Technologies). 2 µg of total RNA were reverse transcribed using oligo dT primer and Superscript III according to the manufacturer' s instructions (Invitrogen). Real-time qPCR was performed on a Bio-rad CFX96 cycler using SsoFast EvaGreen Supermix (Bio-rad). For GSY86, separate PCR reactions for detecting the S. cerevisiae and the S. uvarum MEP2 transcripts were performed, using primers shown in Table S1, in technical triplicates for each of the two biological replicates. Note that the primer pairs for detecting the S. cerevisiae and the S. uvarum MEP2 transcripts were first determined by PCR (using genomic DNA) to be specific for each species. Control PCR reactions (in triplicate, for each biological duplicate) for the reference S. cerevisiae TFC1 and S. uvarum YDR458C genes [97] (primers shown in Table S1) were also performed. For GSY2532, qRT-PCR was performed similarly, except using primers to detect the S. cerevisiae - S. uvarum MEP2 fusion transcript (see Table S1). Competitive chemostat experiments were performed for the interspecific hybrid GSY86 and each of three independently-evolved strains from Vessels A, B, and C, respectively (GSY2532, GSY2533 and GSY2535). Each strain was competed pairwise (in triplicate) against a common GFP-marked F1 hybrid reference strain (GSY2590, see Table 1; it is similar to GSY86 but with the S. cerevisiae parent genome containing the fluorescent GFP marker inserted into the YBR209W locus [72]). Each strain was grown to steady state, as determined by constant optical density for ∼24 h; culture volume for the reference strain was set at 500 mL, and its competitors at 300 mL. 150 ml were removed from each competitor vessel and replaced with 150 mL of the reference strain. The competition was followed for ∼15 generations. Beginning at the time of mixing (t-zero), 1 mL samples of the mixed populations were collected every 8–12 h, spun at 11,000×g for 2 min, resuspended in 0. 5 mL of 1× PBS, then stored at 4°C until FACS analysis. Flow cytometry was performed using a FACSCaliber flow cytometer (Becton Dickinson, San Jose, CA) using a 488 nm laser for excitation of GFP and signal collection using a 530-30 bypass filter. Analysis was performed using CellQuest 3. 3. Selection coefficients were determined from the linear regression of ln [Test/Reference] against generations, using methods developed by [65]. We used ANOVA followed by Tukey' s HSD to test for differences among competition coefficients.

Interspecific hybridization occurs when two different species mate and produce viable offspring. While hybrid offspring are usually sterile, like the mule, which results from a horse–donkey mating, sometimes they are fertile, creating new species. Indeed, many plant and animal species have arisen via this mechanism. Because interspecific hybridization occurs between different yeast species, and because they are such tractable models, yeast are ideally suited for experimentally investigating the genomic consequences of interspecific hybridization. We created an interspecific yeast hybrid by crossing S. cerevisiae and S. uvarum, and then studied genomic changes that occurred as it adaptively evolved in a stressful nitrogen-limiting environment. We discovered that a characteristic rearrangement between the parental species' chromosomes evolved independently many times, and always within a particular gene encoding a protein that imports nitrogen into the cell. Evolved hybrids carrying this rearrangement grew faster under nitrogen-limitation than ancestral hybrids, suggesting that the rearrangement is beneficial in nitrogen-poor environments. Our results suggest that having the genomes of two different species within a cell provides novel sources of variation for evolution to act upon, leading to adaptations that could not occur in either parental species.

Abstract Introduction Results Discussion Materials and Methods

microbial mutation genetic mutation genome evolution chromosome structure and function population genetics microbiology genome sequencing model organisms mutation types microbial evolution forms of evolution microbial physiology chromosome biology biology evolutionary genetics natural selection genetics yeast and fungal models saccharomyces cerevisiae genomics evolutionary biology genomic evolution evolutionary processes genetics and genomics

2013

Recurrent Rearrangement during Adaptive Evolution in an Interspecific Yeast Hybrid Suggests a Model for Rapid Introgression

Helminthic infections are highly endemic in Mozambique, due to limited access to healthcare and resources for disease prevention. Data on the subclinical prevalence of these diseases are scarce due to the fact that an immunological and imaging diagnosis is not often available in endemic areas. We conducted a cross-sectional study on HIV1+ patients from Beira city in order to determine the seroprevalence of cysticercosis, schistosomiasis, toxocariasis and echinoccocosis and its possible interaction with HIV infection. Patients (601) were voluntarily recruited at the Ponta Gea Health Center and their demographic and clinical data were recorded (including CD4+ cell count and antiretroviral regimen). Mean age was 39. 7 years, 378 (62. 9%) were women and 223 (37. 1%) were men. Four hundred seventy-five (475) patients (79%) were already on highly active antiretroviral therapy (HAART), and 90 started therapy after being enrolled in the study. For serological testing we used a Multiplex Western Blot IgG from LDBIO Diagnostics. The overall seroprevalence was 10. 2% for cysticercosis, 23% for schistosomiasis, 7. 3% for toxocariasis and 17. 3% for echinococcosis. Neither age nor the CD4+ count were significantly associated with the seroprevalence of the helminths studied. However, patients with CD4+ between 200–500/µl had a higher seroprevalence to all helminths than those with less than 200/µl cells/and those with more than 500 cells/µl. Female gender was significantly associated with cysticercosis and schistosomiasis, and being in HAART with toxocariasis. Headache was significantly associated with cysticercosis and toxocariasis. There was no association between epilepsy and seropositivity to any of the parasites. The study concluded that a clear understanding of the prevalence and manifestations of these coinfections, how best to diagnose subclinical cases, and how to manage diseases with concomitant antiretroviral therapy is needed. The study of HIV and helminth coinfection is a topic of great interest in endemic regions because little is known about the synergism that may exist between HIV and tisular helminths. Important questions remain regarding the increased susceptibility to helminths, HIV replication enhancement, worsening of HIV-associated neurological disorders, and increased incidence and severity of the immune reconstitution inflammatory syndrome (IRIS) following initiation of antiretroviral therapy. To clarify the interactions that probably exists it is important to determine the seroprevalence of non-intestinal helminths in HIV infected patients. In this study we have selected cysticercosis, schistosomiasis, echinococcosis and toxocariasis. Cysticercosis is emerging as a serious public health problem in the countries of Eastern and Southern Africa especially in rural subsistence farming communities, where raising cattle is not economically feasible [1]. In such areas pigs may range freely, having direct access to human feces from outdoor facilities, and veterinary inspection of meat does not exist or is inadequate, thus facilitating the continuous transmission of the disease. The increasing demand for pork meat in urban areas may result in the transport of infected meat from rural communities to large urban populations. Previous studies of abattoir records indicate the presence of porcine cysticercosis in all provinces of Mozambique [2]. Neurocysticercosis (NCC), the most serious complication of the disease, is associated with seizures, headaches, intracranial hypertension, focal neurological disorders, hydrocephalus, encephalitis, and occasionally with psychiatric manifestations and dementia [3]. Previous serological studies in Mozambique showed that 15 to 21% of apparently healthy adults were positive for cysticercosis, while in neuropsychiatric patients seroprevalence was as high as 51%, [2]. In a study conducted in Angónia District, to evaluate the association between epilepsy and NCC, of 2,023 individuals screened with a T. solium cysticercosis antigen ELISA, 15% were positive. Of these, 47% had a history of epilepsy. Additionally, 43 (57. 3%) out of 75 individuals positive for cysticercus antigens with associated epilepsy, had brain lesions identified as T. solium cysts on computed tomography scan (CT scan) [2]. Urogenital schistosomiasis is a risk factor for contracting HIV in both sexes due to its chronic immunomodulatory effects that lead to a more aggressive infection. In women in particular it causes alterations of the genital mucosa (mucosal edema, abrasion, and ulceration), that persist even after treatment with a specific anthelminthic drug [4]. A survey conducted in Mozambique to assess the prevalence of schistosomiasis in school children found infection rates of approximately 47% S. haematobium in urine and 1. 5% S. mansoni in stools [5]. It is well established that anti IgM and IgG ELISA detection shows much higher prevalences than those obtained from microscopic study, while allowing the diagnosis of coinfection by both species [6]. There is no information about the occurrence of human echinoccocosis in Mozambique although it exists in neighboring countries. In a retrospective study in Tanzania, the incidence has been established in 10 cases per 100. 000 people per year [7]. In South Africa it is an agent of morbidity with an estimated prevalence of about 137 patients hospitalized per year. A significantly increased mortality was found in patients also infected with HIV and TB [8]. A possible interaction between echinococcosis and HIV has been pointed out since a depressed immune response may lead to an increased susceptibility for both pathogens and also to a more severe hydatid disease [9]. We have not found previous data on the prevalence of human toxocariasis in Mozambique. Whilst this infection is frequently overlooked, its presence can have serious consequences. A systematic revision and a meta-analysis of available data support the evidence of a positive association between seropositivity to Toxocara spp and epilepsy. Cognition will be affected during chronic infections with hyperactive behavior in children and severe mental alterations in elderly adults [10]. In experiments on infected mice it was determined that the telencephalon and the cerebellum, (areas involved in memory and coordination), are the preferred places of larvae accumulation. Furthermore, in murine toxocariasis, the neurodegeneration is associated with the emergence of AβPP and phosphorylated tau in the brain [11]. Although factors that induce occasional migration of tisular Toxocara larvae to the brain are not well determined, it is obvious that the immune status is likely to be involved. Patients for this study were recruited at Ponta Gea Health Centre (PGHC), located 5 km from Beira Central Hospital (BCH), the second largest referral hospital, and located in the central part of the country. Those receiving care for HIV-1 infection were approached systematically about participation. Prior to enrollment, a nurse or a research doctor explained the aims of the study to potential volunteers and obtained their written consent. The study was approved by the National Bioethics Committee of Mozambique and the Human Research Protections Program of the University of California, San Diego, US. The sample size was calculated using Statcalc in Epi Info 3. 3. 2 software. At a power of 90% and a confidence interval level of 95%, a sample size of 601 was obtained. The bivariate and unconditional logistic regression analysis was carried out to analyze the association between the socio demographic and clinical variables and the prevalence of the parasites [12]. Data collection and sampling were taken from February 2011 to June 2012. Demographic and clinical data including gender, age, CD4+ count (<200,200–500 and >500), and antiretroviral treatment history, were each recorded and additionally, any clinical signs and symptoms such as headache, epilepsy that might be related to CNS infection. A 5 ml sample of venous blood was taken by venipuncture and blood serum was removed. This was put into 3 aliquots and frozen until transported from PGHC to the Parasitology Laboratory at Faculty of Medicine, Eduardo Mondlane University in Maputo. Here serological testing was carried out to determine the presence of antibodies against Taenia solium larva, Schistosoma spp, Echinococcus spp and Toxocara spp, using the multiplex Western Blot IgG kits for parasites serology from LDBIO Diagnostics (www. ldbiodiagnostics. com). Results were interpreted in each case in accordance with the detailed instructions given by the manufacturer. Patients were divided by sex, age range (18–25,26–35,36–45, +45), and whether or not they had been in HAART previous to the sample collection. We enrolled a total of 601 HIV infected patients with a mean age of 39. 7 years; 378 (62. 9%) were women and 223 (37. 1%) were men. A total of 475 (79%) patients were already on HAART and 90 started HAART after being registered in the study. In 10 cases it was not known exactly when HAART therapy had been commenced. A brain CT scan was performed in 48 patients with positive serology for cysticercosis at Beira Central Hospital (BCH) in order to detect any brain lesions or abnormalities. We defined subclinical cysticercosis as a patient with a positive Western Blot assay result but no overt neurologic signs or symptoms. All results were reported to medical personnel caring for the participants. Western Blot IgG kits indicated that 61 [10. 2%; 47 (77%) females and 14 (23%) males], were positive for cysticercosis 139 [23%; 71 (51%) females and 68 (49%) males], were positive for schistosomiasis, 44[7. 3%; 25 (56. 8%) females and 19 (43. 2%) males], were positive for toxocariasis and 104 [17. 3%; 72 (69. 2%) females and 32 (30. 8%) males], were positive for echinococcosis, and (Table 1). Seven patients were seropositive for two parasites; five of them for both echinococcosis and schistosomiasis one for schistosomiasis and cysticercosis, and the other one for cysticercosis and echinococcosis. No patients were seropositive for three or more parasites. Following analysis of disease prevalence by sex all were found to be higher in women, with gender being significantly associated with cysticercosis (p = 0. 0145) and with schistosomiasis (p = 0. 0006) (Table 2). Being in HAART was significantly associated with toxocariasis (p = 0. 0087). Although seroprevalence to all helminths was more than three times higher in the age groups above 26 years this was not statistically significant, CD4+ count was also not significantly associated with any parasites (Table 2). Of the patients studied, 9 had a history of epilepsy; among them one was seropositive for cysticercosis, two for schistosomiasis, one for toxocariasis and three for echinococcosis. However, there was no association between epilepsy and seropositivity to any of the parasites (Table 1). When stratified according to the CD4+ cell count, we noted that patients with CD4+ cell counts between 200–500 cells/µl had an overall helminthic seroprevalence rate that was 1. 6 to 2. 5 fold higher than patients with a CD4+ count less than 200 cells/µl. Among the same group with CD4+ cell count between 200–500 cells/µl there were 1,8 to 2,8 times more seropositive cases than among patients with CD4+ cell counts greater than 500 cells/µl. However these differences were not statistically significant (Table 3). When we analyzed the relationships among CD4+ cell count, HAART treatment and seropositivity for helminths, we found that, with the exception of schistosomiasis, in the group of patients with CD4+ cell count less than 200 cells/µl, those on HAART were more likely to be seropositive, although this was statistically significant only for echinococcosis 23 (39%) (p = 0. 008) (Table 3). In the group with CD4+ cell counts between 200–500 cells/µl, the seroprevalence rate was higher in patients not on HAART for cysticercosis and for echinococcosis. In the group with CD4+ cell counts higher than 500 cells/µl the seroprevalence was higher for those on HAART therapy for all helminths except for schistosomiasis, (Table 3). Among patients who reported having had headache 19 (31. 1%) were seropositive for cysticercosis and 5 (11. 4%) for toxocariasis. These findings were statistically significant; p = 0. 02 and p = 0. 05 respectively for cysticercosis and toxocariasis, (Table 3). We were able to take a CT scan of only 48 patients seropositive to cysticercosis due to the fact that the CT machine broke down several times during the study. None of these patients showed brain abnormalities. We have used a commercial test for the serodiagnosis due to its economy and reliability. The specific bands within which a result is deemed positive are explained in the manufactures instructions. Nevertheless, it cannot be ruled out that the genetic background of the population, their environmental living conditions or being infected with HIV, has not influenced the results in some way. Seroprevalence for tissue helminths in HIV infected patients was demonstrated to be high, with schistosomiasis being the most frequent affection (23%), followed by echinococcosis (17. 3%), cysticercosis (10. 2%) and toxocariasis (7. 3%). The data analysis is complex as there are many factors involved but the existence of a high percentage of HIV-helminths coinfections is clear. We have conducted previous studies on schistosomiasis and cysticercosis in Mozambique [2], [5], but not investigated the prevalence of toxocariasis and echinococcosis, so one limitation of our study is the lack of seroprevalence data of echinococcosis and toxocariasis in HIV negative population. Consequently these results must be interpreted with care as it is difficult to obtain the right conclusions. We did not search for hydatid cysts in the seropositive patients due to financial limitations. These helminthiasis are chronic conditions that, in the absence of treatment, remain in the body for prolonged periods, although by themselves are not associated with immunosuppression because they are not opportunistic pathogens. As we have indicated one of the most common presentations of AIDS in our environment is uncontrollable diarrhea. Patients attending health services for this reason are routinely prescribed albendazol, mebendazol or cotrimoxazol. We do not know whether these drugs may have in some way influenced the results of this study. CD4+ cell count was not significantly associated with any parasite. However, the global seroprevalence was higher (although not significant) in patients between 200–500 CD4+/µl than in those with count less than 200 or greater than 500 (Table 3). A possible explanation could be that although all groups have the same susceptibility to the infections patients with CD4+ count <200 are highly immunocompromised having lost the ability to produce antibodies, or producing them at minimum levels undetectable with our assay. It is worth considering that since these are chronic conditions and non-opportunistic parasites some of these patients might have been already infected before contracting the HIV. A population study among the non-HIV infected people from the same region would provide data on the prevalence of these conditions at different ages. In the case of patients with CD4+ >500 cells the explanation could be because their immune system is not yet compromised thus they have more ability to clear out the parasites before they become established in the body. The high seroprevalence of schistosomiasis is not a surprise, as it is a common condition in the country [5]. Its significant association with female gender probably contributes further to the risk factors that lead to the higher HIV prevalence in females, due to the alterations in the integrity of the female genital mucosa that favor the transmission of the HIV virus in addition to the immunomodulatory effects of chronic schistosomiasis. The WHO recommended that regular treatment of children with praziquantel needs to be extended to adults and prioritized in national programs as a possible means of further preventing HIV infections in sub-Saharan countries [4]. In a previous serological study of echinococcosis using ELISA and AgB5 a total of 269 sera from apparently healthy young people from a suburban area of Maputo city were negative [13]. In the present study echinococcosis seroprevalence was surprisingly high, 17. 3% (104 patients). In contrast to the high seroprevalence we found in this study, there are no reported cases of human hydatidosis in the country. The only one reported was post mortem, a patient who had been working in South Africa and it was therefore impossible to establish the source of his infection. We did not search for hydatic cysts in liver or lung using ultrasonography or CT scan due to the financial limitation. In neighboring countries like South Africa, Tanzania and Zimbabwe surgical interventions in patients with hydatid cysts are performed. Cross-reactions with cysticercosis can be discarded as only one of the sera tested positive for both helminthiasis. Recent studies have demonstrated that diversity of Echinococcus species and strains is greater in Africa than on any other continent. Variability by molecular epidemiological research has been studied in South Africa where E. granulosus sensu stricto G1–G3, E. canadiensis G7 and E. ortleppi G5 are all known to exist. Their global seroprevalence is 17. 0% with the Eastern Cape Province showing an even higher rate of 30. 4% [14]. Much research is needed in our country in order to clarify which species and strains exist, their infectivity for humans, their clinical manifestations and the hosts implicated in their life cycles in order to establish preventive measures. It has been suggested that HIV and TB induced immune modulation seems to affect the clinical course of echinococcosis [15]. In fact we have found in this study a strong association with HAART in patients with CD4+ cell count less than 200/µl and seropositivity to echinococcosis. However, is difficult to interpret this result because of the lack of data on this disease in our country. It is well known that different Echinococcus strains differ in their preferential location (liver, lungs, etc.) both in humans and in other hosts and can also be incompletely developed in some giving rise to infertile cysts (acefalocystic). In African countries only now are we beginning to study this diversity, which, it is hoped will shed light on human hydatidosis [15], [16]. A recent epidemiological study in Kenya looking at the prevalence of hydatidosis in cattle, sheep and goats shows a significant increase compared with data about three decades ago [17]. Even the introduction of livestock from other countries after the civil war could be a factor involved in echinococcosis in a country like Mozambique, with such an abundant canine population in its rural environment. Cysticercosis seroprevalence did not surprise us because previous studies had already established the problem when identifying what our priorities should be in researching this area [2]. Unfortunately cysticercosis continues to be ignored, underdiagnosed and neglected. As Mozambique is still lacking enough epidemiological data about human taeniasis and cysticercosis, it is impossible to draw firm conclusions on its prevalence and geographical distribution. Its impact on human health in both HIV positive and negative populations in our country is still unknown. If we take into account that patients who are immunocompromised have partially lost the ability to produce antibodies, the seroprevalence to those helminths is probably underestimated. Previous studies in Mozambique showed that in apparently healthy people cysticercosis seroprevalences were much higher than we have determined in this study [2]. Magnetic Resonance Imaging (MRI) is the ideal technique to detect neurocysticercosis lesions, but we have no access to this equipment. Image brain studies are scarce in Mozambique and mainly accomplished by means of CT scanning, a lesser expensive technique. Although only 48 patients underwent CT, we did not find any with brain lesions. This could be due to the small sample size, or related to the genetic variants of T. solium that circulates in our setting. It has been documented that there is a Latino/African genetic variant and an Asian genetic variant of T. solium each correlated to a different clinical presentation. Serological analysis of cyst fluid of T. solium cysticerci obtained in China, Indonesia, Mozambique and Ecuador indicates geographical differences in their banding patterns, since there are some varieties of manifestation of neurocysticercosis with or without subcutaneous cysticercosis and possible differences in the pathology of cysticercosis worldwide [18]. There is growing evidence that long-standing headache is one of the frequent symptoms after epilepsy in patients with NCC. Our study has shown that people with positive serology had significantly more headaches than those with negative serology and this should be taken into account in the future for the diagnosis of cysticercosis. The immunomodulatory mechanisms induced by cysticercosis showed themselves to be very complex [19]. These finding seems logical, being a metacestode that can not only occupy different locations but also be found in various different stages of development, degeneration or calcification. Genetic variances of the host might explain the susceptibility of people with NCC to develop symptomatic diseases, and polymorphisms of toll-like receptor 4 (TLR 4) have been implicated in this process [20]. In Mozambique we do not know what genetic variants circulate and further studies should be directed at identifying the variants we have as we were exposed to European migration as well as Asian migration in the past. Toxocariasis is a zoonotic cosmopolitan infection more prevalent in tropical areas with poor hygiene, where soil is highly contaminated with Toxocara spp eggs from dogs and cats and can survive for a long time in humid soil. There are very few studies on the prevalence of Toxocara ocular and visceral larva migrans in sub-Saharan countries. Seroprevalence is high (44. 6%) among Swaziland children living in rural slums [21], in Cameroon a 36,3% of young people under 20 years were seropositive, and from 14 people with epilepsy 5 were seropositive to Toxocara antigens [22]. We do not have any comparative data on HIV-noninfected population from our country nor any other work on the seroprevalence of toxocariasis in HIV+ patients, but we have found a significant association with being in HAART and also with headache. The low frequency of sera reacting to more than one parasite antigen could be explained by the fact that not all have the same path of transmission and to the impaired immune system that underestimates the seroprevalence due to the lower antibodies titles. Compounding these problems is the fact that even before HIV has been diagnosed, these patients are routinely given albendazole, mebendazole or other medication such as cotrimoxazole to treat complaints such as diarrhea, which is one the most common presentation of AIDS in our settings. Even though there are given albendazole or mebendazol at a certain point for their disease it does not explain at all this low rate of multireactive sera because the dose is insufficient to kill the larvae of T. solium and Echinococcus. As a general result of this study, we conclude that in our wormy world HIV+ patients showed a high seroprevalence versus the four non-intestinal helminths surveyed. These chronic conditions are all derived from environmental contamination with human and animal feces. Much remains to be learned about its synergism with HIV and other associated pathogens, being important to approach in our country the study of echinococcosis and toxocariasis, possibly two new still uninvestigated zoonoses.

In Mozambique many parasitic diseases persist as a result of low living standards and environmental contamination from human and animal fecal waste. Parasites undermine the health of Mozambique' s poorest inhabitants and will be difficult to eradicate without a considerable improvement in the sanitary conditions. Many helminthic infections are undiagnosed because of the need to employ immunologic and image analysis methods, both of which are very costly. We want to know the scope of these co-infections, their possible interaction with HIV, and the course of the disease. We have therefore investigated the seroprevalence against four tisular helminth diseases (cysticercosis, schistosomiasis, toxocariasis and echinococcosis) in a group of HIV infected patients, obtaining data previously absent and relevant conclusions that will enable us to continue working in this field and design strategies in order to improve the quality of life of our people.

Abstract Introduction Methods Results Discussion

medicine and health sciences

2014

A Cross-sectional Serological Study of Cysticercosis, Schistosomiasis, Toxocariasis and Echinococcosis in HIV-1 Infected People in Beira, Mozambique

Lethal acrodermatitis (LAD) is a genodermatosis with monogenic autosomal recessive inheritance in Bull Terriers and Miniature Bull Terriers. The LAD phenotype is characterized by poor growth, immune deficiency, and skin lesions, especially at the paws. Utilizing a combination of genome wide association study and haplotype analysis, we mapped the LAD locus to a critical interval of ~1. 11 Mb on chromosome 14. Whole genome sequencing of an LAD affected dog revealed a splice region variant in the MKLN1 gene that was not present in 191 control genomes (chr14: 5,731,405T>G or MKLN1: c. 400+3A>C). This variant showed perfect association in a larger combined Bull Terrier/Miniature Bull Terrier cohort of 46 cases and 294 controls. The variant was absent from 462 genetically diverse control dogs of 62 other dog breeds. RT-PCR analysis of skin RNA from an affected and a control dog demonstrated skipping of exon 4 in the MKLN1 transcripts of the LAD affected dog, which leads to a shift in the MKLN1 reading frame. MKLN1 encodes the widely expressed intracellular protein muskelin 1, for which diverse functions in cell adhesion, morphology, spreading, and intracellular transport processes are discussed. While the pathogenesis of LAD remains unclear, our data facilitate genetic testing of Bull Terriers and Miniature Bull Terriers to prevent the unintentional production of LAD affected dogs. This study may provide a starting point to further clarify the elusive physiological role of muskelin 1 in vivo. Acrodermatitis enteropathica in humans (OMIM #201100) is an inherited disorder of zinc metabolism. Affected patients display an inflammatory rash, diarrhea and a general failure to thrive [1–3]. This disease is caused by variants in the SLC39A4 gene encoding a zinc transporter that mediates the uptake of dietary zinc in the gut. Clinical signs in patients will ameliorate or even resolve upon oral supplementation with zinc [4]. A similar SLC39A4 associated hereditary zinc deficiency exists in cattle [5]. In Bull Terriers, a related phenotype termed lethal acrodermatitis (LAD) has been reported in the scientific literature as early as 1986 [6]. LAD is inherited as a monogenic autosomal recessive trait. Affected puppies show characteristic skin lesions on the feet and on the face, diarrhea, bronchopneumonia, and a failure to thrive. The skin lesions consist of erythema and tightly adherent scales, erosions or ulcerations with crusts involving primarily the feet, distal limbs, elbows, hocks, and muzzle. Later on, hyperkeratosis of the footpads and deformation of the nails occur. LAD affected dogs also show a coat color dilution in pigmented skin areas. An abnormally arched hard palate impacted with decayed, malodorous food is a characteristic clinical marker for the disease (Fig 1) [6–8]. LAD dogs are immunodeficient with a reduction in serum IgA levels and frequently suffer from skin infections with Malassezia or Candida [9,10]. LAD manifests clinically in the first weeks of life. Affected puppies typically die before they reach an age of two years, either due to infections such as bronchopneumonia or because they are euthanized when their paw pad lesions become very severe and painful. They grow slower than their non-affected littermates and at the age of one year have about half the body weight and size of an unaffected dog [8]. Some, but not all studies found reduced levels of zinc in the serum of LAD affected dogs [6,8, 11]. In contrast to acrodermatitis enteropathica in humans, oral or intravenous supplementation of zinc does not lead to an improvement of the clinical signs in LAD affected dogs [6]. A proteomic analysis reported changes related to inflammatory response in the liver of LAD affected puppies [12]. In the present study, we performed a genome-wide association study (GWAS) followed by a whole genome sequencing approach to unravel the causative genetic variant for LAD in Bull Terriers and Miniature Bull Terriers. We performed a GWAS with genotypes from 78 Bull Terriers and Miniature Bull Terriers. After quality control, the pruned dataset consisted of 22 LAD cases, 48 controls and 76,419 markers. We obtained a single strong association signal with 57 markers exceeding the Bonferroni-corrected genome-wide significance threshold after adjustment for genomic inflation (PBonf. = 6. 5 x 10−7). All significantly associated markers were located on chromosome 14 within an interval spanning from 0. 9 Mb– 10. 6 Mb. The three top-associated markers all had a P-value of 1. 4 x 10−9 and were located between 5. 2 Mb– 5. 9 Mb on chromosome 14 (Fig 2). To narrow down the identified region, we visually inspected the genotypes of the cases to perform autozygosity mapping. We searched for homozygous regions with allele sharing and found one region of ~1. 11 Mb, which was shared between all 22 cases. The critical interval for the causative LAD variant corresponded to the interval between the first flanking heterozygous markers on either side or chr14: 5,248,244–6,355,383 (CanFam 3. 1 assembly). We sequenced the genome of an affected Bull Terrier at 24x coverage and called single nucleotide variants (SNVs) and small indel variants with respect to the reference genome (CanFam 3. 1). We then compared these variants to whole genome sequence data of 3 wolves and 188 control dogs from genetically diverse breeds. This analysis identified five private homozygous variants in the critical interval in the affected dog (Table 1, S1 Table). Four of these five variants were intergenic and classified as “modifier” by the SNPeff software. The remaining fifth variant was located within the 5’-splice site of intron 4 of the MKLN1 gene and its SNPeff impact prediction was “low”. The formal designation of this variant is chr14: 5,731,405T>G or MKLN1: c. 400+3A>C (S1 Fig). We confirmed the presence of this variant by Sanger sequencing (Fig 3). As MKLN1: c. 400+3A>C represented the only plausible candidate causative variant, we genotyped 251 Bull Terriers, 89 Miniature Bull Terriers, and 462 dogs from 62 other breeds for this variant (Table 2). The variant showed perfect association with the LAD phenotype in Bull Terriers and Miniature Bull Terriers (PFisher = 4. 8 x 10−58). All 46 available cases were homozygous for the variant, whereas the unaffected dogs were either homozygous wildtype or heterozygous. The test dogs included a subset of unaffected Bull Terriers and Miniature Bull Terriers from Finland, which were not specifically collected for this study and therefore considered representative for the general population. The 166 Finnish dogs contained 37 heterozygous dogs (22%). The variant was not found in any of the tested dogs from other breeds. To assess the putative impact of the MKLN1 variant on splicing, we analyzed the frequency of the wildtype and mutant sequence motifs in a compilation of 186,630 human 5’-splice sites [13,14]. The canine wildtype sequence TAGgtaagg was identical to the sequence of 276 human 5’-splice sites, while the mutant sequence motif TAGgtcagg occurred in only 3 human 5’-splice sites. The very low frequency of the mutant sequence motif suggested that MKLN1: c. 400+3A>C might affect the efficacy of the splicing process. Several other pathogenic A>C transversions at 5’-splice sites’ position +3 with subsequent exon skipping have been described in the literature [15–17]. We experimentally analyzed MKLN1 transcripts in skin RNA from an LAD affected dog with the homozygous mutant C/C genotype in comparison to a healthy control dog (A/A genotype). RT-PCR with primers located at the exon 2/3 and exon 5/6 boundaries yielded a cDNA fragment of the expected size in the control dog, but not in the LAD affected dog. In the LAD affected dog, a very clean cDNA amplicon lacking exon 4 was obtained. This experiment demonstrated a complete skipping of exon 4 in MKLN1 transcripts as consequence of the genomic MKLN1: c. 400+3A>C variant (r. 312_400del89; Fig 4). If translated, the mutant transcript was predicted to result in a severely truncated protein containing only the first 105 of a total of 735 amino acids of the wildtype protein (p. (Gly105SerfsTer10); S2 Fig). In the present study we identified a splice defect in the canine MKLN1 gene in Bull Terriers with LAD. The combination of GWAS and haplotype analysis localized the causative variant to a relatively small chromosomal region with only a few characterized genes including MKLN1. The splice region variant in MKLN1 was the only plausible variant within this critical interval that showed the expected genotype concordance with the LAD phenotype in a large cohort of more than 300 Bull Terriers and ~500 dogs from other breeds. The identified MKLN1: c. 400+3A>C variant resulted in exon 4 skipping and a frameshift as 89 nucleotides were missing from the mutant transcripts. It therefore seems likely that mutant transcripts are degraded by nonsense-mediated mRNA decay. Considering the strong genetic association of the variant with the phenotype and the fact that we demonstrated a functional defect on the MKLN1 transcript level, we think that our data strongly suggest the causality of the MKLN1: c. 400+3A>C variant for LAD in Bull Terriers and Miniature Bull Terriers. MKLN1 encodes the widely expressed intracellular protein muskelin 1, also known as TWA2. The function of muskelin 1 is only partially understood. It was originally described as a protein that mediates adhesive and cell-spreading responses to thrombospondin 1, an extracellular matrix adhesion molecule [18]. However, different studies suggested that the function of muskelin 1 goes beyond this pathway and are also supported by the fact that muskelin 1, which has homologs in invertebrates and even fission yeast, evolved earlier than the vertebrate-specific thrombospondin 1 [19,20]. Muskelin 1 is a multidomain protein with an N-terminal discoidin domain, a LisH / CTLH tandem domain, and six C-terminal Kelch repeats, which forms homotetramers [21]. The LisH domain was shown to be crucial for muskelin 1 dimerization and cytoplasmic localization, and, together with the head-to-tail interaction via the discoidin domain, also for the tetramerization of muskelin 1 [20,21]. Consistent with its multidomain structure and ubiquitous expression, diverse binding partners have been reported for muskelin 1. It binds prostaglandin EP3 receptor isoform α [22] and heme-oxidase 1, which counteracts inflammatory and reactive oxygen species induced damage [23]. It is part of the CTLH complex, the homolog of yeast E3 ubiquitin ligase, where it binds to RanBPM and Twa I [24–26] and interacts with the cardiogenic transcription factor TBX-20 [27]. In the rat lens, muskelin 1 is a substrate of Cdk5 and interacts with the Cdk5 activator p39 [28]. Also in lens, it was shown that p39 links muskelin 1 to myosin II and stress fibers [29]. Mkln1-/- knockout mice are viable and do not have skin lesions comparable to those in Bull Terriers with LAD. However, they exhibit a subtle coat color dilution phenotype similar to that seen in LAD affected dogs. In these Mkln1-/- knockout mice, muskelin 1 was identified as a protein required for GABAA receptor endocytosis and trafficking in neurons via direct interaction with the α1 subunit of GABAA receptors and the motor proteins dynein and myosin VI. The dilute coat color of Mkln1-/- knockout mice suggested that muskelin is a trafficking factor involved in several different intracellular transport processes, possibly including melanosome transport [30]. The lacking skin lesions in Mkln1-/- knockout mice raise the questions whether muskelin 1 depletion does not result in disease in mice; whether their clinical signs would only manifest at a (much) older age; or whether the sterile environment of the laboratory animals prevented infections and thus the development of skin lesions. In the latter case, LAD would be a primary immunodeficiency disorder, in agreement with the observation of lower IgA levels and higher susceptibility to microbial infection in LAD affected dogs [8–10,12]. Given the diverse known protein-protein interactions of muskelin 1, it is however likely that absence of muskelin 1 leads to dysfunctions beyond the immune system. In humans, an intronic SNV in MKLN1 was associated with urinary potassium excretion in Korean adults and another intronic MKLN1 SNV with early bipolar disorder [31,32]. Furthermore, MKLN1 has been associated with asthma in independent GWASs. A SNV in MKLN1 ranked among the top 100 SNVs associated with childhood asthma in a study sample of 429 affected-offspring trios from a European American population [33]. A different SNV in the 5’-UTR of MKLN1 was associated with asthma in a population including patients with severe or difficult-to-treat asthma [34]. In the ExAC database, only one MKLN1 missense, but no nonsense, frameshift or splice site variants present in a homozygous state were found [35,36]. Furthermore, the probability of loss of function (LoF, specified as nonsense, splice acceptor, and splice donor variants) tolerance was estimated to be 1. 00, indicating that the MKLN1 gene is extremely LoF intolerant [37]. Therefore, it is conceivable that loss of function variants on both alleles might lead to severe phenotypes in humans. To our knowledge, no link between muskelin 1 and zinc or copper metabolism has been reported to date. While acrodermatitis enteropatica in humans and acrodermatitis in cattle clinically resemble LAD in dogs, these diseases may be caused by completely different molecular mechanisms. The fact that findings on zinc levels in the few published studies on LAD affected dogs were contradictory and zinc supplementation did not lead to improvement of lesions [6] support this hypothesis. In conclusion, we identified the MKLN1: c. 400+3A>C variant leading to a splice defect in the MKLN1 gene as candidate causative variant for LAD in Bull Terriers and Miniature Bull Terriers. The molecular pathogenesis of LAD remains unclear. Our data facilitate genetic testing of Bull Terriers and Miniature Bull Terriers to prevent the unintentional breeding of LAD affected dogs. LAD affected dogs may serve as models to further clarify the elusive physiological role of muskelin 1 in vivo. All animal experiments were performed according to the local regulations. The dogs in this study were examined with the consent of their owners. The study was approved by the “Cantonal Committee For Animal Experiments” (Canton of Bern; permits 22/07,23/10, and 75/16). Bull Terriers with their characteristic egg-shaped head were founded as a dog breed in the 1850s in the United Kingdom. Originally, there were no size standards in this breed and smaller dogs were bred as a variety of the regular Bull Terrier. Eventually, two sub-populations formed and the Miniature Bull Terrier with a maximum height of 35. 5 cm was recognized as an independent breed in 1991 by the American Kennel Club (AKC) and in 2011 by the European Fédération Cynologique Internationale (FCI). Therefore, Bull Terriers and Miniature Bull Terriers share a common ancestral gene pool, but represent independent closed populations today. This study included samples from 251 Bull Terriers (41 LAD cases / 210 controls) and 89 Miniature Bull Terriers (5 LAD cases / 84 controls). Case/controls status was based on owners’ reports. We additionally used 462 dogs from 62 breeds, which were assumed to be free of the disease allele (S3 Table). Skin biopsies were taken from two LAD affected Bull Terriers from toe, nose, lip, and forearm and fixed in 10% buffered formalin for 24 hours. Biopsies were processed, embedded in paraffin and sectioned at 4 µm. Skin sections were stained with hematoxylin and eosin. The histopathology was performed by veterinary pathologists (BR, Dipl. -ECVP, and SH). Two further biopsies from comparable sites of the same dogs were submerged in RNAlater solution for subsequent RNA isolation. We isolated genomic DNA from EDTA blood samples. Seventy-eight dogs were genotyped for either 173,662 or 218,256 SNVs on the illumina canine_HD chip. The raw SNV genotypes are available at https: //www. animalgenome. org/repository/pub/BERN2017. 1208/. The initial dataset consisted of 78 dogs and 220,853 markers. Using Plink version 1. 9 [38] we excluded markers that were not located on autosomes or the X chromosome (n = 2,327) and markers with a genotyping rate lower than 90% (n = 49,814). Using the R package GenABEL [39] and the command “check. markers”, dogs with a call rate < 90% (n = 3), ibs > 95% (n = 0), high individual heterozygosity (FDR = 0. 01) (n = 1, included in dogs with low call rate) as well as markers with a maf < 1% (n = 55,931) and a genotyping rate < 90% (n = 10,951) were excluded. Five outliers in the multidimensional scaling plot based on a genomic distance matrix were also removed. In a second quality control step, markers deviating from Hardy-Weinberg equilibrium (FDR = 0. 2) in controls (n = 1,239), markers with a genotyping rate < 90% (n = 0) and maf < 1% (n = 26,215) were excluded, resulting in a final dataset of 70 dogs (22 cases, 48 controls) and 76,419 markers. A polygenic model of the hglm package [40], with a kinship matrix based on autosomal markers in the cleaned dataset as random effect, was estimated and a score test for association using the function “mmscore” was performed. The genomic inflation factor was 1. 16. We corrected for multiple testing using Bonferroni correction with a significance level of 0. 05. QQ plots were created using qqman version 0. 1. 4 [41]. We visually inspected plink tped files for the region of interest on chromosome 14 using Excel and searched for homozygous regions with haplotype sharing in cases with a call rate >90%. The first flanking heterozygous markers on either side of the homozygous region in 22 cases defined the borders of the critical interval. An Illumina PCR-free TruSeq fragment library with 350 bp insert size of an LAD affected Bull Terrier was prepared. We collected 219 million 2 x 150 bp paired-end reads or 24x coverage on a HiSeq3000 instrument. The reads were mapped to the dog reference genome assembly CanFam3. 1 and aligned using Burrows-Wheeler Aligner (BWA) version 0. 7. 5a [42] with default settings. The generated SAM file was converted to a BAM file and the reads were sorted by coordinate using samtools [43]. Picard tools (http: //sourceforge. net/projects/picard/) was used to mark PCR duplicates. To perform local realignments and to produce a cleaned BAM file, we used the Genome Analysis Tool Kit (GATK version 2. 4. 9,50) [44]. GATK was also used for base quality recalibration with canine dbsnp version 139 data as training set. The sequence data were deposited under the study accession PRJEB16012 and sample accession SAMEA4504844 at the European Nucleotide Archive. Putative SNVs were identified in each of 192 samples (S2 Table) individually using GATK HaplotypeCaller in gVCF mode [45]. Subsequently all sample gVCF files were joined using Broad GenotypeGVCFs walker (-stand_emit_conf 20. 0; -stand_call_conf 30. 0). Filtering was performed using the variant filtration module of GATK using the following standard filters: SNVs: Quality by Depth: QD < 2. 0; Mapping quality: MQ < 40. 0; Strand filter: FS > 60. 0; MappingQualityRankSum: MQRankSum < -12. 5; ReadPosRankSum < -8. 0. INDELs: Quality by Depth: QD < 2. 0; Strand filter: FS > 200. 0. The functional effects of the called variants were predicted using SnpEFF software [46] together with the NCBI annotation release 104 on CanFam 3. 1. For the filtering of candidate causative variants in the case, we used 191 control genomes, which were either publicly available [47] or produced during other projects of our group or contributed by members of the Dog Biomedical Variant Database Consortium. A detailed list of these control genomes is given in S2 Table. We used the dog CanFam 3. 1 reference genome assembly for all analyses. Numbering within the canine MKLN1 gene corresponds to the accessions XM_005628367. 3 (mRNA) and XP_005628424. 1 (protein). Numbering within the human MKLN1 gene corresponds to the accessions NM_013255. 4 (mRNA) and NP_037387. 2 (protein). We used Sanger sequencing to confirm the candidate variant MKLN1: c. 400+3A>C and to genotype the dogs in this study. A 797 bp fragment containing the variant was PCR amplified from genomic DNA using AmpliTaq Gold 360 Master Mix (Life Technologies) and the primers CCATGCACTGTAGCCACATC and TGGAAAAGGTTCCACTTGAAAT. After treatment with shrimp alkaline phosphatase and endonuclease I, PCR products were directly sequenced on an ABI 3730 capillary sequencer (Life Technologies). We analyzed the Sanger sequence data using the software Sequencher 5. 1 (GeneCodes). RNA was extracted from skin samples using the RNeasy Fibrous Tissue Mini Kit (Qiagen). The tissue was first finely crushed by mechanical means using TissueLyser (Quiagen), and RNA was extracted by centrifugation following the instructions by the manufacturer. Total mRNA was reverse transcribed into cDNA using the SuperScript IV Reverse Transcriptase kit (Thermo Fisher) with oligo d (T) primers. A PCR on the synthesized cDNA was carried out using primer MKLN1_c_F2, CCTCCCCAGTACTTGATTCTG, located at the boundary of exons 2 and 3, and primer MKLN1_c_R2, TTCCTGTTCACGGTACTTGC, located at the boundary of exons 5 and 6 of the MKLN1 gene. The products were analyzed on a Fragment Analyzer capillary gel electrophoresis instrument (Advanced Analytical). The sequence of the obtained RT-PCR products was confirmed by Sanger sequencing as described above.

Lethal acrodermatitis (LAD) is an autosomal recessive hereditary disease in dogs. It is characterized by poor growth, immune deficiency and characteristic skin lesions of the paws and of the face. We mapped the LAD locus to a ~1. 11 Mb segment on canine chromosome 14. Whole genome sequence data of an LAD affected dog and 191 controls revealed a candidate causative variant in the MKLN1 gene, encoding muskelin 1. The identified variant, a single nucleotide substitution, MKLN1: c. 400+3A>C, altered the 5’-splice site at the beginning of intron 4. We experimentally confirmed that this variant leads to complete skipping of exon 4 in the MKLN1 mRNA in skin. Various cellular functions have been postulated for muskelin 1 including roles in intracellular transport processes, cell morphology, cell spreading, and cell adhesion. Our data from dogs reveal a novel in vivo role for muskelin 1 that is related to the immune system and skin. MKLN1 thus represents a novel candidate gene for human patients with unsolved acrodermatitis and/or immune deficiency phenotypes. LAD affected dogs may serve as models to gain more insights into the function of muskelin 1.

Abstract Introduction Results Discussion Materials and methods

genome-wide association studies animal types medicine and health sciences pathology and laboratory medicine variant genotypes vertebrates pets and companion animals dogs animals mammals genetic mapping signs and symptoms genome analysis sequence motif analysis mammalian genomics zoology research and analysis methods sequence analysis bioinformatics lesions animal genomics haplotypes eukaryota diagnostic medicine heredity database and informatics methods genetics biology and life sciences genomics amniotes computational biology organisms human genetics

2018

MKLN1 splicing defect in dogs with lethal acrodermatitis

Learning the reliability of different sources of rewards is critical for making optimal choices. However, despite the existence of detailed theory describing how the expected reward is learned in the basal ganglia, it is not known how reward uncertainty is estimated in these circuits. This paper presents a class of models that encode both the mean reward and the spread of the rewards, the former in the difference between the synaptic weights of D1 and D2 neurons, and the latter in their sum. In the models, the tendency to seek (or avoid) options with variable reward can be controlled by increasing (or decreasing) the tonic level of dopamine. The models are consistent with the physiology of and synaptic plasticity in the basal ganglia, they explain the effects of dopaminergic manipulations on choices involving risks, and they make multiple experimental predictions. In situations where actions are associated with rewards, knowledge of the reliability of rewards for alternative choices is critical for selecting the optimal action. Normative models have suggested that optimal foraging requires adaptively switching between risk aversion and risk seeking depending on the circumstance [1,2]. Indeed, experimental data suggest that humans and animals tend to seek or avoid choice options with reward uncertainty in different situations [1,3]. To implement such policies, animals and humans need to have estimates of the reward variability associated with different sources, as well as the ability to control how this variability should influence their choices. In addition, knowledge of the reliability of reward feedback is important for learning about the mean reward, as it sets the optimal learning rate. Indeed, in high uncertainty situations, a single new data point should not influence the animal’s previously held estimate as strongly as it would in situations where the uncertainty associated with the data point is fairly low [4]. Furthermore, the estimate of reliability of rewards is helpful in optimizing the exploration-exploitation trade-off [5], because when an animal wishes to find which action yields the highest average reward, it takes more samples to get an accurate estimate of the mean reward for actions with more variable rewards. Hence, such actions should be preferably explored. One of the key regions of the brain underlying action selection is the basal ganglia (BG). The BG is thought to be involved in learning the expected values of rewards that are associated with given actions and in selecting the actions associated with the highest expected values while inhibiting the others. The learning process in BG is facilitated by neurons releasing dopamine (DA), which encode the reward prediction error, defined as the difference between reward obtained and expected [6,7]. This signal allows BG to update its estimates of reward accordingly [8,9]. The pathologies that affect the function of BG influence how it learns or makes decisions in situations involving uncertainty. For instance, a subset of patients with Parkinson’s disease, who suffer from selective death of dopaminergic neurons in the substantia nigra in the midbrain, are impaired in a task involving choices between options with different spreads of their respective reward distributions [10]. When they are on medication (DA agonist), these patients exhibit a well-reported phenomenon of obsessive gambling, in which the patients seem to exhibit a change in their subjective values of risk and reward [11]. This change can be reversed by taking the patients off medication [12]. Additionally, manipulating the levels of dopamine in humans and animals adjusts their decision making under risk [13]. These pieces of evidence suggest that uncertainty is encoded in BG (but one has to note that although BG is the main target of dopaminergic projections, DA neurons also innervate cortex, so some of the effects mentioned above may also have cortical contribution). While computational models have been developed to explain how BG can estimate the expected reward [8,9, 14,15], it is still unclear how the reliability of the reward can be estimated in BG, given its anatomical and physiological properties. Here we show that there exists a class of models consistent with the physiology of BG that can at once learn both the expected reward from a given action and the reliability of the reward, i. e. , the spread of its probability distribution. We then show how the models can use learned information about reward uncertainty in decision making, and how the models can account for the effect of dopaminergic medications on decision making in tasks involving risk. In the next section (“Models”), we review previously proposed models of reinforcement learning in BG, on which our models are built. The new models that can learn reward uncertainty are presented in Section “Results”. Readers familiar with the actor-critic model [16] and Opponent Actor Learning model (OpAL) [15] can skip directly to “Results”. This framework assumes that BG estimates average rewards for selecting different actions. Let Q i (t) denote an estimate of expected reward for selecting the action i on trial t. Let us assume that after selecting the action, a reward r (t) is provided, which comes from a distribution with mean μi and standard deviation σi. We start by considering an abstract Rescorla-Wagner rule [17] for estimating the expected reward for a given action. According to this rule, after receiving a reward, the expected reward is updated in the following way: Q i (t + 1) = Q i (t) + α r (t) - Q i (t) (1) According to the above equation, the change in the estimate of the expected reward is proportional to the reward prediction error (r (t) - Q i (t) ), scaled by the learning rate constant α, where 0 < α < 1. It is intuitive to see why this rule works: If r (t) is underestimated, our estimate Qi will increase (i. e. , Q i (t + 1) > Q i (t) ). If r (n) is overestimated, Qi will decrease, and if r (t) is estimated perfectly (Q i (t) = r (t) ), then Qi will remain the same. In addition, the amount we increment by will be scaled by the magnitude of the prediction error (r (t) - Q i (t) ), so that we learn more quickly when we have a lot of learning to do than when our estimate is quite close to the true mean already. Also note that having α < 1 ensures that the new data point updates our estimate but does not completely replace it (as would be the case if α were in fact equal to 1), an implicit acknowledgement of the existence of uncertainty in the reward and noise in the system. The actor-critic model [16] includes two components: an actor that learns tendencies to select particular actions and a critic that learns an overall value of the current context or state. In the actor-critic model, the value V of being in this state is learned by the critic according to the standard Rescorla-Wagner rule [17] (cf. Eq 1): V (t + 1) = V (t) + α r (t) - V (t) (2) Note that V (t) is updated regardless of which action i is selected, so V (t) is not an estimate of expected reward associated with a particular action, but rather an average reward in the current state. In the standard actor-critic model, after choosing action i, the tendency to choose it, which we denote by Qi, is learned by the actor using the following update rule: Q i (t + 1) = Q i (t) + α r (t) - V (t) (3) According to the above equation, the tendency to choose action i is also modified proportionally to the reward prediction error, i. e. , it is increased if the action resulted in a higher reward than expected by the critic and decreased if the reward was below expectation. The actor-critic model naturally maps on the matrix-patch organization of the striatum [18]. Such mapping assumes that V (t) is encoded in the synapses between cortical neurons selective for the current context and striatal patch neurons, as shown in Fig 1. The patch neurons directly inhibit dopaminergic neurons [19], so that if the dopaminergic neurons also receive input encoding reward, then their activity may encode r (t) − V (t). The actor part of the model is mapped on matrix neurons [18] that send projections to the output nuclei, which in turn project to areas controlling movement, so they can affect which movement is selected. Finally, the dopaminergic neurons modulate plasticity of the synapses of both patch neurons and matrix neurons. It is worth adding that some studies map actor and critic on dorsal and ventral striatum respectively [20], but this mapping is related to the matrix-patch mapping, as the patch neurons are more common in ventral than dorsal striatum [21]. A recent model called Opponent Actor Learning (OpAL) [15] takes into account the fact that the matrix neurons can be subdivided into two groups, which express D1 and D2 DA receptors, respectively. These project through different nuclei of BG, as shown in Fig 1 [22] and have opposite effects on movement initiation [23,24]. In particular, D1 neurons project through the “direct” pathway to the output nuclei, and their activity facilitates movements [25] because they inhibit the output nuclei and thus release thalamus from inhibition. By contrast, D2 neurons project through the “indirect” pathway, and their activity inhibits movement [25]. The OpAL model describes learning about the tendencies to choose or inhibit actions i in a given state, which we will denote by G i (t) (for Go) and N i (t) (for NoGo), respectively. The OpAL model proposes that these tendencies are encoded in the strengths of synaptic connections between the cortical neurons associated with that state and the striatal D1 or D2 neurons selective for action i, respectively [15], as illustrated in Fig 1. In the OpAL model, after selecting action i the synaptic weights are modified according to: G i (t + 1) = G i (t) + α G i (t) r (t) - V (t) (4) N i (t + 1) = N i (t) - α N i (t) r (t) - V (t) (5) Thus if the reward prediction error is positive, the tendency to select the action is increased, while the tendency to inhibit it is weakened, and vice versa. Additionally, in the OpAL model, the reward prediction error is scaled by G i (t) and N i (t), which prevents G i (t) and N i (t) from becoming negative. For example, if G i (t) becomes close to 0, the changes in its value also tend to 0. The OpAL model additionally proposes how the probabilities of actions depend on the weights in Go and NoGo pathways, through a generalized version of the softmax rule [26,27]: P i (t) = exp a G i (t) - b N i (t) ∑ k exp a G k (t) - b N k (t) (6) In the above equation, normalization by the denominator ensures that the P i (t) add up to 1 across all possible actions. Parameters a and b control how deterministic the choice is: when a = b = 0, all actions have equal probability, while with higher a and b, the influence of the learned tendencies on choice increases. The relative value of parameters a and b describes to what extent the neurons in the Go and NoGo pathways contribute to choice (when a = b, both pathways contribute equally; otherwise, one pathway dominates). The rationale for introducing two parameters a and b is that the activity levels of the striatal D1 and D2 neurons are modulated in opposite directions by levels of DA; hence, they can differentially contribute to activity in the output nuclei [15] (see Fig 1). In the models including only the actor, learning about the reward distribution of an individual action is independent of learning about the distribution of another. Thus for simplicity of notation, while introducing the model we will consider just a single context and a single action, and denote the corresponding synaptic weights of D1 and D2 neurons on trial t by G (t) and N (t), respectively. Furthermore, we will denote the mean and standard deviation of reward distribution by μr and σr. The model employing the original Rescorla-Wagner rule (Eq 1) keeps track of an abstract variable Q (t) that describes the overall tendency to select action i, but in BG this tendency is encoded in the synaptic weights of D1 and D2 neurons, G (t) and N (t). So let us relate these variables by: Q (t) = G (t) - N (t) (7) The update rules for the weights in the Actor learning Uncertainty (AU) model have the following form: G (t + 1) = G (t) + α r (t) - Q (t) + - β G (t) (8) N (t + 1) = N (t) + α r (t) - Q (t) - - β N (t) (9) In the equations above, the prediction errors are transformed through threshold-linear functions |x|+ and |x|− which are equal to |x| if x is positive or negative respectively, and 0 otherwise. In other words, |x|+ = max (x, 0), and |x|− = max (−x, 0). Thus if the prediction error is positive, then so is the corresponding term in Eq (8), and G increases, while if the prediction error is negative, then the corresponding term in Eq (9) is positive, and N increases. Furthermore, the decay terms (last terms in Eqs (8) and (9) ) are scaled by a separate constant 0 < β < 1. As we will explain below, the AU model encodes the estimate of mean reward μr in G (t) − N (t), while the estimate of reward spread σr in G (t) + N (t). Before giving a proof for this property, let us first provide an intuition. The AU model encodes the mean reward in G (t) − N (t) due to its similarity with the Rescorla-Wagner rule. In particular, when the reward is higher than expected, G tends to increase, while when the reward is lower than expected, N tends to increase, so in both cases G (t) − N (t) tends to move towards the value of the reward. To gain some intuition for how the model can encode reward uncertainty in G (t) + N (t), it is useful to consider the changes in the weights in two different cases: when the rewards are deterministic, i. e. , of the same magnitude each time the action is selected, and when they are stochastic. In the case of deterministic rewards, on initial trials, reward prediction error will be positive, hence only G will increase but not N, as illustrated in the top left panel of Fig 2. By contrast, in the case of stochastic rewards, on some trials the reward prediction error will be negative. Hence, N will also increase, as illustrated in the top right panel of Fig 2. Finally, the decay terms in the above equations serve to ensure the convergence of the synaptic weights, as in their absence, the update rules would only allow G and N to either increase or stay the same upon every iteration, but never decrease. Let us now show that the AU model can learn expected reward. By subtracting Eq (9) from Eq (8) we obtain: Q (t + 1) = Q (t) + α r (t) - Q (t) - β Q (t) (10) The threshold-linear functions disappear when Eqs (8) and (9) are subtracted, because if the prediction error is positive, the corresponding terms in Eqs (8) and (9) are equal to the prediction error and 0 respectively, so when subtracted give the prediction error. Conversely, if the prediction error is negative, the corresponding terms in Eqs (8) and (9) are equal to 0 and the negative of the prediction error, so when subtracted they also give the prediction error. Comparing Eqs (10) and (1), we note that this update rule is similar to the standard Rescorla-Wagner rule, with an added decay term. For a fixed value of α, the variable Q never converges when σr > 0, but constantly fluctuates. Nevertheless, it is useful to consider a value around which it fluctuates. After sufficiently long learning, the expected change in Q will be zero. In other words, for large enough t, E Q (t + 1) - Q (t) = 0 (11) The value of Q (t) at which Eq (11) holds is referred to as the stochastic fixed point, and we will denote it by Q i *. By combining Eq (10) with Eq (11), we obtain: E α r - Q * - β Q * = 0 (12) Rearranging the terms in the above equation, we see that Q at the stochastic fixed point is equal to: Q * = α α + β E [ r ] (13) Although in the AU model Q* is not equal to the expected reward, it is proportional to it, with a proportionality constant that is equal across all actions. Thus, choosing an action with the highest Q* is equivalent to choosing an action with the highest expected reward. We now show that the AU model learns reward uncertainty. In order to do so, we will analyze how the sum of the synaptic weights evolves. Thus, let us define: S (t) = G (t) + N (t) (14) By adding Eq (9) to Eq (8): S (t + 1) = S (t) + α r (t) - Q (t) - β S (t) (15) From the above equation we see that at the stochastic fixed point: S*=αβE [| r−Q*|]=αβE [| r−αα+βμr |] (16) The above equation implies that when Q* = μr, the sum of G and N is equal to the deviation of the reward from the mean. In S1 Text we illustrate that, when Q* = μr, then S* is directly proportional to the standard deviation or variance of rewards (depending on the shape of the reward distribution). When Q* ≠ μr, S* is not exactly proportional to the deviation of the rewards from the mean. To see more clearly when it approximates the deviation, let us rewrite the above equation as: S * = α β E (r - μ r) + 1 α β + 1 μ r (17) From the equation above, we see that S* becomes proportional to the deviation of rewards when the second term inside the expected value is dominated by the first. This can occur in two cases. First, since the magnitude of the first term increases with σr, while that of the second term is proportional to μr, then S is close to an estimate of the deviation of rewards when σr is relatively high with respect to μr. Fig 2 shows simulations of the model for different reward mean and standard deviations of rewards and illustrates changes in synaptic weights as learning progresses. The simulations shown in different rows correspond to mean reward being positive, equal to 0, and negative, respectively. Note that the difference between G and N always approaches a value proportional to the expected reward. The simulations shown in different columns correspond to progressively higher standard deviation of reward. When μr = 0, the value that G and N approach increases linearly with σr. By contrast, when μr is higher, the encoding of reward uncertainty is less precise. For example, in the top row of Fig 2 we observe that the values of synaptic weights change very little as σr increases from 0 to 2. The increase in weights is slightly higher as σr increases from 2 to 4. Nevertheless, Fig 2 shows that increasing reward uncertainty still results in higher values of both G and N. Note that in each row, the larger the reward uncertainty, the larger G and N. Second, the second term in Eq (17) decreases with the ratio of parameters β α. Thus the lower β is relative to α, the closer S* is to a linear function of the deviation of rewards. This property is illustrated in Fig 3, which plots S as a function of the standard deviation of rewards for different values of β. It is evident in the figure that, on average, S is a monotonic function of σr. Hence, it is worth noting that although S is an estimate of reward uncertainty, it is possible for the neural system to obtain a closer estimate by learning the function mentioned above and thus decode the estimate of reward deviation from S (i. e. , correct the biases of S in estimating σr). However, this function has a flat region for low σr, so that the model’s estimate of the reward deviation will not be precise in that range of σr. For example, one can observe in Fig 3 that when β = α the value of S ≈ 0. 5 arises for a wide range of σr, so knowing that S = 0. 5 we cannot accurately tell the value of σr. The size of the region where σr is not well estimated can be reduced by decreasing β relative to α. Nevertheless, Fig 3 illustrates that there is a trade-off: Lower β α results in a higher magnitude of weights, and thus higher metabolic cost, and lower β also slows learning (see [28] for details). Let us now consider how the mean and spread of a reward distribution, learned by the model described above, can be used by BG in action selection. In the model the tendency to choose or avoid risky options is controlled by the tonic level of DA. Before giving mathematical justification for this property, let us first provide an intuition for it. Fig 4 illustrates states of a network choosing between two options, one safe and the other risky, represented by neurons shown in blue and orange, respectively. In the figure, the strength of cortico-striatal connections is denoted by the thickness of the arrows. Thus both options are associated with positive mean reward (as the connections Gi are thicker than Ni), but the orange option has higher estimated spread of rewards (as the orange connections are thicker than the blue ones). DA is known to activate the D1 or Go neurons and inhibit D2 or NoGo neurons, which is represented in Fig 4 by green arrows and lines ending with circles. The top panel illustrates a situation when the tonic DA level is high. In this case the NoGo neurons are suppressed (indicated by bleak color) and the choice is driven by the activity of the Go neurons. Thus with high DA, the more risky, orange option is more likely to be chosen, as G2 > G1. By contrast, with low levels of DA, the Go neurons are inhibited (bottom panel of Fig 4), and the choice is driven by NoGo neurons. Thus with low DA, the risky option is inhibited (as N2 > N1), and the model is more likely to select the safe option. The above example illustrates that the model has the tendency to choose more risky options when the level of DA is high, and safer options otherwise. Let us now show this property formally. The choice rule of Eq (6) can be rewritten to make the effect of the mean and deviation of reward visible. To do so, we first write Gi and Ni in terms of Qi and Si (defined in Eqs (7) and (14) ): G i (t) = 1 2 S i (t) + Q i (t) (18) N i (t) = 1 2 S i (t) - Q i (t) (19) Substituting the above into Eq (6) we obtain: P i (t) = exp 1 2 U i ∑ k exp 1 2 U k, where U i = (a + b) Q i (t) - (b - a) S i (t) (20) In the choice rule above, the probability of choice depends on a utility function Ui that is a linear combination of mean reward and the deviation of reward (cf. [29,30]). By increasing b relative to a in the above choice rule, one can explicitly control how choice probability is affected by the deviation of rewards. In particular, when b > a, the uncertainty of rewards reduces the probability of selecting the corresponding action, resulting in risk aversion. By contrast, setting b < a increases the probability of choosing actions with uncertain rewards, resulting in risk seeking. Recall that parameters a and b describe in the OpAL model [15] to what extent D1 and D2 neurons contribute to determining choice. Since high levels of DA activate the direct pathway and suppress the indirect pathway, increasing the tonic level of DA will correspond in the model to increasing a and decreasing b, which according to the analysis above would result in more risk-seeking behavior. Thus such modulation provides a mean by which an organism can control whether the action selection should be risk-averse or risk-seeking. The above analysis explains why a tendency for gambling in Parkinson’s patients [12,31] may arise from increasing the level of DA by medications or from deep brain stimulation of subthalamic nucleus (which would also weaken the indirect pathway so would correspond to lowering b). The presented model accounts for the effect of pharmacological manipulations affecting dopaminergic receptors on risk aversion in reinforcement learning tasks. In a particularly comprehensive study [32], rats were trained to choose between 2 levers: pressing one of them resulted in certain delivery of a single food pellet, while pressing another could result either in delivery of 4 pellets or none. The probability of receiving the large reward after the selection of the risky lever was varied across conditions. After the rats were well-trained in the task, they were injected with different drugs, and changes in the fraction of risky choices made were measured. An overall increased tendency to choose the risky option was observed either after injection of D1 agonist or D2 agonist, as shown in Fig 5. Furthermore, the injection of D1 antagonist or D2 antagonist decreased the tendency to choose the more risky option [32]. The fraction of risky choices made in simulations by the AU model is shown by curves in Fig 5. In the simulations, the parameters controlling learning were fixed to standard values (α = β = 0. 1), and only the parameters controlling choice (a and b) were fit to the data. Parameters a and b were fit separately to the data in each panel of Fig 5, as each panel was obtained from a different group of rats. While fitting the model to the data from D1 receptor manipulations, it was assumed that a differed between control and drug conditions, while b did not change. Thus three parameters were fit: acontrol, adrug, and b. We did not enforce any relationship between acontrol and adrug, but as we will explain below, the estimated parameters followed the relationship expected from the known effects of drugs. Analogously, while fitting the model to the data from D2 receptor manipulations, a, bcontrol, and bdrug were fit. For each panel, the values of the three parameters were found that minimized the sum of squared errors between the fraction of risky choices made by the animals and the model in the 8 conditions (4 probabilities of large rewards on and off the drug). The parameters were found using the simplex algorithm [33] implemented in Matlab (function fminsearch). The search was repeated 10 times with different random initial parameter values sampled from the range [0,3]. The model reproduced the fractions of risky choices made by the animals relatively well. Importantly, the overall direction of changes in risky choices and estimated parameters is consistent with the pattern in the data. In particular, in the top panels of Fig 5, the fraction of risky choices is higher in the simulation of the agonist conditions. Furthermore, in the top left panel, estimated parameters satisfied adrug > acontrol (acontrol = 1. 71, adrug = 3. 13, b = 0. 59), which is consistent with the excitatory effect of DA on D1 receptors, while in the top right panel, the estimated parameters satisfied bdrug < bcontrol (a = 2. 72, bcontrol = 1. 86, bdrug = 0. 39), consistent with the inhibitory effect of DA on D2 receptors. Thus the choice behavior may become more risky due to activation of either D1 or D2 receptors, as activation of either of them decreases b − a, which reduces risk aversion in Eq 20. Analogously in the bottom panels of Fig 5, the fraction of risky choices is lower in the simulated condition with antagonists, and estimated parameters satisfy adrug < acontrol for the bottom left (acontrol = 2. 67, adrug = 0. 86, b = 1. 04) and bdrug > bcontrol for the bottom right panels (a = 1. 95, bcontrol = 0. 04, bdrug = 2. 16). It is worth noting in the bottom left panel of Fig 5 that the model reproduces the cross-over of the two curves. It occurs in the simulations because as a is reduced (corresponding to the effect of D1 antagonist), the choice in the model becomes more random (recall from the Models section that a and b also control how deterministic the choice is), so that the fraction of risky choices is closer to 50%. In this task, choosing the risky lever gave higher expected reward in the 100% and 50% conditions while choosing the safe lever had higher mean reward in the 12. 5% condition, and the model simulated with higher a in the bottom left panel of Fig 5 exploited the options with higher expected rewards more. The AU model assumes particular rules for updating striatal synaptic weights, and here we consider whether these rules are consistent with the existing data concerning synaptic plasticity in the striatum. For a synaptic plasticity rule to be plausible, the change in a synaptic weight needs to depend only on the information that can be sensed by a synapse, i. e. , the activity of pre-synaptic and post-synaptic neurons, the levels of neuromodulators released in the vicinity of the synapse, and the synaptic weight itself. Eqs (8) and (9) describe the change in synaptic weights between the neurons encoding current context and those encoding current movement, i. e. , they describe changes in synapses between co-active neurons. This change includes two terms, which are the reward prediction error and decay. As mentioned earlier, a plethora of evidence suggests that reward prediction error (r (t) − Q (t) ) is encoded in phasic changes in DA concentration, which is released in striatum. The proposed weight update rules are consistent with the pattern of synaptic plasticity modulation by DA [34]. It has been observed experimentally that the activation of cortical neurons followed by striatal D1 neurons strengthens the synapses of D1 neurons when the DA level is elevated, and weakens these synapses when the DA level is reduced (Figs 3F and 2E in [34]). Such changes are consistent with Eq (8), because for positive prediction error, the prediction error term will dominate, so G will increase. By contrast, if the prediction error is negative, |r (t) − Q (t) |+ will be equal to 0, and the decay term will dominate, so G will decrease. Conversely, the activation of cortical neurons followed by striatal D2 neurons weakens the synapses of D2 neurons when the DA level is elevated, and strengthens the synapses of D2 neurons when DA level is reduced (Figs 1H and 3B in [34]). Such changes are consistent with Eq (9) for analogous reasons. A critical property of the learning rules allowing encoding reward uncertainty in G (t) + N (t) is the asymmetry in how synaptic weights change for positive and negative reward prediction error. In particular, in the AU model, the change in G is only proportional to the reward prediction error if the error is positive, but not if the error is negative (analogous asymmetry holds for N). It is easy to check that if such asymmetry were not present (i. e. , nonlinear functions of predictions errors were removed from Eqs (8) and (9) ), then G (t) + N (t) would no longer encode the spread of reward distribution. Such asymmetry may arise in striatal synapses from the observed differences in the affinity of DA receptors, such that a higher DA concentration is necessary to activate D1 receptors than D2 receptors [35]. Fig 6 shows how the probability of D1 and D2 receptor activation depends on DA concentration in a biophysically realistic model of DA release [36]. Simulation of that model based on activity of DA neurons in vivo [37] suggested that the baseline DA level in striatum is in a sensitive range of both D1 and D2 receptors (as illustrated by the dashed line in Fig 6). Due to the arrangement shown in Fig 6, an increase in DA level has a larger effect on the activation of D1, while a decrease in DA has a larger effect on D2 receptors. According to Fig 6, the decrease in DA level may still have some small effect on the binding probability of D1 receptors (analogously the increase in DA may have a small effect on D2 receptors). Hence the complete lack of effect of a decrease (increase) in DA level on D1 (D2) neurons’ plasticity may seem inconsistent with the above analysis. Nevertheless below we show that for learning reward uncertainty, it is sufficient that there exist an asymmetry in the dopaminergic effects on the receptors, i. e. , that the increase in DA level affect plasticity of D1 neurons more than D2 neurons (and the opposite for a decrease in DA level). The Equations describing the AU model can be generalized to include more complex functions of reward prediction error: G (t + 1) = G (t) + α r (t) - Q (t) + - ϵ r (t) - Q (t) - - β G (t) (21) N (t + 1) = N (t) + α r (t) - Q (t) - - ϵ r (t) - Q (t) + - β N (t) (22) where ϵ is a constant such that ϵ < 1. As synaptic weights cannot be negative, whenever G (t+1) or N (t+1) computed from the above equations is negative, it is set to 0. A potential advantage of using such functions of prediction error is that after each feedback iteration, they drive changes in both G and N, and thus potentially result in faster learning. When ϵ = 0, the above model reduces to the AU model. We now show that with these functions, the model can still encode expected reward and reward uncertainty. Subtracting the above two equations gives: Q (t + 1) = Q (t) + α (1 + ϵ) r (t) - Q (t) - β Q (t) (23) Hence at the stochastic fixed point: Q * = α (1 + ϵ) α (1 + ϵ) + β E [ r ] (24) Thus the differences in the synaptic weights of D1 and D2 neurons encode scaled relative values of actions, which are also sufficient to choose the action with the highest value. Similarly adding Eqs (21) and (22) we obtain: S (t + 1) = S (t) + α (1 - ϵ) r (t) - Q (t) - β S (t) (25) Hence at the stochastic fixed point: S * = α (1 - ϵ) β E r - Q * (26) Using the analysis applied earlier to the AU model, we see that the sum of the weights of D1 and D2 neurons encodes a scaled version of deviation of the reward, under analogous conditions to those for the AU model (i. e. , σr is relatively high with respect to μr, or β is relatively small with respect to α (1 + ϵ) ). However, when ϵ > 0, the weights G (t+1) or N (t+1) computed from Eqs (21) and (22) may become negative, but negative synaptic weights are not allowed in the model, so the calculations of the fixed points above are only valid for ϵ sufficiently small so that G (t+1) and N (t+1) are not negative. To illustrate how this generalized AU model encodes reward uncertainty, the left panel in Fig 7 shows the results of simulations in the same setting as in Fig 3, but with a fixed value of β = 0. 1, for different values of parameter ϵ. The figure shows that when ϵ = 0. 5, the model also encodes reward uncertainty, but the encoding is less accurate than for ϵ = 0. In particular, when S is equal to a certain value, we can infer σr more precisely from the left panel in Fig 7 for ϵ = 0, as the range of σr resulting in the certain value of S is narrower for ϵ = 0 (e. g. , S = 0. 75 for σ ∈ [0. 6,1]) than for ϵ = 0. 5 (e. g. , S = 0. 75 for σ ∈ [1,2]). In this section we show that the actor-critic model after small extension can learn both the mean and spread of rewards associated with actions. The model uses the same rule for the update of the critic (Eq (2) ), and the plasticity of synapses of D1 and D2 neurons is described by equations similar to those for the AU model, but in which the prediction error is based on the reward estimated by the critic: G i (t + 1) = G i (t) + α r (t) - V (t) + - α G i (t) (27) N i (t + 1) = N i (t) + α r (t) - V (t) - - α N i (t) (28) For simplicity, in the above equations we set the decay constant β = α, which will also allow relating the model to advantage learning [39,40]. We will refer to a model with the actor described by the above equations, with the critic by the standard Rescorla-Wagner rule of Eq (2), and with the OpAL choice rule of Eq (6), as the Actor-Critic learning Uncertainty (ACU). We now show that the ACU model estimates both mean and spread of rewards associated with action i, which we denote by μi and σi, respectively. To see that the mean rewards are encoded in the difference between Gi and Ni, we subtract the above equations, and using Eq (7), we obtain: Q i (t + 1) = Q i (t) + α r (t) - V (t) - α Q i (t) (29) This update rule differs from that of the original actor-critic model of Eq (3) in that it includes a decay term, and the rule is known as advantage learning [39,40] (for reasons that will become apparent below). Let us now find the value the vicinity of which Qi approaches, by noting that at the stochastic fixed point the following condition must hold: E α r - V * - α Q i * = 0 (30) Rearranging the terms in the above equation, we see: Q i * = μ i - V * (31) Namely, Qi at the stochastic fixed point is equal to the expected reward for action i relative to the overall average reward in the current state (this quantity has been termed the advantage of action i). Note that knowing the relative values of the actions available in a given state is sufficient for selecting the action with the highest value. The value of the state V* is equal to the average value of all actions weighted by how frequently they are selected: V * = ∑ i P i * μ i (32) In this model, the sum of G i (t) and N i (t) also approximates reward uncertainty. Adding Eqs (27) and (28) we obtain: S i (t + 1) = S i (t) + α r (t) - V (t) - α S i (t) (33) At the stochastic fixed point, the expected change in the sum of weights should be equal to 0, hence: E α r - V * - α S i * = 0 (34) Rearranging terms, we see that the sum of weights Gi and Ni at the fixed point is: S i * = E r - V * (35) The above equation implies that when V* = μi, the sum of Gi and Ni is equal to the deviation of the reward from the mean. We now consider three situations when V* is close to μi. First, when only one action is available, and chosen on all trials, then V* = μ1, and hence S 1 * ∼ σ 1. This property is illustrated in the right panel of Fig 7, where black dots show the uncertainty estimated by the ACU model in simulations with a single action. Note that S is proportional to reward uncertainty for the entire range of σr, so with a single action, the ACU model can accurately encode uncertainty for a wider range of σr than the AU model (cf. black points in left and right panels of Fig 7). Second, when a few actions i ∈ I have similar mean rewards, while other actions j ∈ J give much lower rewards, then Pj ∈ J are close to 0. In this case, V* is equal to a weighted average of μi ∈ I, but since we assumed that all μi ∈ I are similar, then V* is close to μi for i ∈ I. Hence the ACU model estimates well the spread of reward distribution for actions with the highest mean reward, i. e. , those most frequently selected. It may not estimate the spread of other actions, but this does not matter, as these actions are typically not selected anyway. Different rows in Fig 8 show simulations of the ACU model for different reward distributions and illustrate changes in synaptic weights as learning progresses. In the first simulation, the two actions have the same mean reward, and it can be seen in the top row that the value V converges to the expected reward. For each action, Gi and Ni converge to values equal to each other, because the ACU model encodes in Gi − Ni the relative value of actions which are equal to 0 here. In the simulation, the second action has uncertainty twice as high as the first one, and indeed one can see in the top row of Fig 8 that G2 + N2 converges to a value twice as high as G1 + N1. In the simulation illustrated in the bottom row of Fig 8, the first action has a smaller expected reward. The model learns to select the second action on a great majority of trials, which results in the expected reward V converging towards the mean reward of the second action. The model estimates well the deviation of rewards associated with the second action—note that G2 + N2 is similar in both rows of Fig 8. Finally, the model does not estimate well the deviation of reward of the first action, but this does not matter, as this action is very rarely selected. Third, the ACU model can still estimate reward uncertainty for actions with lower mean rewards than other actions available, if the uncertainty is sufficiently large. To understand this property, it is helpful to rewrite Eq 35 as: S i * = E (r - μ i) + (μ i - V *) (36) When σi is sufficiently larger than |μi − V*|, the first term in the above equation will dominate over the second, and Si will be more closely proportional to σi. In summary, the AU and ACU models differ in the conditions under which their ability to estimate reward uncertainty is limited. The AU model does not precisely estimate the reward uncertainty in situations where the standard deviation of rewards is small relative to their mean. The ACU model has a limited ability to estimate uncertainty only in a subset of these situations, i. e. , when the reward uncertainty is small and additionally the mean value of the action is substantially lower than for other actions available in the corresponding state. Finally, it is worth mentioning that the learning rule of the ACU model can be generalized as described in the previous subsection, such that the weights of the actor are modified according to Eqs 21 and 22 but with Q replaced by V. The grey dots in the right panel of Fig 7 show that the uncertainty estimated by such a generalized ACU model is still proportional to the true variability of rewards but is encoded less precisely than in the original ACU model. Furthermore, a simulation of the ACU model analogous to that shown in Fig 5 produced qualitatively similar behavior as the AU model; thus, an increased tendency to take risky options with a high level of DA is a general property of a class of models encoding reward uncertainty in G + N. We also investigated the behavior of the OpAL model [15] in the presence of reward uncertainty. Fig 9 shows simulations of the OpAL model in the same tasks used for the ACU model in Fig 8. Top rows of Fig 9 show simulations of a task in which the two actions have the same mean reward but differ in reward deviation. In the initial trials, in which the reward prediction error is positive, Gi increase exponentially. The exponential increase arises due to the multiplication of prediction error by G or N in Eqs 4 and 5, which results in a rate of weight changes that is proportional to the weights themselves. Once the reward prediction becomes equal to 0 on average, the weights start to decay towards 0. The weights have a stochastic fixed point at Gi = Ni = 0 in the OpAL model, because when Gi = Ni = 0, there are no changes in weights according to Eqs 4 and 5. In the task simulated in the top panel of Fig 9, this fixed point was attractive, and all weights of the actor eventually approached 0. It is interesting that this decay was faster for the option with higher uncertainty, as for this option the larger fluctuations in the reward prediction error drove the weights to the fixed point faster. In the task simulated in the bottom panel of Fig 9, this fixed point was attractive only for the action with the higher value, while for the other action, Ni increased with time. It is evident from Fig 9 that the OpAL model does not encode reward uncertainty in the weights close to convergence, and the dynamics of weight changes is much more volatile than in the ACU model (note that the range of vertical axes in Fig 9 is two orders of magnitude higher than in Fig 8). Furthermore, when two actions have equal mean reward, as in the top panels of Fig 9, after extensive training, all weights Gi and Ni converge to 0, so the probability of choosing a more risky option becomes exactly 0. 5, according to Eq 6, irrespective of the values of parameters a and b. Hence in this case, the probability of a risky choice predicted by the OpAL model is not dependent on the level of DA. The OpAL model is able to capture the effects of dopaminergic medications seen in a series of experiments [14,41,42], which as we will see below, are challenging for the AU and ACU models. These experiments were designed to test the effects of DA on learning from positive and negative feedback, but in these studies the feedback uncertainty also varied between choice options. During these experiments the participants were presented with Japanese characters, were asked to choose one them, and subsequently received feedback indicating whether their choice was correct. For clarity, let us consider a simplified version of the task. Assume that during the training phase, the participant is presented on each trial with 3 letters which we will refer to as A, B and C. The probability of obtaining “Correct” feedback after selecting each of the 3 options is 0. 8,0. 2 and 0. 5 respectively. After the training, the participant is presented with a choice between A and C, or with a choice between B and C. The fraction of A vs. C trials in which the participant chooses A has been interpreted as a measure of learning from positive feedback (as stimulus A was associated with the highest probability of “Correct” feedback). Conversely, the fraction of B vs. C trials in which the participant does not choose B has been interpreted as a measure of learning from negative feedback (as stimulus B was associated with the highest probability of “Incorrect” feedback). It has been observed that Parkinson’s patients on dopaminergic medications exhibit higher accuracy in choosing A than in avoiding B, while the opposite pattern is present off medications [14]. Furthermore, it has been suggested that this effect is dependent on the medication state during testing rather than during encoding [42]. The OpAL model is able to replicate these effects [15]. While simulating learning in this task, we assumed that the model receives a reward of r = 1 when “Correct” feedback is given, and no reward r = 0 after “Incorrect” feedback. The top left panel in Fig 10 shows the weights learned by the OpAL model. As expected, Gi increase with the probability of reward, while Ni decrease. Importantly, the relationship between weights and reward probability is non-linear. This non-linearity arises from the multiplication of prediction error by Gi or Ni in Eqs 4 and 5, which as mentioned above, results in an exponential growth of the weights and thus magnification of weights with high values. The bottom right panel in Fig 10 illustrates how the values of the weights affect behavior during test. In the simulated on medication condition, the choice is primarily affected by weights Gi (Eq 6). Thus the accuracies in choosing A and avoiding B depend on |GA − GC| and |GB − GC|, respectively. Since |GA − GC|> |GB − GC| in the top left panel, the probability of choosing A is higher than the probability of avoiding B on medications in the bottom left panel. In the simulated off medication condition, the choice is primarily affected by weights Ni, and hence the model is better at avoiding B than choosing A for analogous reasons. The choice pattern in the bottom left panel of Fig 10 is qualitatively consistent with that observed in experimental studies [14,41,42]. The top panel in the middle column of Fig 10 shows the weights learned by the AU model. Here also, Gi increase with reward probability, while Ni decrease. However, in the AU model the sum of weights Gi + Ni is highest for option C, which gives reward on 50% of trials and thus has highest reward variance. Consequently, the relationships between weights and reward probability are concave for the AU model, rather than convex as they were for the OpAL model. This results in the opposite effect of DA on choosing A and avoiding B relative to the OpAL model (cf. left and middle panels in the bottom row of Fig 10). The right panels of Fig 10 illustrate that the behavior of the ACU model is qualitatively similar to that of the AU model. However, the predicted effect of medications on choice probability in ACU is smaller than in AU, because the relationships between weights and reward probability are more linear for ACU. This occurs because ACU estimates the deviation of reward from the mean across all trials (Eq 35) rather than from the mean reward for a given option, as in AU. The OpAL model also described the dependence of learning rates α for Gi and Ni on the level of DA [15]. Simulations of the AU and ACU models indicate that increasing the learning rate for Gi (or Ni) scales up the learned values of Gi (or Ni) but does not change the convexity/concavity of the relationship between weights and reward probability, and hence does not change qualitatively the predicted effects of DA during testing on the probability of choosing A and avoiding B. In summary, the simulations of the AU and ACU models produced qualitatively different patterns of effects of dopaminergic medications on choosing A and avoiding B than observed experimentally [14,41,42]. A critical feature of the OpAL model that allows it to capture the experimentally observed effects is the multiplication of prediction error by G or N in Eqs 4 and 5, but it is this very property that also caused unrealistically volatile weight changes in simulations of Fig 9. It is interesting to ask under what assumptions the pattern of weights in the top left panel of Fig 10 (that allows reproducing the effects of medications on choosing A and avoiding B) could be obtained in a model learning reward uncertainty. In our simulations we assumed that “Correct” and “Incorrect” feedback were mapped on rewards of 1 and 0. However, it is unclear if the brain simply maps abstract feedback on the reward. It is possible that instead the brain infers that option C is unpredictable and does not engage in learning about it, which would result in relatively low GC and NC, as in the top left panel of Fig 10. This interpretation together with the AU (or ACU) model predicts that if an actual (e. g. , monetary) reward is given as feedback, the effect of dopaminergic medications on choosing A and avoiding B should reverse (or be very small). This interpretation is consistent with a result of experiments employing a modified version of the Japanese letter task with more salient feedback, i. e. , smiling and sad faces, in which no effect of medications was found [43]. However, to fully test this interpretation, further studies are needed that could for example use explicit monetary reward. We discuss here the relationships between predictions of the models and experimental data, including behavior and neural activity. Since in this paper we presented several models, it will be important to distinguish in the future which of them provide the best description of learning uncertainty in the basal ganglia. To differentiate between predictions specific to individual models and common to other models, we will use the term “the models” to refer to a class including all models introduced in this paper. We already demonstrated in the Results section that the models account for the effect of pharmacological manipulations affecting dopaminergic receptors on risk aversion in reinforcement learning tasks in rats. The studies investigating the effect of DA on human decisions involving risks use two types of paradigms: one in which the mean and spread of rewards associated with choice options are explicitly described to the participant before each decision, and one in which they are gradually learned from feedback. Since human behavior is very different in these paradigms [44], and the models assume that the mean and deviation of rewards are learned in cortico-striatal synapses, below we only focus on studies involving learning from experience. The most commonly used paradigm in such tasks is the Iowa gambling task in which participants choose between decks of cards differing in reward variance. In agreement with the models, Parkinson’s patients receiving dopaminergic medications choose the risky decks more frequently than healthy controls, but this effect is not present in patients that have not been put on medications yet [45], or who stopped receiving medications [46]. The models introduced in this paper do not describe behavior in decision tasks in which information about risks associated with different options is explicitly presented before each trial. It is likely that processing information about uncertainty in such tasks involves different neural mechanisms and circuits than those learning about reward uncertainty over many trials. The models are also consistent with the results of a recent study showing that optogenetic activation of striatal D2 neurons decreases the probability of choosing options with high reward variance [47]. Optogenetic activation of D2 neurons corresponds to a scenario illustrated in the bottom panel of Fig 4, where the choice is primarily driven by D2 neurons, and thus the risky option is inhibited. The AU and ACU models differ in the predicted activity of DA neurons when the reward exactly matches the expected reward in tasks where only one action is available. In the ACU model, DA response is assumed to carry (r − V) where V* = E[r], so when r = E[r], DA neurons should not change their firing rate. By contrast, in the AU model the DA release is assumed to encode (r − Q) where Q* < E[r] (see Eq 13), so when r = E[r], DA neurons should increase their firing rate above baseline. Experimentally observed DA responses after expected rewards differed between experimental studies. For example, DA neurons were found to maintain their activity in classical conditioning in some studies [6,7], while an increase was observed in others [48,49]. Thus, more research is necessary to establish factors determining DA response to expected reward. The AU model predicts that learned synaptic weights in BG are insensitive to small standard deviations of reward; thus, it predicts that an individual’s choices are not affected by small enough uncertainty in reinforcement learning tasks. By contrast, the ACU model predicts that biases in estimation of reward uncertainty should only be present for actions with mean rewards much lower than those of other actions. The models predict that overall activity in striatum should be higher during choice between options with high reward variance than during choice between options with lower reward variance but similar mean, because in the models the spread of rewards is encoded in Gi + Ni, so higher reward variance should increase the activity of both D1 and D2 striatal neurons. This prediction could be easily tested using functional MRI. The models predict that synaptic plasticity will depend on the current value of the weight itself (i. e. , Gi or Ni), because the weight update rules include decay terms proportional to the weights themselves. Thus the models predict that the stronger the weight of a synaptic connection, the higher the amplitude of induced long-term depression. Such dependence of plasticity on the value of weights has been observed in neocortex [50], and it would be very interesting to see if it is also present in cortico-striatal synapses. In addition to the models presented in this paper, reward uncertainty can be learned by a wide family of models in which the decay terms are proportional to the estimated uncertainty, and these models were analyzed in [28]. The models in this family can learn reward uncertainty even for small deviations. However, to implement such learning rules, the information about the uncertainty would need to be provided to a synapse, e. g. , by a second neuromodulator. The models in this family predict that the release of this neuromodulator would need to be dependent on uncertainty and promote long-term depression of cortico-striatal synapses. Three different neuromodulators have been proposed to encode information about estimated (or expected) reward uncertainty: tonic DA [51], acetylcholine [52], and serotonin [30]. All of these neuromodulators have been shown to affect risky decisions [12,53–55]. However, we have not found support in existing experimental data for predictions of our models employing multiple neuromodulators, hence we did not include them in this paper. In previous reinforcement learning models that described learning about uncertainty [30,56], the estimate of reward variance was updated on each trial proportionally to “variance prediction error”, which is equal to the difference between the square of reward prediction error and the current estimate of variance. An interesting model describing how such learning could be implemented in BG [57] suggested that the variance of rewards is encoded in striatal neurons co-expressing D1 and D2 receptors. This model assumed that such neurons could increase their weights both when the prediction error is highly positive (like D1 neurons) and when it is strongly negative (like D2 neurons). However, the neurons co-expressing D1 and D2 receptors form only a small proportion of striatal neurons [58], and the models we propose describe learning of reward deviation in the great majority of striatal projection neurons that express either D1 or D2 receptors. An interesting reinforcement learning model has also been proposed in which choosing risky options can be avoided without explicitly learning the spread of reward distributions for different options [59]. In this model, the weight update rules are modified such that Qi is decreased when action i leads to a reward with higher variance. This model is efficient when the desired level of risk aversion is known and fixed before the learning starts, but unlike the models presented in this paper, it does not allow the trained system to be easily switched from risk aversion to more neutral or risk seeking behavior. Reward uncertainty is also likely to be estimated in the cortex. A particularly interesting model [60] describes how the variance of any feature of the stimulus (including reward) can be estimated in a neural circuit with organization similar to that of the neocortex, and it has been shown how this learning about variance can be implemented with local Hebbian plasticity [61]. It is highly likely that the mechanisms of learning uncertainty in neocortex and striatum can operate in parallel. Furthermore, these two structures may estimate complementary measures of dispersion: the cortical model [60,61] estimates variance, while the models presented here estimate the mean absolute deviation (which is less affected by outliers). In this paper we focused on one particular type of uncertainty associated with variable rewards in a stationary environment, which is typically called “expected uncertainty” [52]. But there is also another type of uncertainty connected with rapid changes (or volatility) of mean reward, referred to as “unexpected uncertainty” [52]. It is likely that there are complementary neural mechanisms which estimate unexpected uncertainty. For example, it has been proposed that striatal cholinergic tonically active interneurons detect changes in reward contingency and increase the learning rate following such changes [62]. Areas beyond BG can also be involved in this process, as the activity in other brain regions has been shown to track reward volatility [63] and volatility prediction errors [64]. Finally, let us discuss the relationship of the ACU model to advantage learning [39,40]. As mentioned in the Results section, the ACU model estimates the mean reward using the advantage learning rule; thus, the ACU model also provides a description of how this abstract rule may be implemented in the BG circuit. The advantage model was originally introduced to reconcile reinforcement learning models with animals’ innate tendency to approach highly rewarding stimuli [39,40]. The central feature of the advantage model (also inherited by the ACU model) is that as learning progresses, the value V represented by the critic approaches the value of the best action in the current state, while the advantage Qi of this action approaches 0. This property describes a transition from an instrumental action selection to a stimulus-response habit, as in the trained state the action selection is implemented in the advantage model by the innate tendency to approach high value states [39,40]. The ACU model has an analogous property that in the absence of reward uncertainty, Gi decreases towards 0 as learning progresses. Selection under such circumstances is primarily driven in the ACU model by D2 neurons, as suboptimal actions have large Ni, and thus are inhibited. This agrees with a recent proposal of D2 neurons being critical for choosing among actions [65]. It would be possible to also include in the ACU model the tendency to approach high value states, by including additional terms in the softmax choice rule, as in [66]. In the advantage and ACU models, the actor encodes the mean reward relative to the overall reward in the current state (Eq (31) ). So although the actor has the information necessary to choose which action is best in the current context, it does not know whether it is worth selecting it at all (e. g. , whether any μi > 0). The information on whether it is worth making a movement in the given state (i. e. , on the average value of actions chosen by the actor) is encoded in the critic. Thus the models suggest that patch neurons, which the critic has been mapped onto, should also be involved in movement initiation. It is intriguing that patch neurons project to the dopaminergic neurons [19], so one could ask whether they may communicate the information on the value of making a movement via dopaminergic neurons. This idea is consistent with DA controlling the vigor of movements [67].

To maximize their chances for survival, animals need to base their decisions not only on the average consequences of chosen actions, but also on the variability of the rewards resulting from these actions. For example, when an animal’s food reserves are depleted, it should prefer to forage in an area where food is guaranteed over an area where the amount of food is higher on average but variable, thus avoiding the risk of starvation. To implement such policies, animals need to be able to learn about variability of rewards resulting from taking different actions. This paper proposes how such learning may be implemented in a circuit of subcortical nuclei called the basal ganglia. It also suggests how the information about reward uncertainty can be used during decision making, so that animals can make choices that not only maximize expected rewards but also minimize risks.

Abstract Introduction Models Results Discussion

learning medicine and health sciences neurochemistry decision making nervous system dopaminergics social sciences electrophysiology neuroscience learning and memory simulation and modeling synaptic plasticity cognitive psychology cognition neuronal plasticity research and analysis methods developmental neuroscience animal cells neurochemicals biochemistry cellular neuroscience psychology cell biology anatomy synapses physiology neurons biology and life sciences cellular types cognitive science neurophysiology

2016

Learning Reward Uncertainty in the Basal Ganglia

P10 is a small, abundant baculovirus protein that accumulates to high levels in the very late stages of the infection cycle. It is associated with a number of intracellular structures and implicated in diverse processes from occlusion body maturation to nuclear stability and lysis. However, studies have also shown that it is non-essential for virus replication, at least in cell culture. Here, we describe the use of serial block-face scanning electron microscopy to achieve high-resolution 3D characterisation of P10 structures within Trichoplusia ni TN-368 cells infected with Autographa californica multiple nucleopolyhedrovirus. This has enabled unparalleled visualisation of P10 and determined the independent formation of dynamic perinuclear and nuclear vermiform fibrous structures. Our 3D data confirm the sequence of ultrastructural changes that create a perinuclear cage from thin angular fibrils within the cytoplasm. Over the course of infection in cultured cells, the cage remodels to form a large polarised P10 mass and we suggest that these changes are critical for nuclear lysis to release occlusion bodies. In contrast, nuclear P10 forms a discrete vermiform structure that was observed in close spatial association with both electron dense spacers and occlusion bodies; supporting a previously suggested role for P10 and electron dense spacers in the maturation of occlusion bodies. We also demonstrate that P10 hyper-expression is critical for function. Decreasing levels of p10 expression, achieved by manipulation of promoter length, correlated with reduced P10 production, a lack of formation of P10 structures and a concomitant decrease in nuclear lysis. The Baculoviridae is a diverse family of insect-specific viruses that contain circular, supercoiled dsDNA genomes ranging from 80 to 180 kilobase pairs (kbp) [1,2]. Baculoviruses are characterised by rod shaped virions between 250–400 nm in length and 30–70 nm in diameter [3,4] that are embedded in protective paracrystalline protein structures known as polyhedra or occlusion bodies (OBs) [5]. Protection of virus particles within OBs is primarily restricted to invertebrate-specific viruses and is thought to relate to the need to preserve virus during diapause or when host larval numbers are scarce [6–9]. The nature of baculoviruses has led to their widespread use in agriculture, as a bio-pesticide [6] and as a platform for gene expression and recombinant protein production [10,11]. The expression system has exploited the highly efficient promoters of the p10 and polyhedrin (polh) genes, both of which have been found to be non-essential for the propagation of the virus in cell culture [12–14]. Over the course of a baculovirus infection, whether in larval tissue or in cultured insect cells, cell nuclei become enlarged and, in the very-late stages, become packed with the para-crystalline OBs that are readily visible under the light microscope [15]. Ultimately, nuclei disintegrate releasing OBs or polyhedra into the liquefied larval remains or cell culture medium [16,17]. The major component of OBs is the 30 kDa polyhedrin protein that is produced under control of the very late, hyper-expressed polh promoter [18]. During the final stage of maturation, OBs acquire a polyhedral envelope (PE) /calyx, which largely consists of a polyhedral envelope protein (PEP), PP34 in Autographa californica multiple nucleopolyhedrovirus (AcMNPV) [19–21], and a carbohydrate matrix. The process for PE formation is still the subject of speculation. However, both P10 and electron dense spacers (EDS) associate with or contain pp34 (39–40) and appear to play a key role in PE formation [5,15,22,23]. In particular, it has been shown that in the absence of P10 or pp34, OBs do not fully mature leaving a rough and pitted surface [22]. The abundant P10 protein is produced from the other very-late, hyper-expressed gene found in the baculovirus genome [10]. Early observations of baculovirus-infected cells by transmission electron microscopy (TEM) noted the formation of large fibrous structures in both the nucleus and cytoplasm during the later phases of infection [5,23–25]. Initially it was proposed that these fibrous structures represented the polyhedrin protein in its physical state prior to formation of OBs; largely due to the close association with developing OBs. It was not until later, with the characterisation of polyhedrin and a 10-kDa protein, referred to as P10 [10,11,26–28], that these fibrous structures were identified as separate to OBs [29]. Evidence from TEM and confocal microscopy suggest P10 is associated with a number of intracellular structures with potential roles not only in OB maturation [15,19,22], but also nuclear stability [30] and mediation of nuclear disintegration [17]. Immunofluorescence studies on AcMNPV-infected Trichoplusia ni (TN-368) or Spodoptera frugiperda (Sf9) cells, using anti-tubulin and anti-P10 antibodies, observed that the initial thinner P10 structures co-localised with microtubules [27,28]. These studies suggested that cross linking with microtubules may promote the reorganisation of the cytoskeleton that is observed in baculovirus-infected cells and may be a prerequisite for P10 self-assembly [30,31]. Interestingly, however, the thicker P10 structures found from 48 hpi do not co-localise with microtubules [30]. Instead, confocal microscopy indicated that they form a perinuclear network or ‘cage’ enveloping the infected cell nucleus [28]. To date, the mechanisms underlying the formation and role of these structures is unknown. Most of our knowledge of the structures associated with P10 has come from studies using confocal microscopy [28]. Understanding of the detailed ultrastructure may help elucidate how and why these structures form. To enable visualisation of the three dimensional (3D) ultrastructure, we chose to use serial block-face scanning electron microscopy (SBFSEM) due to the combination of resolution that can be achieved and total cellular volume that can be imaged relatively rapidly. SBFSEM is an emerging technique that is rapidly gaining wider applications in many areas of biology [32,33]. We report here the first use of SBFSEM to study baculovirus structures in infected insect cells. The remarkably detailed images show the formation of two independent and distinct P10 structures, one within the nucleus and one within the cytoplasm that is perinuclear. Additionally, our study has enabled the reconstruction of OBs and EDS and their close association with the nuclear form of P10. The motivation for this study was to elucidate the functional roles of P10 during AcMNPV infection of cultured insect cells and to understand the mechanisms by which it can influence diverse processes. This includes events such as the maturation of OBs and nuclear disintegration, often referred to as nuclear lysis, for the release of mature OBs from the assembly location (nucleus) in to the cell culture medium (in vitro) or haemocoel (in vivo). Our results enable us to propose a unique model for the structural reorganisation of P10 over the course of the virus infection cycle that explains the multiple roles attributed to this small, abundant viral protein. Trichoplusia ni (TN-368) cells were infected with AcMNPV at a multiplicity of infection (MOI) of 5 plaque forming units (pfu) /cell, and processed by embedding in resin for SBFSEM imaging at 24,48,72 and 96 hours post-infection (hpi). Image capture was optimised and typical baculovirus-infected cell structures such as the virogenic stroma (VS), P10, electron dense spacers (EDS) and occlusion bodies (OBs) were observed (Fig 1). SBFSEM data were collected by imaging the cut surface of the resin-embedded AcMNPV-infected cells using a Gatan 3View system (Gatan, UK) in combination with a Zeiss Merlin compact VP scanning electron microscope (SEM, Zeiss, UK). A backscattered electron signal was used, which provided atomic contrast of the stained cells that was comparable to data collected using a TEM (Fig 2Ai). A diamond knife mounted within the specimen chamber was set to automatically slice the sample, with scanned images of the block face being captured between each 100nm slice [34,35]. The subsequent stack of images (500–700) were aligned and segmented to produce 3D models of cellular and virus structures at different stages in the viral replication cycle as represented by 24,48,72 and 96 hpi (Fig 2Aii and 2Aiii; S1–S5 Videos). The P10 structures within AcMNPV-infected cells [36] were modelled at each time-point. Two distinct P10 fibrous structures were observed; a cytoplasmic perinuclear cage and a nuclear vermiform mass (Fig 2Aii). At 24 hpi, cytoplasmic P10 (light blue) was observed as thin filamentous branching structures. In contrast, at the same time point, nuclear P10 (dark blue) was observed as a single thin worm-like or vermiform structure (Fig 2Aii, 24 hpi). The two structures were separated by the host cell nuclear membrane (Fig 2Aiii, dark grey). At 48 hpi, the cytoplasmic P10 was demonstrably thicker than at 24 hours and had formed an elaborate network that enwrapped the nucleus (Fig 2Aii and 2Aiii, 48 hpi). At this time point, nuclear P10 had also increased in size, forming a thick elongated ‘worm’ (Fig 2Aii, 48 hpi). By 72 hpi, the cytoplasmic perinuclear P10 structure had remodelled to a form with fewer but thicker tubules (Fig 2Aii and 2Aiii, 72 hpi). However, at 72 hpi nuclear P10 (Fig 2Aii and 2Aiii, 72 hpi) appeared similar in morphology to that observed at 48 hpi. By 96 hpi, a significant change in the morphology of the cytoplasmic P10 was apparent (Fig 2Aiii). The perinuclear cage structure was still present, but it lacked the extensive network of thick tubules observed at 48 and 72 hpi (Fig 2Aii and 2Aiii, 96 hpi). At this very late stage of the infection cycle, most P10 had coalesced to form a large, polarised aggregate, with a few remaining thin fibrils branching from it (Fig 2Aii and 2Aiii, 96 hpi). Simultaneously nuclear P10 maintained its characteristic long vermiform architecture (Fig 2Aii and 2Aiii, 96 hpi) and at this very late stage was surrounded by virogenic stroma and densely packed OBs. To investigate the changes in P10 over time, we estimated the diameter (nm) and relative cell volume (%) of the P10 structures. Representative diameter measurements were acquired manually in AMIRA using 3D volume rendered images of cytoplasmic P10 fibrils/tubules (n = 10) and nuclear P10 (n = 10) structures. These measurements were acquired from duplicate AcMNPV-infected cells modelled at 24,48,72 and 96 hpi (Fig 2Bi). AMIRA-generated 3D models of AcMNPV-infected cells were also used to calculate volumes of the cell nucleus, cytoplasm, nuclear P10 and cytoplasmic P10 to give an accurate representation of the changes that occurred during the virus replication cycle. In this study, the volumetric measurements are given as a percentage of the total cell volume (Fig 2Bii). The combination of diameter and volumetric measurements from 24 to 96 hpi confirmed that both cytoplasmic and nuclear P10 structures increased in size as the infection cycle progressed. At 24 hpi, cytoplasmic and nuclear P10 structures had a mean diameter of 380 (± 186) nm and 370 (± 44) nm, respectively and together accounted for just 0. 1% of the total cell volume (Fig 2Bi and 2Bii). By 48 hpi, the mean diameter of cytoplasmic P10 fibrils had nearly tripled in size to 1180 (± 367) nm, with nuclear P10 structures also increasing to 1680 (± 370) nm (Fig 2Bi). Volumetric measurements indicated a combined relative percentage increase to 6% of total cell volume. Cytoplasmic P10 structures continued to increase in diameter through 72 hpi but at this stage also demonstrated a wide range of diameters from 980 to 5520 nm (mean of 2547 ± 735 nm) with nuclear P10 recording a mean diameter of 2520 nm (Fig 2Bi). Volumetric readings confirmed these data, and at 72 hpi the P10 structures comprised 11% of the total cell volume (Fig 2Bii). Measurements of the polarised aggregate of cytoplasmic P10 at 96 hpi correlated to the features observed both in 2D (Fig 2Ai) and 3D (Fig 2Aii and 2Aiii) microscopy, with the structure measuring as much as 14,150 nm diameter (mean 6258 ± 4511 nm) with the thin branching filaments at 520 nm. At this stage the diameter of nuclear P10 structures had increased to 2940 (± 530) nm (Fig 2Bi) and total P10 structures accounted for 17% of the relative percentage volume of the infected cell (Fig 2Bii). Analysis of AcMNPV-infected TN-368 cells using confocal microscopy showed the formation of P10 associated with thick peri-nuclear structures that encapsulated the nucleus at 72 hpi (Fig 3, Confocal, i and ii, arrow). During the later phase of infection at 96 hpi, P10 was associated with a thick cage-like structure around the nucleus and observed as a thick band (Fig 3, Confocal, iii and iv, arrow). The thinner cytoplasmic filaments had condensed and disassociated from the periphery of the nuclear membrane. This is in agreement with previously documented results [30]. These morphological structures observed using confocal microscopy are consistent with the 3D reconstruction of P10 using SBF-SEM (Fig 3, SBF-SEM). To test the hypothesis that hyper-expression of p10 is critical for nuclear lysis, we constructed a series of AcMNPV p10 promoter deletions within recombinant viruses (Fig 4). Controls for this experiment included recombinant viruses that either contained the full length p10 promoter with the native P10 coding sequence (Ac_P10Rescue) or lacked the coding sequence (AcΔp10). After infection of cells with the control viruses or viruses containing promoter deletions, we showed using Coomassie staining (Fig 5Ai) and western blot analysis (Fig 5Aii) that there was a progressive decrease in P10 accumulation as base pairs within the p10 promoter sequence were deleted (Fig 5Ai and 5Aii, lanes 3–7). The viruses used in this study contained 4,8, 12,16 or 20 base pair deletions in the p10 5’ non coding leader sequence beginning at the -1 position relative to the ATG. Viruses designated Ac_P10prl-16 and Ac_P10prl-20 yielded levels of P10 that were below the detection limits for Coomassie staining. Interestingly, Ac_P10prl-12 displayed higher expression levels of P10 than that expected from neighbouring deletions, with a band density more comparable to AcMNPV and Ac_P10Rescue (Fig 5Ai and 5Aii, lane 5). AcMNPV cathepsin was used as a marker to ensure equal loading on samples for the Western blot (Fig 5Aii). As expected, a P10-associated band was not detected for AcΔp10 either by Coomassie staining or Western blot (Fig 5, lane 8). By comparing band density on replicate Western blots (n = 3), a reduction in P10 expression with progressive removal of nucleotides from the p10 promoter was confirmed (Fig 5B). A number of functional roles have been suggested for P10 [reviewed in 37] as described above. Most distinctive in cell culture is the role of P10 in nuclear lysis or disintegration to facilitate dispersal of OBs [15,17]. In the absence of P10, nuclear lysis does not occur and OBs are not released into the culture medium [15]. To determine whether hyper-expression of p10 is required for nuclear lysis and release of OBs, T. ni Hi5 cells were infected with wild-type AcMNPV, Ac_P10Rescue, AcΔp10, Ac_P10prl-4, Ac_P10prl-8, Ac_P10prl-12, Ac_P10prl-16 and Ac_P10prl-20 at 5 MOI and imaged at 7 days (d) pi using a light microscope (Fig 5D). A visual observation of virus-infected cells at 7 dpi indicated that the number of OBs released was directly related to the level of P10 synthesis (Fig 5B–5D). To confirm these results, the number of free OBs observed in a field of view at 7 dpi were enumerated using a haemocytometer (n = 10 for each virus) at 100X magnification (Fig 5C). The quantitative data confirmed the visual inspection results that Tni Hi5 cells infected with recombinant viruses lacking progressive deletions in the p10 generated decreasing numbers of free OBs in the media (Fig 5C; P<0. 05; one-way ANOVA). However, Ac_P10prl-12 had similar numbers of free OBs as counted for both AcMNPV (P>0. 9999) and Ac_P10Rescue (P>0. 9918). No signs of nuclear lysis or free OBs were observed in AcΔp10-infected cells (Fig 5C and 5D). This confirmed a relationship between P10 accumulation (Fig 5A nd 5B) and OB liberation (Fig 5C and 5D) and suggests modulation of P10 levels has a direct effect on nuclear lysis. To further investigate the relationship between levels of P10 and nuclear lysis to release OBs, we infected TN-368 cells with the promoter deletion viruses Ac_P10prl-4 and Ac_P10prl-20 and examined the cells using confocal microscopy to observe any effects on P10 structure formation. We observed that reduced p10 expression from the promoter deletion viruses (Fig 5A and 5B) impacted formation of P10 structures (Fig 5E). Reduced levels of expression was correlated with a delay in, or absence of, the formation of the thicker perinuclear tubular structures that form the cage-like structures characteristic of P10 in the very late stages of infection [30]. The control wild-type AcMNPV and Ac_P10Rescue virus formed a P10-associated cage-like structure around the nucleus as expected. In Ac_P10prl-4-infected TN-368 cells, P10 was observed only as thinner, peri-nuclear cytoplasmic filaments with some branching out in the cytoplasm (Fig 5E). In Ac_P10prl-20-infected cells, few P10 structures were detected and none were observed in a peri-nuclear location (Fig 5E). The data support the hypothesis that reduction of P10 levels (Fig 5A and 5B) is concomitant with reduction in nuclear lysis (Fig 5C and 5D) and the confocal analysis (Fig 5E) suggests that impaired development of P10 structures may provide the mechanistic link. In addition to nuclear lysis or disintegration, it has also been suggested that P10 plays a role in calyx or PE formation, the final stage in the maturation of OBs [15,20,22]; although no one has hitherto been able to describe how P10 may be involved in such diverse functions. High-resolution SEM images of OBs isolated from Ac_p10Rescue (Fig 6Ai and 6Aii) and AcΔp10-infected cells (Fig 6Aiv and 6Av) showed that there were readily distinguishable morphological differences in OBs in the presence (Ac_p10rescue) or absence (AcΔp10) of P10. The OBs extracted from Ac_p10Rescue-infected TN-368 cells predominantly presented a characteristic polyhedron shape with a smooth calyx; a few mis-formed or immature OBs were also detected (Fig 6Ai and 6Aii). Thust OBs extracted from AcΔp10-infected cells exhibited a spherical phenotype that lacked the smooth outer layer (calyx/PE) and often resulted in OBs with a pitted surface (Fig 6Aiv and 6Av). This pitted surface is characteristic of an incomplete calyx/PE and subsequent loss of occlusion derived viruses (ODVs) during preparation of OBs; indicating a lack of ODV stability within OBs in the absence of a complete polyhedron envelope [38]. Measurements of OB diameter (data in S1 Fig), confirmed by a one-way ANOVA (P<0. 05), showed a statistically significant difference (P<0. 0001, n = 3) between OBs extracted from AcΔp10-infected cells (mean = 2. 521 μm) and either AcMNPV (mean = 2. 896 μm) or Ac_p10Rescue (mean = 2. 915 μm). This gives a 12. 7% and 13. 6% reduction in the diameter of the OBs extracted from AcΔp10-infected cells when compared to AcMNPV or Ac_p10Rescue respectively. A Tukey’s multiple comparisons post-hoc analysis confirmed the mean size of the AcΔp10 OBs to be statistically different to both the AcMNPV (P = <0. 0001) and Ac_p10Rescue (P = <0. 0001). The OBs derived from AcMNPV and Ac_p10Rescue showed no statistical variation (P = 0. 9569) in mean OB size. We found from TEM images of Ac_p10Rescue- and AcΔp10-infected cells that in the absence of P10, EDS failed to wrap around the periphery of OBs (Fig 6Aiii and 6Avi). Instead, EDS were observed to wrap onto themselves, forming self-aggregated spherical structures. This suggests that P10 may play a chaperone-like role in transferring the EDS to the surface of maturing OBs and supports the theory that P10 has a role in the formation of the calyx/PE. Furthermore, Fig 6 Aiii and 6Avi, demonstrated that EDS were present in the absence of P10, confirming that P10 is not a pre-requisite for EDS formation. To analyse the extent of the association between nuclear P10, EDS and OBs, we reconstructed 3D models of EDS (red), nuclear P10 (blue) and OBs (black) at different time-points after infection (Fig 6B). The presence of EDS were first observed at 24 hpi, identified only as small patches associated with the thin nuclear P10 fibrous body (Fig 6Bi and 6Bii). The OBs were absent or immature at 24 hpi (Fig 6Bi). By 48 hpi, extensive areas of nuclear P10 were coated with EDS, with some OBs also showing early signs of association with EDS (Fig 6Bi–6Biii). At 72 hpi, large areas of P10 nuclear structure were thickly coated with EDS with some appearing to form aggregates that branch off the main P10-associated EDS (Fig 6Bii). In multiple cells observed at 72 hpi, EDS were also aggregated in a layer covering large regions of the OB surface (Fig 6Biii), suggestive of OBs being encapsulated by the EDS. By 96 hpi, the majority of the nuclear P10 and OB surfaces were coated with at least one layer of EDS (Fig 6Bii and 6Biii). Interestingly, the 3D models of AcMNPV-infected cells indicated a continued production of EDS through-out virus infection resulting in multiple layers of EDS around a single OB at 96 hpi. At this time, we estimated that 80% of OBs were associated with EDS and approximately 50% of OBs were completely encapsulated within a layer of EDS. These models highlight the intimate relationship between P10, EDS and OB structures and provide convincing evidence of a chaperone-like role for P10 in the formation of the polyhedral envelope or calyx. Previous studies using anti-tubulin and P10-specific antibodies showed that P10 structures associate with microtubules and could have a role in the reorganisation of the cytoskeleton as well as the initiation of P10 structure formation [30,31]. Depolymerisation of the microtubule network with colchicine prevented formation of P10 filaments and a yeast two-hybrid experiment identified an interaction between P10 with host-cell tubulin [31]. However, it was noted that the thicker P10 tubules that form later, and which merge to form a perinuclear cage-like structure, are not closely associated with microtubules [30,40]. Similar P10 filamentous structures have been observed with entomopoxvirus formed by the filament-associated late protein of entomopoxviruses (FALPE) [9]. FALPE shares structural and functional similarities to baculovirus P10 with a close association to the entomopoxvirus occlusion bodies [41]. The SBFSEM images generated in this study enabled us to create detailed 3D models of cytoplasmic P10, which confirm the formation of the perinuclear cage structure proposed in the earlier confocal microscopy data. Furthermore in this study, we show that the perinuclear cage structure is dynamic and re-models over time, eventually coalescing into a polarised mass of P10 that is concomitant with the time of cell lysis (around 96 hpi in cell culture). From these data we can propose a model (Fig 7) in which cytoplasmic P10 first assembles into filaments using a scaffold of microtubules. These later (48 to 72 hpi) develop into thicker, longer tubules that interact to form a perinuclear network or cage-like structure that may stabilise the nucleus allowing time for OBs to develop and mature. Finally at around 96 hpi, as the cage-like structure continues to remodel, P10 withdraws from the tubules coalescing into a polarised mass. At some stage in this final process, our model predicts that the integrity of the nuclear membrane is destabilised causing nuclear lysis and release of OBs, which in nature would assist in OB dispersal for horizontal transmission after other viral factors have liquefied the host (Fig 7). Our model is supported by the timing of this event that coincides with nuclear lysis of AcMNPV-infected TN-368 cells as observed by light microscopy [15]. Our model is also consistent with observations made during these studies (Fig 5D), and historically, that deletion of p10 results in both the absence of P10 structures and the abrogation of nuclear lysis [15,17]. One earlier study examining the role of structural domains of the P10 protein through the construction of deletion mutants, concluded that the process of nuclear lysis continued in the absence of an identifiable P10 structure [14] and this conflicts with our proposed model, in which cell lysis is dependent on the formation of P10 structures. However, more recent studies have presented TEM and confocal images of cells infected with viruses containing similar deletions to those described in [14]; these data clearly showed that P10 structures were formed in these cells [42,43], which is consistent with our model. Early reports on P10 proposed a role in OB maturation, which was a result of TEM studies on baculovirus structures [5,23] that observed P10 in close association with EDS and OBs [19,22,29,44,45]. These OBs were observed to be surrounded by EDS and it was suggested that EDS aided formation of the PE/calyx by a process of condensing and compression [39]. Even though a number of studies have commented on an association between P10, EDS and the PE/calyx [19,22,29,45], there was little evidence for a mechanism to explain the role of P10 in calyx formation, with at least one study sceptical of the involvement of P10 [17]. Using high-resolution SEM images of OBs, this study provides further evidence that P10 is required to form an intact PE/calyx and 3D modelling of P10 provides new evidence for a close spatial and temporal relationship between P10, EDS and OBs. Using the acquired data, we show an increase in formation and association of EDS with both P10 and OBs during the later phase of infection (48–72 hpi). Nuclear P10 becomes almost entirely coated with EDS that appear to stack up upon each other and by 96 hpi, multiple OBs are fully encapsulated by EDS. Our data supports the suggestion [35] that P10 plays a chaperone-like role in transferring the EDS to the surface of maturing OBs as clearly depicted in the images in Fig 6B. It is important to note that p10 deletion mutants show no impairment in the production of polyhedral envelope protein (pp34) or EDS production [19] but fail to form an intact calyx [22] as confirmed here by SEM (Fig 6Ai). In the absence of P10, we also noted that the EDS failed to wrap around the periphery of OBs and instead formed self-aggregated spherical structures (Fig 6Aiv). This confirms a requirement for P10 in the formation of the PE. It is also possible that the principle function of P10 is to maintain nuclear integrity via the formation of its cage structure to enable the maturation of the occlusion body with an intact calyx. A process that provides protection from environmental hazards and limits loss of ODV. This study reports the first instance of targeted control of p10 expression through modification of its promoter, and provides evidence that hyper-expression of p10 is required for P10 to fulfil its biological role in mediating nuclear lysis. In our study, we detected a direct correlation between the amount of P10 protein synthesised and the ability of cells to undergo nuclear lysis, with a critical amount of P10 required to maximise nuclear lysis and hence aid OB dispersal. Previous studies examining the p10 promoter also observed a reduction in expression with progressive deletions made to the 5’ non-coding region of the promoter [46,47]. However, in these studies the p10 coding region had been replaced with that of the CAT reporter gene so the effect on the levels of P10 protein was not recorded. Tni Hi5 cells infected with Ac_P10prl-12 gave comparable levels of P10 protein to that detected in both AcMNPV and Ac_P10Rescue, and we observed formation of normal P10 structures and full nuclear lysis. We have no explanation for why a promoter deletion should result in wild type levels of protein production, but its sequence was confirmed in the recombinant virus. In contrast, the other promoter deletions resulted in low levels of P10, few or no recognisable P10 structures and low levels of nuclear lysis. The relatively high levels of p10 expression observed in Ac_P10prl-12 were not previously documented [46,47]. We can derive from this study that manipulation of the p10 promoter decreased P10 protein levels that resulted in a diminished P10 cage structure with a concomitant decrease in nuclear lysis and abrogation of OB release into the culture medium. It has long been debated why both p10 and polh are under control of the strong, very late promoters that are hyper-expressed in baculovirus-infected cells. Large quantities of polyhedrin are required to form the paracrystalline array that comprises the structure of OBs into which ODV are embedded and protected [48,49]. P10, which is also produced in large quantities, is non-essential for the production of orally infective ODV in vivo [12,15]. However, that there has been a strong evolutionary pressure to preserve the characteristic P10 structure indicates an advantage for p10 [50,51]. One suggestion is that expression of p10 promotes the dispersal and spread of the viral progeny [50]. Our data suggest that large quantities of P10 are required to form the cytoplasmic cage-like structure that provides a mechanistic action to mediate nuclear lysis and OB release. In conclusion, the 3D modelling of AcMNPV structures from a high-resolution data set identified the formation of two independent P10 structures in cultured cells. Cytoplasmic P10 forms a dynamic perinuclear cage-like structure that we propose plays a crucial role in mediating nuclear lysis and OB dispersal. Hyper-expression of p10 is required for cage formation and consequently nuclear lysis. The nuclear form of P10 comprises a single vermiform structure that most likely plays a chaperone-like role in facilitating the maturation of OB by promoting the transfer of EDS to form the polyhedral PE/calyx. Cells used were derived from Tn Hi5 (Invitrogen) and TN-368 [36] and were maintained at 28°C in ESF921 (Expression Systems) or TC100 media (Invitrogen) supplemented with 10% foetal bovine serum, respectively. AcMNPV clone 6 virus was propagated and amplified using standard methods [52]. The construction of a p10-knockout virus, AcΔp10, rescue virus, Ac_p10rescue (S2 Text), and series of promoter deletions in the 5’ untranslated sequence of p10 mRNA (S3 Text) were produced by co-transfecting Sf9 cells with Bsu361-digested AcΔp10_lacZ DNA and designated transfer vector (S1 Text), followed by plaque-purification and amplification of viruses [52]. Oligonucleotide sequences are provided in S2 Table. AcMNPV-infected TN-368 cell pellets at 0,24,48,72 and 96 hpi were washed with PBS and placed into primary fixative (2% paraformaldehyde, 2. 5% glutaraldehyde and 0. 1% tannic acid in 0. 1M sodium cacodylate buffer, pH 7. 4). Six 5-second microwave bursts at 5 second intervals (Russel Hobbs, Digital microwave, 800W) were applied to the cell pellets and fixative solution for rapid microwave fixation [53], followed by a 15 minute incubation at room temperature. The cell pellet was washed four times by re-suspending in 0. 1M sodium cacodylate buffer (pH 7. 4) at room temperature for 5 minutes followed by low speed centrifugation. Secondary fixation was performed with 1% osmium tetroxide in 0. 1 M sodium cacodylate buffer, followed by six 5-second microwave blasts and a 1 hour incubation at room temperature. The cell pellet was rinsed x5 with distilled water, centrifuged to prevent sample loss, and then 2% osmium tetroxide in 0. 1 M sodium cacodylate buffer was applied and incubated for 40 minutes at room temperature. After washing in water again, samples were dehydrated in an ascending ethanol series to a final concentration of 100%, with an intermediate incubation overnight at 4°C in 70% ethanol containing 2% uranyl acetate. After dehydration, the cell pellet was infiltrated and embedded in Epon 812 resin and polymerised at 70°C for 18 hours. The resin embedded samples were sectioned and imaged using a Gatan 3View system including the 3View2XP stage and 3VBSED detector (Gatan, Abingdon, UK) in combination with a Zeiss Merlin compact VP SEM (Zeiss, Cambridge, UK). The sample was imaged using an accelerating voltage of 4 kV, aperture size of 30 μm and dwell time was 3 μs. Variable pressure was between 30–45 Pa. Pixel size was 6. 9–15 nm for 500–700 sections at 100nm each. See S1 Table for imaging parameters for each time point. The raw image datasets were acquired using a digital micrograph format (dm4; Gatan, Abingdon, UK). The images were then converted to obtain stacked image files (mrc) for data processing using a free iMOD [54] and AMIRA (ThermoFisher) software packages. Structures were selected in segmentation, using both automatic (thresholding and interpolation) and manual tools to define viral structures. Structures were modelled based on their classic morphological features. The 3D volume rendered images were used for precise measurements. P10 diameter data were taken at 10 representative sites for cytoplasmic P10 fibrils/tubules and 10 representative sites for nuclear P10 per cell. The data were analysed using GraphPad Prism 7 (GraphPad Software Inc. , USA). SEM was performed on OBs extracted from baculovirus-infected TN-368 cells (at 5 MOI) at 7 dpi as previously described [45]. Purified OBs were fixed in 4% (v/v) formaldehyde in PIPES buffer for 1 hour, washed once in PIPES buffer and then dehydrated in an ascending ethanol series to 100%, for 10 minutes each. The dehydrated OBs were densely seeded onto glass coverslips and sputter-coated with gold (Automatic sputter coater, Agar Scientific, Stansted, UK). The samples were imaged using a Zeiss Merlin Compact VP SEM using an accelerating voltage of 4kV. Post-acquisition image processing was performed with Image J and analysed using GraphPad Prism 7. Proteins were separated in 15% gels using the Mini-PROTEAN system (Bio-Rad) as previously described [28] and stained with Coomassie blue or immuno-stained using rabbit anti-P10 (CFELDSDARRGKRSSK, 1: 500; Genscript) antisera with alkaline phosphatase conjugated secondary antibody (1: 30,000; Sigma-Aldrich). Colorimetric images were taken using a ChemiDoc MP imaging system (Bio-Rad). Immunofluorescence staining of infected TN-368 cells using rabbit anti-P10 antibody (1: 700) and Alexa-fluor 488 -conjugated anti-rabbit secondary antibody (1: 1000) was as described previously [30,55]. For image acquisition, a Zeiss LSM 510 meta laser scanning microscope using the 488nm excitation line of the argon laser was used. Fluorescence was detected using a 488/543 nm dichroic beam splitter and a 505-530nm band pass filter for Alexa-Fluor 488. Images were acquired using an oil immersion objective; Plan-Apchromat 63X (1. 4 numerical aperture). Post-acquisition image processing and Z-stack image projections were processed using LSM 5 image Browser (Zeiss, Cambridge, UK). Data were analysed using software package GraphPad Prism 7. Statistical analysis was performed using both one- and two-way ANOVA with a Tukey’s multiple post-hoc test. A p-value of <0. 05 was considered statistically significant.

High-resolution 3D electron microscopy has revealed the complexity of structures formed by P10, a small 10kDa protein that accumulates to very high levels in baculovirus-infected cells. We demonstrate the formation and presence of two distinct, possibly unique, P10 structures that account for the diverse roles associated with this small protein. In the cytoplasm, a peri-nuclear cage-like structure matured into a polarised mass of P10. Remodelling of the cage provides evidence for a mechanism to effect nuclear lysis and release of occlusion bodies to promote dispersal. Over a similar time period, an independent vermiform P10 structure forms and matures within the cell nucleus. It is widely known that in the absence of P10, occlusion bodies do not fully mature. Our data suggest a mechanism for occlusion body maturation with P10 facilitating the envelopment of occlusion bodies with electron dense spacers. The P10 structures formed require vast quantities of P10 protein providing a rationale for the hyper-expression of this hitherto obscure viral protein.

Abstract Introduction Results Discussion Materials and methods

plant anatomy microtubules biological cultures microbiology light microscopy viral structure plant science microscopy cell cultures protein structure confocal microscopy cellular structures and organelles flower anatomy cytoskeleton research and analysis methods proteins scanning electron microscopy molecular biology cytoplasm calyx biochemistry cell biology virology electron microscopy biology and life sciences macromolecular structure analysis

2019

In cultured cells the baculovirus P10 protein forms two independent intracellular structures that play separate roles in occlusion body maturation and their release by nuclear disintegration

Is it possible to extract tethering forces applied on chromatin from the statistics of a single locus trajectories imaged in vivo? Chromatin fragments interact with many partners such as the nuclear membrane, other chromosomes or nuclear bodies, but the resulting forces cannot be directly measured in vivo. However, they impact chromatin dynamics and should be reflected in particular in the motion of a single locus. We present here a method based on polymer models and statistics of single trajectories to extract the force characteristics and in particular when they are generated by the gradient of a quadratic potential well. Using numerical simulations of a Rouse polymer and live cell imaging of the MAT-locus located on the yeast Saccharomyces cerevisiae chromosome III, we recover the amplitude and the distance between the observed and the interacting monomer. To conclude, the confined trajectories we observed in vivo reflect local interaction on chromatin. What can we learn about the local environment, the external and internal forces and the chromatin itself from the motion of a chromatin locus? This motion can be driven by local diffusion and/or forces between monomers of the model polymer [1–3]. Monomers motion is highly correlated due the polymer hierarchy of relaxation times [4,5], leading in particular to anomalous diffusion [6,7]. This anomalous behavior is well documented for chromatin loci [8–10] and we propose here to examine the effect of local external interactions on a locus motion. Much of the chromatin dynamics is reflected in the motion of a single chromosomal locus and conversely, a locus motion allows probing the chromatin dynamics [11,12] at tens of nanometers and millisecond scales resolution [13–15]. When this motion is described as a free or confined Brownian motion, classical statistical tools such as the mean square displacement (MSD) and radius of confinement [16–18] can be used to extract the values of physical parameters. Other methods have been developed to extract kinetic rates about molecular events from forces imposed in pulling experiments [19,20] or in the context of atomic force microscopy [21,22]. Polymer models can account for various forces acting on chromatin, such as bending elasticity, internal rigidity, torsion and Lennard-Jones interactions [2]. In addition, the chromatin fiber can experience local fluctuations driven by ATP [23,24], identified by micrometer long-range coherent [25] and active motion [26]. Other interactions can be due to repulsive forces or self-avoiding interactions with other chromatin parts, attractive forces driven by anchoring a locus at a nuclear pore [27] or tethering to the spindle pole body through the centromere [28] or with other chromosomes mediated by protein-protein interactions. While these interactions are local and extend to tens or hundreds nanometers, they can influence the polymer dynamics and in particular on this polymer, even if positioned far away from the interacting site (Fig 1a). We present here a method based on polymer models and statistical analysis of single particle trajectories, to estimate the local interactions acting on chromatin (Fig 2a). A sufficiently large ensemble of single tagged locus trajectories is the key ingredient of the method. When applied forces are stationary over the time course of the trajectory recording, we extract interactions or their mirror deterministic forces by deriving formulas that link the empirical velocity distribution of a locus to forces applied to a distant single monomer. The present method allows distinguishing external forces applied on a single monomer from intrinsic forces acting on monomers. The principle and the difficulty of the method can be understood as follows: for a single stochastic particle modeled by the Smoluchowski’s limit of the Langevin equation, the velocity of the particle v is proportional to a force f applied on the particle plus an additional white noise, summarized as γv=f+γ2Dw˙, (1) where γ is the friction coefficient, D the diffusion coefficient and w is the normalized Wiener process. Thus by averaging over the ensemble of velocity realizations, it is possible to recover the first moment, which is the force field [29]. However, for a polymer chain, there are internal forces between monomers and thus, the difficulty that needs to be resolved here, as the data are measured at a single monomer, is to separate the internal forces acting on the measured monomer from the external ones acting on a monomer further away. This problem is resolved here, but the inversion formula to recover the force depends on the polymer model. When the external applied force is the gradient of a quadratic potential (second inversion formula) we explicit the formula analytically and show that the motion of the observed monomer is characterized by an effective force, with an effective elastic spring constant kc that we compute. We simulate a Rouse polymer [4], which serves as a model for the chromatin structure [8,30]. The locus motion cannot simply be approximated as an Ornstein-Uhlenbeck (OU) process, with an effective harmonic potential well, but we show that the effective force acting on the observed monomer decays with the distance along the chain between the interacting and the observed monomer. The effective spring constant kc decays slower with this distance for a general class of polymers (β–polymer [31]) compared to Rouse. Applying the present approach to live cell-imaging data of the MAT-locus in yeast [32], which appears to be constrained shows that confined trajectories can either be due to local crowding or to direct interactions. Using Single Particle Trajectories (SPTs), we extract forces acting on that locus and show that trajectory localization is mediated by direct forces. This result validates the model predictions and the relation between the strength of a force applied on the chromatin locus and the radius of confinement. We conclude that local forces and not only crowding do confine chromatin motion. The present approach can further be applied to other situations, such as yeast telomeres anchored to the nuclear periphery [32], changes in single locus dynamics or repositioning following the induction of double-stranded DNA breaks. When an external force, which is the gradient of the potential Uext (R) is applied to a Rouse polymer, the interaction is described by the energy ϕ (R) = κ 2 ∑ j = 2 N (R j - R j - 1) 2 + Uext (R), (2) where R = (R1, R2, …, RN) is the ensemble of monomers, connected by a spring of strength κ = dkB T/b2. b is the standard-deviation of the distance between adjacent monomers [4], kB the Boltzmann coefficient, T the temperature and d the dimensionality (dim 2 or 3). In the Smoluchowski’s limit of the Langevin equation [33], the dynamics of monomer Rn is described by d R j d t = - D ∇ R j ϕ (R) + 2 D d w j d t, (3) for j = 1, … N and each wj is an independent d-dimensional white noise with mean zero and variance 1, D is the monomer diffusion coefficient. We will describe specifically the field of ∇Rj Uext (R) in the next subsection. When the chromatin motion is described by Rouse chain, the effective diffusion coefficient can be estimated from data. We shall choose a reference monomer Rc, which represents the tagged locus. One of the key results of the present analysis is the following formula, which links the velocity or first moment of the monomer of Rc (averaged over all realizations) to the polymer configuration distribution: General inversion formula: lim Δ t → 0 𝔼 { R c (t + Δ t) - R c (t) Δ t | R c = x } = - D ∫ Ω d R 1. . ∫ Ω d R N (∇ R c ϕ) P (R | R c = x), (4) where 𝔼{. ∣Rc = x} denotes ensemble averaging under the condition that the tagged monomer is at position Rc = x. Formula 4 is generic and does not depend on the particular expression of the external forces acting on the polymer. Moreover, we do not impose here any restriction on the domain Ω where the polymer evolves. The polymer is reflected on the boundary ∂Ω. The conditional probability P (R∣Rc = x) is computed from equilibrium probability distribution function (pdf) P (R1, R2, …, RN), which satisfies the Fokker-Planck equation (FPE) in the phase space Ω ×. . Ω ⊂ ℝ3N, 0 = Δ P (R) + ∇ · (∇ ϕ P (R) ), (5) with boundary condition ϕ ∂ P ∂ n i + P ∂ ϕ ∂ n i = 0 for R i ∈ ∂ Ω for i = 1. . N, where ni is the normal vector to the boundary ∂Ω at position Ri. A permanent force located at position μ can be approximated at order two by a harmonic well. We suppose that this force is applied to monomer n. The force applied on Rn is the gradient of the harmonic potential (Fig 1a) U ext (R n) = 1 2 k (μ - R n) 2, (6) where k is the force constant. The monomer n that experiences the force is different from the tagged monomer c and we shall assume that n < c. As we shall see now, this potential well affects the dynamics of the entire polymer and specifically the observed locus c. To extract the strength of the potential well applied on monomer n, from the measured velocity of locus c, we derive an analytical expression for formula 4. First, the force acting on monomer c, when its position is x is given by F R c = x c = - ∇ R c ϕ (R c - 1, R c, R c + 1) R c = x = - κ (x - R c - 1) - κ (x - R c + 1), (7) where the potential ϕ is defined in Eq (2). We take for now a potential well localized at the origin (μ = 0) in Eq (6). Moreover, the pdf at equilibrium is the Boltzmann distribution, conditioned on Rc = x, that is P (R | R c = x) = 𝓝 e - ϕ (R 1, . . . , R c - 1, x, R c + 1, . . , R N), (8) where the normalization factor is 𝓝 - 1 = ∫ Ω. . ∫ Ω ∏ i ≠ c P (R | R c = x). (9) Finally, computing Gaussian integrals (see S1 Text for details) we find that the normalization factor is (for μ = 0, otherwise we need to replace x by x − μ), 1N=[ (2π) N−1κ2−N (κ+| c−n |k) ]3/2e−x2 (κ2+ (c−n+1) kκ) 2 (κ+ (c−n) k). (10) Substituting Eqs (7) – (10) into Eq (4), we obtain (S1 Text) an explicit inversion formula for the mean velocity of monomer c. Second inversion formula: lim Δ t → 0 E { R c (t + Δ t) - R c (t) Δ t | R c (t) = x } = - D k c n x, k c n = k κ κ + (c - n) k. (11) Expression 11 is one the key result here: it links the average velocity over empirical trajectories of the observed monomer c to a permanent force applied on monomer n. The coefficient kcn depends on the harmonic well strength k, the inter-monomer spring constant κ and is inversely proportional to the distance ∣n − c∣ between monomers n and c along the chain. Furthermore, the steady state variance Rc = limt → ∞ Rc (t) of the monomer’s position (see S1 Text) can be related to the dimension d and the coefficient kcn by ⟨ R c 2 ⟩ = d k c n, (12) when ⟨Rc⟩ = 0. Relation 12 is reminiscent of long time asymptotic of classical Ornstein-Uhlenbeck processes. The dynamics of monomer Rc generated by Brownian simulations is shown in Fig 2b and 2c. In the limit of large k (pinned monomer), an analogue of formula 12 was used for analyzing chromatin organization [28] and DNA [34]. Inversion formula 1 assumes the Boltzmann distribution for the single monomer and that the entire polymer has reached equilibrium at the time scale of the simulation or the experiment (from Eq 8). Finally, formula 1 reveals how internal and external polymer forces mix together to influence the monomer velocity. It also shows the explicit decay of the force amplitude with the distance between the observed and forced monomer. We now study the consequences on the motion of a DNA locus of two forces acting on two monomers, located on two opposite sites of the tracked locus. The two monomers n and m (n < m) are interacting with two distinct potential wells applied at positions μn and μm (Fig 1a), the total potential energy of the Rouse polymer is U ext (R) = 1 2 k n (R n - μ n) 2 + 1 2 k m (R m - μ m) 2, (13) In that case, the average steady state position of the tagged monomer c can be computed exactly and is given by (see S1 Text for details) ⟨ R c ⟩ = { μ n k n (κ + | m - c | k m) + μ m k m (κ + | c - n | k n) k n k m | m - n | + (k n + k m) κ, n < c < m, k n μ n κ + k m μ m (κ + | m - n | k n) k n k m | m - n | + κ (k n + k m), n < m < c and similarly to the previous inversion formula, we can relate the velocity of Rc to the applied forces, summarized in this new formula Third inversion formula lim Δ t → 0 E { R c (t + Δ t) - R c (t) Δ t | x ˜ } = - D k c n m x ˜, (14) where x ~ = x − ⟨ R c ⟩ and k c n m = { k c n + k c m, for n < c < m (2 κ + | m - n | k) k κ κ 2 + | 2 c - m - n | k κ + | (m - n) (c - m) | k 2, for n < m < c where kcn and kcm are given by Eq (11) (see S1 Text). For n < m < c, in the limit m − n ≫ 1, we obtain the limiting formula kcnm ∼ ∣c − m∣−1 κ. Thus, the spring coefficient depends on the distance to the closest anchoring point only. However, when n < c < m, the effective spring coefficient depends on the distance between the two wells. Finally, the variance of the monomer position with respect to its mean position Eq (14) is given by ⟨ (R c - ⟨ R c ⟩) 2 ⟩ = d k c n m. (15) The computations are described in the S1 Text. We conclude at this stage that the distance scanned by the tagged monomer is proportional to the distance to the anchoring point (see Fig 1b and 1c). Several interacting forces can certainly be considered, but for a given locus, the two adjacent neighboring interacting monomers are probably enough to characterize the motion, because other forces should be screen by these proximity forces. We shall now extend the inversion formula to other polymer model with a prescribed anomalous exponent. Some refinement of the chromatin dynamics can be accounted for by a class of polymer models (β-polymer), generalizing the classical Rouse model. These polymer models account for long-range interactions between monomers, that decay with the distance along the chain [31]. Moreover, the characteristic of this class of model is to specify long-range forces acting on monomers so that a given monomer has a prescribed anomalous exponent [31]. Conversely, once the anomalous exponent is measured, it is then possible to construct a polymer with such given exponent. In that context, deriving an inversion formula for such polymer models is key to relate the velocity of a tagged locus to the external force, where the difficulty is to subtract the long-range internal forces between monomers, associated with the β-polymer to the total force and thus to recover the external forces applied to a different monomer than the one observed. We recall that for a polymer of N monomers, the dynamics of monomer c is govern by R c = α 0 c u 0 + ∑ p = 1 N - 1 α p c u p, (16) where α p c = { 1 N, p = 0 2 N cos ( (c - 1 / 2) p π N), otherwise. (17) and d u p d t = - D p κ ˜ p u p + 2 D d w ˜ p d t, (18) where D0 = D/N and Dp = D (p > 0), w p ˜ are white noises with mean zero and variance 1, the coefficients are κ ~ p = 4 κ sin (p π 2 N) β (2 > β > 1). At intermediate time, the cross-correlation function of a locus behaves as ⟨ (R c (t 0 + t) - R c (t 0) ) 2 ⟩ ∝ t α, (19) with α = 1 − 1 β [31]. When a gradient force (see Eq (6) ) acts on monomer Rn of a β-polymer, the expectation of the velocity of monomer c (c > n) is: Generalized inversion formula: lim Δ t → 0 E { R c (t + Δ t) - R c (t) Δ t | R c (t) = x } = - D k c n (β, N, l, m) x, (20) where μ = 0 and k c n (β, N, l, m) = A c, c - ∑ l, m ≠ c A l, c A m, c C ˜ l, k - 1, (21) where C ~ is a block matrix, the i-th block of which is C ˜ j, k i = A j, k i + k δ i, n δ j, n, (22) and [31] A j, k = ∑ p = 0 N - 1 κ ˜ p α p j α p k. (23) To conclude, inversion formula Eq 20 for a β-polymer is similar to the one derived for a Rouse polymer Eq (11), but the dependency with the parameters is now implicit. Numerical simulations of Eqs 21–23 reveal that the apparent spring constant kcn (β, N, l, m) decays slower with the distance ∣c − n∣ (between the interacting and the observed monomer) for smaller β (Fig 3a and 3b). When the chromatin experiences several interactions between distant sites along the chain, the external interactions propagate along the chain. In the previous section, we showed how to extract the local interaction between the underlying polymer and the surrounding environment from the trajectories of an observed locus. When the force is applied far away from the tagged locus, it is possible to recover the strength of the force and the distance where it was applied from the statistics of trajectories. The three inversion formulas we derived above can be used for different polymer models. In this section, we apply these formulas to extract parameter from numerical simulations and then we present a computational method to recover forces (chromatin interactions) from trajectories of the MAT-locus imaged in living yeast cells. We have shown here how to extract from single locus trajectories, chromatin tethering mediated by interactions with its surrounding environment. The presented method allows recovering an external force applied on chromatin although this one occurs far away from the observed locus. We note that this analysis is valid, although the recorded trajectories are possibly shorter than the relaxation time of the anchored chromatin. However, it is not yet possible to discern the forces from a locus positioned between two different interacting potential wells from the one generated by a single force located far away from the observed locus. In the complex nuclear environment, interactions of different strength can be randomly and transiently distributed along the chromatin. However, the resulting force on a single locus should mostly be generated by the sum of the two nearest interacting forces (derived from two stable potential wells). The distribution of the spring values kc shown in Fig 4c can be attributed to different interaction strength (k—Eq 6) or to the distances between the observed locus and the nearest interacting wells Eq (15). Other traps beyond the two nearest ones should certainly have an additional but lower contribution that needs to be estimated. A refined description of interactions on the chromatin would require monitoring simultaneously several loci. The present approach is also applicable for higher order organized polymer, modeled by β-polymers and we extracted here in vivo interactions of the chromatin with other nuclear element that were reflected in the motion of the MAT-locus. These interactions are responsible for constraining the locus in a small fraction of the nucleus. The motion of the chromatin is driven by both thermal fluctuations and by active ATP-dependent forces [25]. While our modeling is relevant to extract an interaction that does not change during the time acquisition of the trajectory, the spring constant kc that would be extracted during an active chromatin motion could be differentiated from the thermal one by projecting the dynamics perpendicular to the direction of motion. Finally, the present approach could also be used to study how chromatin modifications occurring during gene transcription or double stranded DNA repair affect the dynamics of a given locus. Yeast strains used in this study are all derivatives of the JKM179 strain [11] which is MATα ade1 leu2-3, leu2-112 lys5 trp1: : HisG ura3-52. The strain was obtained through insertion of both a Lac operator array (256 lacOp repeats), a Nup49-mCherry fusion and a non-tetramerizing lac repressor-GFP fusion under the HIS3 promoter into JKM179. To serve as a static reference point in the nucleus, the Spc42 protein was fused to yEGFP. All insertions or deletions were verified by PCR and phenotypic assays.

Is it possible to recover the local environment, the external and internal forces acting on a polymer from a single locus trajectories? To study this question, we resolve this reverse cell biology problem by developing a method that uses in vivo live single locus trajectories to extract physical forces applied on chromatin. We applied the method to the statistics of the S. cerevisiae MAT-locus motion and recover tethering forces acting on the chromatin. The local confinement of a chromatin locus can either be due to crowding or to local interactions with partners such as the surface of the nuclear membrane, other chromosomes or nuclear bodies that cannot be directly measured. We conclude here that confined trajectories of a single chromatin locus can be generated by local tethering interactions. This approach is applicable to cells under various conditions, such as during double-stranded DNA break repair.

Abstract Introduction Results Discussion Materials and Methods

2015

Analysis of Single Locus Trajectories for Extracting In Vivo Chromatin Tethering Interactions

Ochollo is a village in southern Ethiopia burdened with cutaneous leishmaniasis (CL), where Phlebotomus pedifer is the only vector for Leishmania aethiopica and hyraxes are confirmed reservoir hosts. A detailed description of the different players of transmission, and the ecology and seasonality of the vector needs to be established in order to accomplish efficient control programs. Between March 2017 and February 2018, a monthly sandfly collection was carried out in different habitats and records of temperature and humidity were taken. Rodents and hyraxes were trapped in the dry and wet season. All samples were screened for Leishmania kinetoplast DNA (kDNA). Positive samples were further processed for determination of the Leishmania species and the species of the sandfly/small mammal that was found infected. Additionally, the species of 400 sandfly specimens from different habitats and seasons was identified. 17,190 Sergentomyia and Phlebotomus sandflies were caught and showed an overall kDNA prevalence of 2. 6%, all were L. aethiopica infections only found in P. pedifer. The overall sandfly and P. pedifer abundance peaked in the dry season and was negatively correlated with the %RH. The kDNA prevalence varied over the months and was negatively correlated with the temperature. Total sandfly abundance did not differ between the sampled habitats, but P. pedifer was the distinct predominant species only in caves. Moreover, significantly more infected sandflies were found in caves. Only 1/192 rodents were kDNA positive, while 20. 0% (5/25) of Heterohyrax brucei were found infected. This study suggests that caves may be a source of multiplication of the infection. If an outdoor control program would be considered, it would be useful to focus on caves in the wet season, when the sandfly abundance is lowest. The captured rodent species appear not important for transmission and the contribution of hyraxes in transmission should be further investigated. Cutaneous leishmaniasis (CL) is a vector-borne disease caused by protozoa of the genus Leishmania and is listed as one of the neglected tropical diseases (NTD) [1]. It is characterized by nodules or ulcerative skin lesions, which can lead to secondary infections and disfiguring scars. In the Old World, female sandflies of some species of the genus Phlebotomus are the vector for Leishmania, as they can transmit the parasites when they take a blood meal [2,3]. Since the parasite’s host range includes other mammals aside from humans in Ethiopia, CL is a zoonosis [4]. The annual CL incidence in Ethiopia is estimated around 20 000 to 50 000 cases, which is probably an underestimation [5]. Endemic foci of CL are located in the (mid-) highlands [6], widespread in the country [7,8]. The main causative agent for CL is L. aethiopica [9] and occasionally L. major [10] and L. tropica [11]. The most common vectors are P. pedifer and P. longipes [8,12], although one study also obtained L. aethiopica promastigotes from P. sergenti [13]. A steep slope (>2. 15 degrees), an altitude between 1700m and 3500m and an average yearly rainfall between 1300 and 1700 mm were found positively correlated with the occurrence of CL [14]. In general, areas with CL are characterized by a temperate climate and rugged environments with cliffs, serving as a favorable habitat for sandflies and hyraxes [15]. Rock hyrax (Procavia spp.) and bush hyrax (Heterohyrax spp.) are thought to be the reservoir hosts for L. tropica and L. aethiopica in Ethiopia [8,12,15], Kenya [16–18] and other African countries [19–22]. In northern Africa, rodents are known to be potential reservoirs for L. major [23–25] and L. tropica [26]. Likewise, Ethiopian rodents and bats have recently been found positive for L. tropica DNA [27,28]. This suggests that rodents and other small mammals, whose burrows can be used by sandflies as resting and breeding sites, might be alternative animal hosts for L. aethiopica. Resting sites for phlebotomine sandflies can be almost anything, as long as it is relatively cool and humid [29]. The places identified so far are cracks in basalt cliffs, fissures and holes in walls, barns, caves used by hyraxes, rodent burrows, termite nests, soil cracks, tree trunks, etc. [12,29,30]. Ochollo is a village in the mid-highlands of southern Ethiopia where CL is mainly seen among children [7,31]. A study on primary school children in 2014 showed that 4% of the study population had active lesions, 59. 8% of them had scars and 1. 5% had both, making it a considerable public health problem. Scars and lesions were predominantly localized above the neck with the highest occurrence on the cheeks [7]. One study performed a species typing on skin scrapings of 35 CL patients, confirming the hypothesis that L. aethiopica causes CL [32]. Former studies described two Phlebotomus species in Ochollo: P. pedifer and P. ashfordi. P. pedifer is the most abundant species, shows anthropophilic behavior and is so far the only incriminated vector, though, this information was based so far on relatively small sample sizes [12,33,34]. The parasite species from five naturally infected sandflies was determined and turned out to be L. aethiopica [33]. During a study in 1973 in Ochollo, P. pedifer was found to live in close association with hyraxes in cracks in rocks and sandfly blood meals were found to come from humans and hyraxes [9]. This suggested that hyraxes, which are numerous in Ochollo, might play an important role in transmission of Leishmania parasites. Ashford et al. found four out of 19 Heterohyrax brucei (21%) naturally infected with Leishmania parasites, and thereby hypothesized that hyraxes might even be able to sustain the parasite in the hyrax population, without a human factor [12]. In this study, we evaluate whether the situation of sandflies and hyraxes is still similar to 45 years ago, we investigate whether also rodents are likely as natural hosts for Leishmania aethiopica in the region and we describe spatial and temporal variation in the abundance of (infected) sandflies. Since CL is a zoonosis and the ecology and transmission dynamics are unique to each area, such knowledge must be obtained to lay the groundwork for adequate disease control. Animal trapping and sample collection were conducted with authorization of the appropriate institutional and household authorities. Handling of the animals was carried out according to the 2016 Guidelines of the American Society of Mammalogists for use of small mammals in research and education. Permission from EWCA (Ethiopian Wildlife Conservation Authority) was not required according to the Ethical Clearance Committee of Arba Minch University, since our study site was a village which is not part of a protected area and the captured species are not endangered. Ochollo is a village situated approximately 20 km north of Arba Minch at 6°11’N, 37°41’E. It lies on the western side of the Ethiopian Rift Valley, at an altitude of approximately 2100 m. It has a temperate climate, with rainy seasons in February and March (average rainfall 400 mm), and between June and September (average rainfall 600 mm) [31]. The population consists of approximately 5000 people, living on top of hills and across steep slopes. The hills give rise to caves and crevices and the landscape is overall rocky and relatively densely vegetated. Sampling was carried out in three different habitats: caves, rocky areas and stone fences. Caves (Fig 1A) are highly abundant in Ochollo and are situated in the proximity of houses. They are located on cliffs (Fig 1B), where hyraxes are abundant. Rocky areas (Fig 1C) are a cluster of large boulders, with dark and humid crevices in between, where hyraxes are seen at daytime. Stone fences (Fig 1D) are manmade walls around household compounds and footpaths, constructed by stones from the surrounding area. The stones are covered with moss, indicating a high humidity. There are many small openings between the rocks, leaving space for potential resting sites of rodents and sandflies. Sandflies were collected from March 2017 until February 2018, for three consecutive nights between dusk and dawn in the three habitats every month (S1 Table). Sandflies were collected using CDC miniature light traps (John W. Hock Company, USA) and two types of sticky traps covered with sesame oil; sticky traps made from an A4 format cardboard plate covered with white papers and plastic were placed straight at the entrance of crevices, while white laminated A4 format papers were used to fold into cavities (S1 Fig). The traps were equally distributed over the sample sites belonging to the three habitats: caves, rocky areas and stone fences. Female sandflies were sorted out under the microscope. The head, wings and legs were disposed, and the thorax and abdomen were stored in 97% ethanol at -20°C until further analysis. Rodents were trapped with Sherman live traps (76 x 89 x 229 mm, Sherman Live Trap Co. , Tallahassee, FL, USA); in March, April, May, August and September 2017,400 traps were distributed in three different habitats during three consecutive days, baited with peanut butter and flour and checked daily at sunrise and just before sunset. Hyraxes were collected by local hunters using traditional snare traps. Mammals were initially identified based on taxonomic features, weighed and measured [35]. Venous blood was collected on filter paper (dry blood spots) and stored at -20°C. Ear samples were collected (4mm x 4mm) and preserved in 97% ethanol at -20°C. I-buttons (HQMatics, the Netherlands), small climate loggers, were placed at the sampling places to record the temperature and % relative humidity (%RH) on an hourly basis during the whole year. A short description of the weather conditions was recorded on the sampling days. The statistical analysis was performed with R version 3. 4. 3. Statistical tests were considered significant when the p-value < 0. 05. Small mammals were captured and subsequently screened with different PCR approaches (Table 1). We caught 192 rodents, of which one ear sample, derived from Mus mahomet, was positive for kDNA. The sample was not positive for ITS-1, thus the Leishmania species could not be determined. Of the three shrews (Crocidura olivieri) and four bats (Nycteris spp.) that were captured, none were found infected. From 25 captured hyraxes, which were all H. brucei, five (20. 0%) ear samples gave a positive result for kDNA. Of these five samples, four were positive for ITS-1 and when aligned in GenBank, all were L. aethiopica. Lesions were never seen on the animals’ nose or ears. None of the dry blood spot samples was positive. An overview of the sandfly data is presented in S1 Table. In total, 17190 (Sergentomyia and Phlebotomus) sandflies, 8410 females (48. 9%) and 8780 males (51. 1%), were caught between March 2017 and February 2018. Of the selection of 400 specimens from different habitats collected in January and July, 293 (73. 3%) were P. pedifer (Table 2), the others were mostly Sergentomyia spp. and a very small number of Phlebotomus that matched with < 95% identity in GenBank with different Phlebotomus species. The overall species composition did not differ between months (χ62=1. 03, p = 0. 598). Of 8410 females, 1065 were excluded for parasite screening, because of contamination or inhibition prior to or during DNA extraction (S1 Table). The reactions were determined as invalid, because positive and negative extraction controls did not give correct results. The overall percentage of sandflies positive for kDNA was 2. 6% (187/7345). Assuming an overall proportion of 73. 3% P. pedifer in the total sandfly population (Table 2), the proportion of the kDNA positive P. pedifer is 3. 5% (187/5384). Of the 187 kDNA positive samples, 162 were positive for ITS-1 (86. 6%, S3 Fig), of which 58 samples were successfully sequenced (35. 8%). The established species was L. aethiopica in all cases. The COI sequence of some morphologically identified P. pedifer sandflies was used as a reference sequence to align the obtained sequences with. Of the 187 kDNA positive samples, 148 were positive for COI (79. 1%, S4 Fig), of which 147 samples were successfully sequenced (99. 3%). All of the kDNA positive sandflies were identified as P. pedifer. An overview was made for the overall monthly number of captured sandflies and proportion of kDNA positive females (Fig 2A and 2B). Fluctuations in the abundance of sandflies ranged between 779 and 2498 sandflies per month and was significantly different over the months (F11,160 = 6. 99, p < 0. 001). The sandfly abundance was higher in the recorded dry season, with peaks in January, February and April. The population was lower from June to October, with the lowest point in August, which corresponds with the documented rainy season. Since we found no difference in the proportion of P. pedifer between the extreme months (Table 2), we assume that the seasonal patterns for the total sandfly population can be taken as a proxy for those in P. pedifer alone. The kDNA prevalence among female sandflies varied significantly over the months (χ112=29. 73, p = 0. 002), fluctuating from 0. 5% to 5. 0%, however the Tukey test could not show pairwise differences between months. There were peaks in kDNA prevalence rates in July and August and November. Similarly, the number of infected sandflies was highest in July and November, but also in February a relatively high number of infected sandflies was obtained, while the prevalence rate was rather low. A steep decline was observed in September, in both absolute numbers and prevalence. However, only 34% of the specimens from this month were successfully tested. The monthly temperature and % RH were recorded in Ochollo from March 2017 to February 2018 (Fig 2C and S1 Table). Average values were used, since the recordings of different habitats were indistinguishable. The mean monthly temperature ranged between 17. 0°C and 22. 6°C over the year, while the % RH stretched between 40. 2% and 87. 3%. Mean temperature and % RH had an inverse pattern. The mean % RH rose from March to August and dropped from August to February. It showed a steep decline from November to February and had its lowest value in February, while the sandfly abundance peaked in January and February (2498 and 2439 sandflies respectively). The temperature fell from March to August and from then on increased again. It dropped to its lowest recordings in July and August (17. 1°C and 17. 0°C), while the prevalence then peaked highest (4. 8% and 5. 0%). In September and October, the temperature increased (21. 0°C and 21. 1°C) and simultaneously the prevalence decreased (0. 5% and 1. 6%). In November, a slight drop in temperature (about 2°C) was recorded and the number of positive sandflies increased from eight in October to 27 in November. The months that were recorded to have had rainfall during the sandfly collection were May, July, August, September and October. The correlations between the mean temperature and % RH, and the average monthly number of sandflies and kDNA prevalence were obtained by GLMMs and GLMs respectively. Temperature did not have a correlation with the sandfly abundance (F1,217 = 0. 37, p = 0. 955), but the % RH showed a significant negative correlation (F1,217 = 5. 95, p = 0. 015). There was no interaction effect, meaning that the type of habitat (cave, stone fence or rocky area) did not influence the association between temperature and % RH and the population. Temperature was negatively correlated with kDNA prevalence (χ12=3. 91, p = 0. 048), while the % RH showed a borderline non-significant positive correlation with the kDNA prevalence (χ12=3. 79, p = 0. 051). The type of habitat did not influence the association between temperature and the kDNA prevalence, but there was an interaction effect between the type of habitat and the mean % RH (χ22=6. 87, p = 0. 032). The average monthly sandfly abundance at each habitat type is presented on a logarithmic scale (Fig 3, left panel). Sandflies were present in each habitat during the whole year. The monthly variation of the sandfly abundance was significantly different at each habitat type (F22,160 = 2. 40, p < 0. 001), as shown by the crossing lines and different peaks and drops for each habitat type, indicating that the sandfly population was not predominantly present at one of the three habitats. From a total number of 400 sandflies, 200 from July 2017 and 200 from January 2018, the number and proportion of each sandfly species in the different habitats are presented in Table 2. The three-way interaction of the model on the selection of 400 sandflies of which the species was determined, indicated that the abundance of sandflies in the different habitats was significantly different in July 2017 and January 2018 (χ02=23. 62, p < 0. 001). Only in caves, the population mainly consisted of P. pedifer in July (93/118,78. 8%) and January (117/124,94. 4%), while raw numbers and proportions of P. pedifer were lower at rocky areas and stone fences. The monthly kDNA prevalence among tested females (Fig 3, right panel) followed a similar pattern at each habitat type (χ222=32. 04, p = 0. 077) and the habitat had a significant correlation with the prevalence (χ22=55. 01, p < 0. 001). Throughout the whole year, infected sandflies were present at caves, in contrast with the stone fences and rocky areas, where there were several months without infected sandflies. At all three habitats, a peak in prevalence occurred in July or August. At stone fences and rocky areas, the maximum prevalence was reached in November. Fig 4 presents which habitat was preferred by (infected) sandflies. According to the Tukey test, caves (4. 27%) had a significantly higher prevalence than stone fences (0. 97%, p < 0. 001) and rocky areas (1. 53%, p-value <0. 001), which is also reflected by the black curve (Fig 3 left panel) that is, except for November, consistently higher than the other two. Raw numbers are presented at the bottom of S1 Table. We demonstrated that a high proportion of H. brucei, which are abundant in Ochollo, were asymptomatically infected with L. aethiopica, confirming the results of Ashford et al. and Lemma et al. [8,12]. Similarly, Procavia spp. are described to carry L. tropica parasites in Ethiopia [12,15], Kenya [17] and other African countries [19–22], with prevalence ranging between 3. 5% and 27% with seasonal variations. However, since the nomenclature L. aethiopica was only named as such in 1973, the aetiology of CL in Ethiopia was believed to be L. tropica, implying that the infections found in Ethiopia were in fact most probably due to L. aethiopica [9]. The high abundance and infectivity rate of hyraxes in Ochollo, in combination with the fact that they can become up to 13 years old, suggests that hyraxes might play a considerable role in the transmission of zoonotic CL, as suggested already 45 years ago. Yet, it remains to be evaluated how long the parasite can be sustained in a hyrax and how efficient it can be transmitted to P. pedifer [44]. Our study also assessed whether other small mammals could contribute to the transmission of CL. Only a single Mus mahomet was found kDNA positive, indicating that the captured rodent species do not play a major role as a source of infection in southern Ethiopia. Abebe et al. found one ground squirrel (Xerus rutilus) naturally infected with L. aethiopica in Aba Roba (1200 m), a VL endemic area, where there has never been a human case of CL [45]. Except for this report, to our knowledge L. aethiopica has not been found in rodents [8,12,46,47]. In contrast, studies carried out in different areas in Ethiopia found L. tropica in Arvicanthis sp. , Gerbillus nanus and Acomys spp. [13,28] and L. major in Arvicanthis niloticus [48]. We did not obtain enough samples from bats and shrews to draw any meaningful conclusions from this study. It would be interesting to acquire knowledge about the blood meal sources of P. pedifer in the area. If particular rodent species or other mammals appear to be a dominant blood source, new work could target these species. All dry blood spots samples were negative for kDNA, suggesting that these samples should not be used for molecular detection of L. aethiopica. Based on a selected subsample of 400 sandflies, it can be concluded that P. pedifer (73. 3%) is the predominant sandfly species in Ochollo. The rest of the sandflies mainly belong to species in the subgenus Sergentomyia. Only 1. 5% of the sandflies in our traps were other Phlebotomus species. The exact species could not be determined, because sequences were not available in GenBank, but most probably it was P. ashfordi [33,34]. Overall, 2. 6% of the sandflies (or an estimated 3. 5% of P. pedifer) captured between March 2017 and February 2018 in Ochollo were positive for kDNA. CL has been described as a disease of childhood in Ochollo, as about 65% of primary school children had scars or active lesions [7]. The infection at young age might be due to the children’s or sandflies’ behavior, but it could also be explained by the high prevalence and sandfly abundance, which indicate that there is a high intensity of Leishmania transmission, increasing the risk of early exposure to an infectious sandfly. Studies on the behavior of children and indoor/outdoor biting behavior of sandflies should be carried out to find out where and when transmission takes place. Two prior studies carried out in Ochollo described that promastigotes were found in 5. 4% (2/37) [12] and 1. 67% (5/359) [33] of the dissected sandflies. They only dissected a small number of P. pedifer females, in contrast with the current research, where all sandflies (Phlebotomus and Sergentomyia) were tested. The former two studies proved that P. pedifer is a vector for Leishmania by dissecting the midgut of the sandflies, while this study intended to evaluate whether other sandfly species could also carry Leishmania parasites. Therefore, we opted to screen a large sample size with a real-time PCR targeting kDNA. Leishmania parasites have a concatenated network of kDNA minicircles, which are highly abundant, making it a very sensitive fragment to target [49]. P. pedifer was the only infected sandfly species in Ochollo carrying exclusively L. aethiopica parasites. There is very limited knowledge about this vector, although this is a prerequisite for efficient vector control strategies. To know when and where to deal with the vector population, the seasonality and habitat preference were determined. The relative humidity seemed to be a good indicator for rainfall, as rainfall on the days of sampling was recorded in most months with a high relative humidity. The seasons did not appear like previously described for this area [31], confirming that seasons are changing, as mentioned by inhabitants. Our data illustrate that the sandfly abundance and Leishmania infection in Ochollo were present throughout the whole year and exhibit a distinct seasonality. Since sandflies were not collected in the same number of caves, stone fences and rocky areas every month, this might give a slightly biased representation of the monthly abundance of sandflies. The % RH was found negatively correlated with the sandfly abundance, and since the % RH is a good proxy for rainfall, there is a decline in the sandfly abundance when the rainfall increases. This was reflected by the sandfly abundance, which peaked during the dry months and dropped in the rainy period. With P. pedifer accounting for nearly 75% of all sandflies, in the wet as well as in the dry season, it can be stated that the species’ population decreases with increasing rainfall. A study in the Mt. Elgon region in Kenya by Mukhwana et al. observed the same fluctuation in the abundance of P. pedifer, showing a peak in the two dry seasons and a drop in the two wet seasons [50]. Although that study was only based on 657 sandflies, it supports the current findings. No other studies were carried out on seasonality in areas where P. pedifer is responsible for CL transmission. P. longipes is the other main vector for L. aethiopica in Ethiopia and shows morphological and ecological similarities with P. pedifer [9,51]. However, contradictory to the obtained results for P. pedifer, an increase in rainfall was accompanied with a rise in the P. longipes population in a similar study conducted in Kutaber [12]. Increasing temperatures appeared to be accompanied with a drop in the number and proportion of kDNA positive sandflies, but not as distinct as the correlation between temperature and sandfly abundance. The low prevalence rate in September should be interpreted with caution. Due to inhibition, only 34% of the female sandflies caught in September were successfully tested, which might slightly lower the strength of the prevalence estimation within this month. Studies have been done to investigate the effect of temperature on the development of Leishmania parasites in sandflies [52], as well as the metabolism of the sandflies [53]. Metabolic processes are slower at lower temperatures, so there is a delay in defecation when temperatures are lower, providing more time for Leishmania to establish an infection in the midgut [52,53]. Contradictory to our results, the seasonality of P. pedifer infected with L. aethiopica in the Mt. Elgon region in Kenya showed two drops, concurrent with the rainy seasons [50]. However, based on the tables and figures provided in the paper, it is unclear how the presented results were calculated: only 21 females were infected with promastigotes, yet four trapping sites over a period of 12 months were described, resulting in 48 separate prevalence values. None of them were zero, raising doubts about the presented results. In Kutaber, the infection of promastigotes in P. longipes varied considerably over 14 months, but seemed to be independent of seasonal conditions [12]. Insight in the spatial abundance of the vector is important for potential future outdoor vector control strategies. The overall sandfly population in Ochollo was not predominantly abundant at a particular habitat, though, hyrax feces suggested to serve as larval food, were only present at caves and rocky areas. It was remarkable that in caves, on average almost 87% of the sandflies were P. pedifer, while at stone fences and rocky areas, which are situated closer to the people’s houses, only half or even less than half of the population was P. pedifer. Other papers describe without statistical analysis that the preferred habitat for P. pedifer and P. longipes are cracks in basalt cliffs (here referred to as caves) [12,33,54,55]. The infected sandfly population was evidently more present at caves, where they were found throughout the whole year, while stone fences and rocky areas had several months without kDNA positive sandflies, implying that caves could be the source of multiplication of the infection. A study in Mt Elgon Region in Kenya observed that CL cases as well as P. pedifer were mainly found near caves and concluded from this that human infection with L. aethiopica by P. pedifer is happening near caves. It must be mentioned though, that no other possible habitats of P. pedifer were considered in that study [55]. At first sight, there seemed to be more CL cases close to the caves in Ochollo, but more research is required to investigate where the transmission particularly happens. Data on vectors and reservoirs of CL in southern Ethiopia are very limited. Until now, there are no efficient intervention programs, partially due to a potentially important zoonotic component and ecological factors associated with the disease that are left aside. Moreover, CL is moving towards new areas with susceptible people, because people are building settlements and cultivate closer to habitats of hyraxes and sandflies [8]. Our results can provide guidance for disease management programs in areas in southern Ethiopia and Kenya with a similar ecological context. Although there is still need to investigate potential indoor transmission, if outdoor vector control would be considered, it would be a good idea to focus on caves in the beginning of the wet season, when the population is at its minimum. The role of hyraxes in CL transmission should be further investigated to assess whether they should be included in control programs. Destroying hyrax habitats close to human settlements is almost impossible in Ochollo, since the majority of the houses are surrounded by basalt cliffs in a range of 200 m. Shooting hyraxes or biological control of the population is believed not to be effective and might result in an increase in human-vector contact [4,8]. Rather than focusing on the hyraxes, attractive toxic sugar treated barrier fences around caves could be used to prevent sandflies from going towards human dwellings to acquire a blood meal [56]. Also, people should be made aware of the risk they face when building settlements or residing in the proximity of caves [57]. Additional information is required about the biting behavior of P. pedifer to decide whether indoor residual spraying and insecticide impregnated bed nets could be beneficial.

Zoonotic cutaneous leishmaniasis poses a considerable health problem in Ethiopia. Efficient disease control can only be accomplished when all players of transmission are well understood and taken into account. The aim of our study was to investigate in a village in the south, called Ochollo, whether rodents also harbor the parasite and to assess the different potential vector species (sandflies) and their spatial and seasonal distribution. We established that the rodent species we captured are probably no hosts for Leishmania aethiopica, but confirmed that hyraxes are abundant and that a high percentage of them is positive for L. aethiopica. Based on a very large sample size, we found that Phlebotomus pedifer is the only vector in the area. We discovered that the general sandfly and specific P. pedifer abundance are lowest in the wet season and negatively correlated with humidity. We also demonstrated that the sandfly abundance is equally distributed among different habitats, but P. pedifer and infected sandflies mainly reside in caves. Altogether, we suggest that if outdoor sandfly control methods would be considered, it would be useful to carry it out in the wet season in or around caves. The role of hyraxes in disease transmission should be further investigated.

Abstract Introduction Methods Results Discussion

geomorphology invertebrates medicine and health sciences landforms hyraxes topography vertebrates sand flies parasitic diseases animals mammals parasitic protozoans diptera protozoans leishmania insect vectors cellular structures and organelles infectious diseases kinetoplasts phlebotomus disease vectors insects arthropoda caves rodents eukaryota cell biology earth sciences biology and life sciences species interactions amniotes organisms

2019

Ecology and seasonality of sandflies and potential reservoirs of cutaneous leishmaniasis in Ochollo, a hotspot in southern Ethiopia

Functional brain networks detected in task-free (“resting-state”) functional magnetic resonance imaging (fMRI) have a small-world architecture that reflects a robust functional organization of the brain. Here, we examined whether this functional organization is disrupted in Alzheimer' s disease (AD). Task-free fMRI data from 21 AD subjects and 18 age-matched controls were obtained. Wavelet analysis was applied to the fMRI data to compute frequency-dependent correlation matrices. Correlation matrices were thresholded to create 90-node undirected-graphs of functional brain networks. Small-world metrics (characteristic path length and clustering coefficient) were computed using graph analytical methods. In the low frequency interval 0. 01 to 0. 05 Hz, functional brain networks in controls showed small-world organization of brain activity, characterized by a high clustering coefficient and a low characteristic path length. In contrast, functional brain networks in AD showed loss of small-world properties, characterized by a significantly lower clustering coefficient (p<0. 01), indicative of disrupted local connectivity. Clustering coefficients for the left and right hippocampus were significantly lower (p<0. 01) in the AD group compared to the control group. Furthermore, the clustering coefficient distinguished AD participants from the controls with a sensitivity of 72% and specificity of 78%. Our study provides new evidence that there is disrupted organization of functional brain networks in AD. Small-world metrics can characterize the functional organization of the brain in AD, and our findings further suggest that these network measures may be useful as an imaging-based biomarker to distinguish AD from healthy aging. Alzheimer' s disease (AD) is a neurodegenerative disorder characterized by progressive impairment of episodic memory and other cognitive domains resulting in dementia and, ultimately, death. Imaging studies in AD have begun a shift from studies of brain structure [1], [2] to more recent studies highlighting focal regions of abnormal brain function [3]–[6]. Most recently, fMRI studies have moved beyond focal activation abnormalities to dysfunctional brain connectivity. Functional connectivity is defined as temporal correlations between spatially distinct brain regions [7]. PET studies, restricted to across-subject connectivity measures, have shown that AD patients have decreased hippocampus connectivity with prefrontal cortex [8] and posterior cingulate cortex [9] during memory tasks. Using fMRI, we demonstrated that AD patients performing a simple motor task had reduced intra-subject functional connectivity within a network of brain regions—termed the default-mode network—that includes posterior cingulate cortex, temporoparietal junction, and hippocampus [10]. Bokde et al. reported abnormalities in fusiform gyrus connectivity during a face-matching task in subjects with mild cognitive impairment—frequently a precursor to AD [11]. Three recent studies have reported reduced default-mode network deactivation in MCI and/or AD patients during encoding tasks [12], [13] and during a semantic classification task [14]. Celone et al also reported increased default-mode network deactivation in a subset of “less impaired” MCI patients. In addition to analyzing functional connectivity during task performance, functional connectivity has also been investigated during task-free (“resting-state”) conditions. Task-free functional connectivity MRI detects interregional correlations in spontaneous blood oxygen level-dependent (BOLD) signal fluctuations [15]. Using this approach, Wang et al. found disrupted functional connectivity between hippocampus and several neocortical regions in AD [16]. Similarly, Li et al. reported reduced intrahippocampal connectivity during task-free conditions [17]. Most recently Sorg et al. [18] reported reduced resting-state functional connectivity in the default-mode network of MCI patients. Although evidence is accumulating that AD disrupts functional connections between brain regions [19], it is not clear whether AD disrupts global functional brain organization. Graph metrics–the clustering coefficient and the characteristic path length—are useful measures of global organization of large-scale networks [20]. Graphs are data structures which have nodes and edges between the nodes. The clustering coefficient is a measure of local network connectivity. A network with a high average clustering coefficient is characterized by densely connected local clusters. The characteristic path length is a measure of how well connected a network is. A network with a low characteristic path length is characterized by short distances between any two nodes. Small-world network is characterized by a high clustering coefficient and a low characteristic path length [20], [21]. In a graphical representation of a brain network, a node corresponds to a brain region while an edge corresponds to the functional interaction between two brain regions. Functional connectivity networks of the human brain derived from electroencephalograms (EEGs), magnetoencephalograms and task-free fMRI data exhibit small-world characteristics [22]–[24]. In a recent EEG study, Stam et al. reported that small-world architecture in functional networks in the brain is disrupted in AD [25]. Here we examined the global functional organization of the brain in AD by (1) creating whole-brain functional connectivity networks from task-free fMRI data, (2) characterizing the organization of these networks using small-world metrics, and (3) comparing these characteristics between AD patients and age-matched controls. We hypothesized that global functional brain organization would be abnormal in AD. Further, given the need for a reliable, non-invasive clinical test for AD [26], we sought to determine whether a small-world metric obtained from task-free fMRI data might provide a sensitive and specific biomarker in AD. Demographic data is shown in Table 1. Subject groups did not differ significantly in age (p = 0. 73), gender distribution (p = 0. 62), or years of education (p = 0. 58). The mean MMSE was significantly lower (p<0. 0001) for the AD group (22. 14) compared to the controls (29). We first examined graph metrics obtained for the functional brain networks constructed by thresholding (threshold values ranged from 0. 01 to 0. 99 with an increment of 0. 01) the wavelet correlation matrix computed at three scales (frequencies in the range from 0. 01 to 0. 25 Hz) for the AD group and the control group (see Figure 1). For both groups, the mean degree was highest at Scale 3 for a wide range of correlation thresholds (0. 01<R<0. 7). The mean characteristic path length (λ) for both groups, when controlled for the degree of the network, was low (1<λ<1. 27) and showed similar trends at all the scales. The clustering coefficient (γ) for both groups, when controlled for the degree of the network, was highest at Scale 3. Due to higher mean γ values, the small-world measure σ (γ/λ), when controlled for degree of the network, was highest at Scale 3 for both groups. The small-world property (σ>1) showed a linear increase in small-worldness as the threshold increased (degree decreased). σ values for higher correlation thresholds are difficult to interpret, as at higher threshold values, graphs of functional brain networks have fewer edges (smaller degree) and tend to split into isolated sub-graphs. Graph metrics such as clustering coefficient, characteristic path length, and small-world property do not meaningfully characterize network structures that are not composed of a single, large group of interconnected nodes [20]. Since functional connectivity and small-world properties were salient at lower-frequencies (0. 01 to 0. 05 Hz) for the AD group and the control group, we only report results for this frequency interval in subsequent analyses. In the frequency interval between 0. 01 to 0. 05 Hz, we examined λ and γ values in the two groups. For group comparison, we controlled for the average correlation value (r). r is different across groups. Thus, for a given correlation threshold, the number of edges in the graph are likely to be less in AD, resulting in high λ and low γ values. To ensure that graphs in both groups had the same number of edges, individual correlation matrices were thresholded such that the resultant graph had exactly K′ edges. K′ is the average number of edges in the graph obtained by thresholding individual correlation matrices with R = ri (ri is the average correlation value for subject i, i = 1 to 39). The value of K′ selected according to this procedure was 40 for both the groups. Mean λ, mean γ, and mean σ values for the networks of the AD group and control group were derived by thresholding the correlation matrices such that the network has K′ (= 40) edges (shown in Figure 2). Results were: (i) No significant differences in the mean λ values were observed, Mean γ values in the AD group were significantly lower than in the control group (p<0. 01), and (iii) Mean σ values in the AD group were significantly lower than in the control group (p<0. 01). We examined global efficiency (Eglobal) values obtained for the functional brain networks constructed by thresholding (threshold values ranged from 0. 01 to 0. 99 with an increment of 0. 01) the wavelet correlation matrix computed at three scales (frequencies in the range from 0. 01 to 0. 25 Hz) for the AD group and the control group (see Figure 3A). The mean Eglobal for both groups, when controlled for the degree of the network, was low (0. 77<Eglobal<1) and showed similar trends at all the scales. In the frequency interval 0. 01 to 0. 05 Hz (scale 3), mean Eglobal values for the AD group and the control group for the networks derived by thresholding the correlation matrices such that the network has K′ (= 40) edges are shown in Figure 3B. No significant differences in the mean Eglobal values were observed. Here, we examined whether γ (normalized clustering coefficient) might prove sufficiently sensitive and specific to serve as a biomarker for AD. Using the cut-off value (γ = 1. 57) that maximizes sensitivity and specificity, γ correctly classified 14 out of 18 controls and 15 of 21 AD subjects, yielding 72% sensitivity and 78% specificity respectively. A receiver operating characteristic curve for various cut-off values is shown in Figure 4. The Area Under the Curve for the ROC was 0. 754 (95% CI Area 0. 602 to 0. 906). Figure 5 shows a plot of γ for each of the four regions, for the AD group and the control group as a function of the correlation threshold. In the left and the right hippocampus, the fitted growth curve was significantly lower (p<0. 01) in the AD group, compared to the control group, reflecting lower clustering coefficient values for a range of threshold values from 0. 1 to 0. 6. A similar analysis in the left and right precentral gyrus, revealed no significant differences in the clustering coefficient values. Across the four regions, no significant differences in the clustering coefficient values were observed for correlation threshold values >0. 6, mainly due to the large variance observed at higher threshold values. This analysis was extended to the remaining 86 regions of the whole brain functional network (see Table S1 to find regions that showed significant differences in clustering coefficient values between the two subject groups). To determine whether the differences observed in γ values reflect true differences and not artifacts of different average correlation values, we repeated our analysis by computing γ values as a function of the number of edges in the graph. Mean γ values of four anatomical regions of interest for the AD group and the control group for networks derived by thresholding the individual correlation matrices such that the network has K′ edges were computed and compared. Results were consistent with the initial analysis–significantly lower clustering coefficient values (p<0. 01) in the left and right hippocampus in AD, and no significant differences in the left and right precentral gyrus. We next examined regional correlation values (connectivity) in the two groups. Results show that compared to the control group, the AD group had decreased correlation values (1) within the temporal lobe, (2) between the temporal lobe and thalamus, (3) between the temporal lobe and corpus striatum, (4) between the thalamus and occipital lobe, and (5) between the thalamus and other parts of the frontal lobe, but increased correlations (1) within the prefrontal areas, (2) within other parts of frontal lobe, (3) between the prefrontal areas and other parts of the frontal lobe, and (4) between other parts of frontal lobe and the corpus striatum. To determine if our findings were robust–reproducible across datasets–we repeated our entire analysis on a second resting-state fMRI dataset (rest2 scans) acquired from the same set of subjects. Results were consistent with previous analysis (performed on data from rest1 scan): (i) Functional brain connectivity and small-world metrics including the global efficiency were salient in the low frequency interval–0. 01 to 0. 05 Hz (Scale 3), (ii) No significant differences in the mean λ values were observed, (iii) Mean γ values in the AD group were significantly lower than in the control group (p<0. 01), (iv) Mean σ values in the AD group were significantly lower than that in the control group (p<0. 01), (v) No significant differences in the mean Eglobal values were observed, and (vi) significantly lower clustering coefficient values were found in the left and right hippocampus in AD, with no significant differences in the left and right precentral gyrus. In this study, we investigated whether global functional brain organization is disrupted in AD. To our knowledge, this is the first study to examine alterations in global functional organization and connectivity in AD patients using fMRI data. Graph metrics–clustering coefficient and characteristic path length—were used to measure and characterize global functional organization in the brain. The main finding of our study is that functional brain networks in AD consistently showed lower clustering but similar characteristic path lengths compared to controls, which suggests disrupted global functional organization in AD. Our findings also suggest that small-world network characteristics might be useful as an imaging biomarker for AD. The characteristic path lengths were low (λ∼1) and showed no significant differences between the AD group and the control group, suggesting short distances between distinct brain regions in both groups. This finding suggests an organization consisting of multiple short alternative paths between nodes in functional brain networks in both groups. The most interesting finding of our study was the lower levels of clustering observed in the AD group. Clustering coefficient is a measure of local efficiency or the fault-tolerance of a network [21]. The difference in clustering coefficients in the AD group as compared to the control group was observed at a correlation threshold at or near a subject' s average correlation (to ensure an equivalent number of edges across subjects), and the clustering coefficient was significantly lower in the AD group, suggesting loss of local efficiency in AD. Similarly, values for σ, a measure of small-worldness, were significantly lower in the AD group compared to the control group, suggesting loss of small-world properties in AD. Analysis of global efficiency in functional brain networks showed that the networks exhibit small-world properties indicated by smaller Eglobal values compared to random networks, but this measure was not significantly different. This finding parallels results obtained with measures of characteristic path length. Regional analysis of differences in clustering coefficients as a function of correlation thresholds showed that the left and the right hippocampal regions differed significantly between groups. In contrast, the clustering coefficient of the precentral gyrus did not differ between groups. This suggests disrupted connectivity from the hippocampus to other regions of the brain in AD. This finding is consistent with our previous study [10] showing that AD reduced functional connectivity of the hippocampus within a specific network of regions—the default mode network [27], [28] that includes the posterior cingulate and lateral temporoparietal cortices. It is also consistent with the study by Wang et al. [16] showing altered hippocampal connectivity to several neocortical regions in the early stages of AD. Other studies have reported decreased intrahippocampal synchrony of low frequency BOLD fluctuations [17] during a task-free scan. Taken together, these findings point to significantly altered local and global hippocampal network connectivity in AD. Analysis of the group differences in the regional connectivity across several broadly defined anatomical regions demonstrate that AD patients not only showed decreased intratemporal, temporo-thalamus, temporo-corpus striatum, thalamo-occipital and thalamo-frontal connectivity but, surprisingly, also showed increased intrafrontal, frontal-prefrontal, and fronto-corpus striatum connectivity. These findings are in line with the recent study by Wang et al. [29] which not only reported decreased connectivity between a number of regions, but also increased prefrontal connectivity in AD. As suggested by fMRI studies showing increased prefrontal activation in AD during task performance [30], these findings suggest that patients with AD may rely on increased prefrontal connectivity to compensate for reduced temporal connectivity. An intriguing (and testable) hypothesis is that the ability to make such compensatory changes in frontal lobe connectivity may account in part for the “cognitive reserve” phenomenon [31] that allows some patients to perform better than others despite equivalent pathological burdens. Small-world characterization is well-suited for analyzing anatomical and functional brain networks at the system level because these networks are complex and optimally connected to minimize information processing costs [32], [33]. Anatomical connectivity networks of the brain obtained from tracer studies in the primate cortical visual system [34], primate cerebral cortex [35], and macaque cortex [36] have been shown to exhibit small-world characteristics. Functional connectivity networks of human brain constructed from EEG as well as MEG data have also been shown to have small-world architecture [22], [23]. Salvador et al. [37] built a whole-brain functional connectivity network from task-free human functional MRI data. This network of intrinsic, task-free functional interactions between 90 cortical regions was also shown to have small-world properties–high clustering coefficient and low characteristic path length. The small-world architecture was confirmed by Achard et al. , who also reported that the small-world properties were salient in the frequency interval 0. 03 to 0. 06 Hz [24], [32]. These findings suggest that the structural and functional organization of the brain has a small-world architecture; these characteristics may assist in robust and dynamic information processing. Recently, Stam et al [25]. reported that the architecture of whole-brain functional networks derived using scalp EEG is disrupted in AD. They observed that a 21-node network constructed using EEG data collected from subjects with AD showed loss of small-world properties characterized by longer characteristic path length with relative sparing of the local clustering. Table 2 provides a comparison of results obtained from our study to all of the above-mentioned results on the small-world characterization of functional brain networks. Our results are largely comparable to small-world metrics reported by Salvador et al. also using task-free fMRI in healthy human subjects [24], [37]. The small-world metrics reported by Stam et al. analyzing beta-band EEG in controls and AD subjects are also largely consistent with our results [25]. It is interesting to note that whereas we report similar characteristic path lengths but different cluster coefficients between AD and controls, the EEG study found the converse (characteristic path lengths differed between AD subjects and controls but cluster coefficients did not). We believe that this discrepancy may be related to significant volume conduction in scalp EEG data [38] which may reduce sensitivity to detect differences in short-range connectivity while enhancing the relative sensitivity to detect differences in long-range connectivity. Other methodological differences may also contribute–the use of synchronization likelihood as their association measure, which unlike wavelet correlation is sensitive to non-linear coupling. Also, the poor spatial resolution of scalp EEG limits the network to mainly cortical regions, unlike our fMRI study where the network comprised of cortical as well as sub-cortical regions, which is a relative strength of our study. To address the extent to which clustering coefficients serve as a sensitive biomarker to distinguish AD from healthy aging, we examined γ values in the two subject groups. The clustering coefficient is a measure of efficiency in network connectivity. It distinguished AD subjects from controls with a sensitivity of 72% and specificity of 78%. These values approach the sensitivity and specificity reported for other imaging biomarkers [10], [39]–[41] and are close to the range considered clinically relevant by a recent Working Group on biomarkers in AD [42]. With some improvements in the technique—decreasing the number of nodes in the network for example—the clustering coefficient may therefore prove to be an effective biomarker for AD, though prospective studies will be required to validate its effectiveness. In addition to its promise as a diagnostic aid, the clustering coefficient merits investigation as a functional marker of response to treatment. This study has two main limitations. First, in evaluating its efficacy as a biomarker, it will be critical to assess this metric not only in AD and normal subjects, but in subjects with non-AD dementias and related conditions to ensure that these findings are specific to AD and not to dementia or other neurodegenerative disorders more generally. The second limitation pertains to the fact that most of the AD patients (14 of 21), and none of the controls, were taking an acetylcholinesterase inhibitor. Similarly, 12 of 21 AD patients, and none of the controls, were taking memantine, an NMDA-receptor antagonist. While we doubt that these differences in medication exposure could account for the differences in clustering coefficients in AD subjects we cannot exclude that possibility in the current study. In conclusion, we have demonstrated that fMRI-derived functional brain networks in AD show loss of small-world properties. Our findings suggest that cognitive decline in AD is associated with disrupted global functional organization in the brain. Twenty one subjects with AD and eighteen age-matched control subjects participated in this study after giving written, informed consent. For those AD patients who were unable to give informed consent, written, informed consent was obtained from their legal guardian. The study protocol was approved by the Stanford University Institutional Review Board. The AD subjects (10 males, 11 females) ranged in age from 48 to 83 (mean age 63. 97) with 12 to 22 years of education (mean years of education 15. 89). The subjects were recruited from memory disorder clinics in Stanford University and the University of California San Francisco (UCSF). All AD subjects met the NINDS-ADRDA criteria for probable AD [43]. One subject had a presenilin-1 mutation; a second subject' s mother had a presenilin-2 mutation (the subject herself did not wish to be tested). Diagnosis of three other subjects has since been confirmed at autopsy. ApoE status was known for 4 additional AD subjects: one was homozygous for the ApoE 4 allele and 3 were heterozygous for the ApoE 4 allele. The control subjects (10 males, 8 females) ranged in age from 37 to 77 (mean age 62. 84) with 12 to 21 years of education (mean years of education 16. 53). Study subjects were recruited from several sources (partners of AD patients, participants in a longitudinal study of normal aging at UCSF, and Stanford research staff). Control subjects denied any significant neuropsychiatric disease or memory trouble, were not taking any psychoactive medicines, and had to have a Mini Mental State Examination (MMSE) score of 27 or more. 14 of 21 AD patients were taking an acetylcholinesterase inhibitor. And, 12 of 21 AD patients were taking memantine, an NMDA-receptor antagonist. The MMSE score of the AD group ranged from 12 to 29 (mean MMSE score 22. 14) and the MMSE score of the control group ranged from 27 to 30 (mean MMSE score 29). Each subject underwent an MMSE, a structural MRI scan, and a task-free fMRI scan. For the task-free scan, subjects were instructed to keep their eyes closed and try not to move. The scan lasted for 6 minutes (rest1 scan). All the subjects (except for one control subject and two AD subjects) also underwent another task-free scan that lasted for 6 minutes (rest2 scan) and was acquired immediately after the first task-free scan. Functional images were acquired on a 3-T General Electric Signa scanner using a standard whole-head coil. Twenty-eight axial slices (4 mm thick, 1mm skip) were acquired parallel to the plane connecting the anterior and posterior commissures and covering the whole brain using a T2* weighted gradient echo spiral in/out pulse sequence (TR = 2000 msec, TE = 30 msec, flip angle = 80° and 1 interleave) [44]. To aid in the localization of functional data, a high resolution T1-weighted spoiled grass gradient recalled (SPGR) 3D MRI sequence with the following parameters was used: 124 coronal slices 1. 5 mm thickness, no skip, TR = 11 ms, TE = 2 ms, and flip angle = 15°. Data (rest1 and rest2 scans) were preprocessed using statistical parametric mapping (SPM2) software (http: //fil. ion. ucl. ac. uk/spm). The first 8 image acquisitions of the task-free functional time series were discarded to allow for stabilization of the MR signal. Each of the remaining 172 volumes underwent the following preprocessing steps: realignment, normalization, and smoothing. Normalization was to the Montreal Neurological Institute (MNI) template and smoothing was done with a 4 mm full width half maximum Gaussian kernel to decrease spatial noise. Excessive motion, defined in our lab as greater than 3. 5 mm of translation or 3. 5 degrees of rotation in any plane, was not present in any of the task-free scans. The preprocessed task-free functional MRI datasets were parcellated into 90 regions using anatomical templates defined by Tzourio-Mazoyer et al. [45]. A task-free fMRI timeseries was computed for each of the 90 regions by averaging all voxels within each region at each time point in the time series, resulting in 172 time points for each of the 90 anatomical regions of interest. These regional fMRI time series were then used to construct a 90 node whole-brain task-free functional connectivity network for each subject. Wavelet analysis was used to construct correlation matrices from the regional fMRI time series data. These matrices described frequency-dependent correlations, a measure of functional connectivity, between spatially-distinct brain regions. Correlation matrices were then thresholded to generate a whole-brain functional connectivity network. Wavelets are mathematical functions that transform the input signal into different frequency components [46]. Wavelets are appropriate methods for the analysis of task-based as well as task-free fMRI signal [24], [47]. In our study, we applied a maximum overlap discrete wavelet transform (MODWT) to each of the 90 regional time series from each subject to obtain the contributing signal in the following three frequency components: scale 1 (0. 13 to 0. 25 Hz), scale 2 (0. 06 to 0. 12 Hz), and scale 3 (0. 01 to 0. 05 Hz). To account for a relatively small number (172) of data points per time series for low frequency correlation analysis, the vector representing the time series beyond its boundaries (<0 and >172) was assumed to be a symmetric reflection of itself. At each of the three scales, wavelet correlations between signals in the 90 anatomical regions were determined by computing the correlation coefficient between the transformed signals at that scale. For each subject, a 90-node, scale-specific, undirected graph of the functional connectivity network was constructed by thresholding the wavelet correlation matrix computed at that scale. If the wavelet correlation value between two anatomical regions represented by nodes i and j in the network exceeded a threshold then an edge was drawn between node i and node j. There is currently no formal consensus regarding threshold selection, so we computed networks for threshold values from 0. 01 to 0. 99 with an increment of 0. 01. Once a whole-brain functional connectivity network was constructed from the correlation matrix, we characterized this network in terms of its small-world properties. Small-World properties of a network are described by the clustering coefficient and the characteristic path length of the network. The clustering coefficient and characteristic path length of functional brain networks generated from the task-free fMRI data obtained from 21 AD subjects and 18 age-matched controls were computed. The clustering coefficient of every node was computed as the ratio of the number of connections between its neighbors divided by the maximum possible connections between its neighbors. The clustering coefficient (C) of the network was calculated as the mean of the clustering coefficients of all the nodes in the network. The mean minimum path length of a node was computed as the average of minimum distances from that node to all the remaining nodes in the network. The characteristic path length (L) of the network was the average of the mean minimum path lengths of all the nodes in the network. The clustering coefficient and path length of nodes completely disconnected with the network were set as 0 and Inf respectively, and these nodes were excluded while computing C and L. To evaluate the network for small-world properties, we compared the clustering coefficient and the characteristic path length of the network with corresponding values (Cran, Lran) obtained and averaged across 1000 random networks with the same number of nodes and degree distribution [48]. Degree of a network is a measure of its connectivity. The degree of every node was computed by counting the number of edges incident on that node. Small world networks are characterized by high normalized clustering coefficient γ (C/Cran) >1 and low normalized characteristic path length λ (L/Lran) ∼1 compared to random networks [24]. A cumulative metric σ–the ratio of normalized clustering coefficient (γ) to the characteristic path length (λ), a measure of small-worldness–is thus greater than 1 for small world networks. Small-world networks are characterized by high clustering coefficient and low characteristic path length. These small-world metrics, particularly the path length, are not meaningful when the graph contains disconnected nodes. To address this issue, we ensured that only small-world metrics computed on connected graphs were considered in our analysis. Specifically, the algorithm used to choose the correlation threshold (R) guaranteed that disconnected graphs were excluded from the analysis. Also, in the node-wise clustering coefficient comparison analysis, we only considered thresholds from 0. 1 to 0. 6. We chose these thresholds because beyond 0. 6 the network gets divided into disconnected subset of nodes. To determine if our characteristic path length findings were robust and reliable, we computed efficiency of functional brain networks. It has been previously reported that efficiency as a graph metric (1) is not susceptible to disconnected nodes, (2) is applicable to unweighted as well as weighted graphs, and (3) is a more meaningful measure of parallel information processing than path length [49]. Efficiency of a graph (Eglobal-net) [50] is inverse of the harmonic mean of the minimum path length between each pair of nodes, Lij, and was computed as, (1) To evaluate the network for its global efficiency of parallel information processing, we compared the global efficiency of the network (Eglobal-net) with corresponding values (Eglobal-ran) obtained and averaged across 1000 random networks with the same number of nodes and degree distribution. A network with small-world properties is characterized by global efficiency value that is lower than the random network–Eglobal (Eglobal-net/Eglobal-ran) <1. In the frequency interval 0. 01 to 0. 05 Hz, we next examined small world metric values of four anatomical regions of interest in the two groups. These four regions included the left hippocampus, the right hippocampus, the left precentral gyrus, and the right precentral gyrus. These were chosen because we hypothesized significant differences in the hippocampus (a region targeted early in AD), but not in the precentral gyrus (which is typically spared even in the advanced stages of AD) [51]. This regional profiling analysis was performed on the clustering coefficient (and not the path length) because only the former differed significantly between the AD and control groups. Growth curve modeling, with an intercept (baseline), linear and quadratic terms, was used to compare the clustering coefficient values for threshold values from 0. 1 to 0. 6 in the two subject groups. We chose these thresholds because beyond 0. 6 the network divides into disconnected subsets of nodes and small-world metrics are then no longer meaningful [20]. This analysis was performed using the Mplus software (http: //www. statmodel. com). Growth curve models describe change (growth) with respect to a control variable. They are well-suited for analyzing group-level differences in biomedical data, particularly in cases where capturing and analyzing individual growth trajectories is important. In our study, the growth trajectories of clustering coefficient of a subject carry important information about the variance within the group and needs to be incorporated in the model. The coefficients of growth curve models capture the baseline performance, instantaneous growth rate, and the acceleration of the variable of interest–γ. We then examined regional correlation values (connectivity) in the two groups. Wavelet correlation values of 4005 pairs of anatomical regions were first z-normalized and then compared between the two subject groups. T-test with a false discovery rate of 0. 01 was used to test if the difference was significant. For the frequency range 0. 01 to 0. 05 Hz, the correlation values of 108 pairs of anatomical regions out of a total 4005 pairs were significantly lower in the AD group as compared to the control group while only 42 correlation values showed a significant increase in the AD group (p<0. 01, corrected for multiple comparisons). To get an idea of average differences in the global functional organization in the two groups, we investigated the regional connectivity at a coarser level of granularity. Ninety anatomical regions of our network were grouped into eight higher-level anatomical regions using the grouping defined by Tzourio-Mazoyer et al. [45]. The prefrontal lobe region consists of the superior frontal gyrus (dorsolateral, orbital, medial, medial orbital), the middle frontal gyrus, the middle frontal gyrus (orbital), the inferior frontal gyrus (opercular, triangular, orbital), the olfactory gyrus, the gyrus rectus, and the anterior cingulate. The other parts of frontal lobe region consists of the precentral gyrus, the supplementary motor area, the median cingulate, and the rolandic operculum. The occipital lobe region consists of the calcarine fissure, the cuneus, the lingual gyrus, the superior occipital gyrus, the middle occipital gyrus, and the inferior occipital gyrus. The temporal lobe and the medial temporal region consists of the superior temporal gyrus, the temporal pole (superior, middle), the middle temporal gyrus, the inferior temporal gyrus, the heschl gyrus, the fusiform gyrus, the hippocampus, the parahippocampal gyrus, and the amygdala. The parietal lobe region consists of the postcentral gyrus, the superior parietal lobule, the inferior parietal lobule, the supramarginal gyrus, the angular gyrus, the precuneus, the paracentral lobule, and the posterior cingulate gyrus. The corpus striatum region consists of the caudate nucleus, the putamen, and the pallidum. Each higher level anatomical region consists of regions from both the hemispheres. Differences in mean correlation coefficients for 4005 pairs were aggregated into 32 pairs and the resulting differences were then normalized. (see also [52]). In the aggregation step, the number of decreased (−1) or increased connectivities (+1) for each of the 32 pairs (= (8×8) /2) was counted. For example, to identify differential connectivity between the prefrontal lobe region and the occipital lobe region the number of decreased or increased connectivities between all pairs of sub-regions belonging to the prefrontal lobe region and occipital lobe region was counted. Since each brain region has a different number of sub-regions, the aggregated differential connectivity count was normalized by the number of possible connections between pairs of sub regions belonging to the two brain regions under investigation.

Alzheimer' s disease (AD) is a brain disorder characterized by progressive impairment of episodic memory and other cognitive domains resulting in dementia and, ultimately, death. Functional neuroimaging studies have identified brain regions that show abnormal brain function in AD. Although there is converging evidence about the identity of these regions, it is not clear how this abnormality affects the functional organization of the whole brain. In order to characterize the functional organization of the brain, our approach uses small-world measures, which have also been used to study systems such as social networks and the internet. We use graph analytical methods to compute these measures of functional connectivity brain networks, which are derived from fMRI data obtained from healthy elderly controls and AD patients. The AD patients had significantly lower regional connectivity, and showed disrupted global functional organization, when compared to healthy controls. Moreover, our results indicate that cognitive decline in Alzheimer' s disease patients is associated with disrupted functional connectivity in the entire brain. Our findings further suggest that small-world measures may be useful as an imaging-based biomarker to distinguish AD from healthy aging.

Abstract Introduction Results Discussion Materials and Methods

neurological disorders/alzheimer disease mathematics/statistics computer science/systems and control theory radiology and medical imaging/magnetic resonance imaging

2008

Network Analysis of Intrinsic Functional Brain Connectivity in Alzheimer's Disease

Beriberi occurs in Vientiane, Lao PDR, among breastfed infants. Clinical disease may be the tip of an iceberg with subclinical thiamin deficiency contributing to other illnesses. Thiamin treatment could improve outcome. A cohort of 778 sick infants admitted during one year without clinical evidence of beriberi were studied prospectively and erythrocyte transketolase assays (ETK) performed. Biochemical thiamin deficiency was defined both in terms of the activation coefficient (α>31%) and basal ETK activity <0. 59 micromoles/min/gHb. Of the 778 infants, median (range) age was 5 (0–12) months, 79. 2% were breastfed, 5. 1% had α>31% and 13. 4 % basal ETK<0. 59 micromoles/min/gHb. Infants ≥2 months old had a higher frequency of biochemical markers of thiamin deficiency. Mortality was 5. 5% but, among infants ≥2 months old, mortality was higher in those with basal ETK<0. 59 micromoles/min/gHb (3/47,6. 4%) than in those with basal ETK≥0. 59 micromoles/min/gHb (1/146,0. 7%) (P = 0. 045, relative risk = 9. 32 (95%CI 0. 99 to 87. 5) ). Multivariate regression analysis indicated that infant age ≥2 months and fewer maternal years of schooling were independently associated with infant basal ETK<0. 59 micromoles/min/gHb. Clinically unapparent thiamin deficiency is common among sick infants (≥2 months old) admitted to hospital in Vientiane. This may contribute to mortality and a low clinical threshold for providing thiamin to sick infants may be needed. Beriberi, or clinically apparent thiamin deficiency, may present with a variety of syndromes including peripheral neuropathy, myocardial dysfunction, encephalopathy, hypoglycemia and lactic acidosis. With the advent of mechanical rice milling in the late 19th century, beriberi became a dominant public health problem in Asia, responsible for a considerable mortality, especially amongst infants [1]–[8]. This stimulated a large research effort and identification of the aetiology led to changes in diet, supplementation and targeted health programs to ensure adequate thiamin intake. The incidence of thiamin deficiency in the more accessible, wealthier parts of Asia declined and interest in the disease waned. However, there are recent reports suggesting that it remains an important public health problem in Asia, especially amongst more vulnerable people, such as refugees [8]–[10], the elderly [11] and infants [10], [12], [13]. Treatment with parenteral thiamin is simple, inexpensive and highly effective but supplementation has been much more contentious [14]). In the Lao PDR (Laos), beriberi was noted in 1930 [15] and acknowledged as an important disease in the 1960s and 1970s [16], [17]. It was rediscovered in the early 1990s as an important cause of infant death in the capital, Vientiane [13], [18]. Thiamin deficiency is conventionally assessed using functional assays for the thiamin-dependent erythrocyte transketolase enzyme in washed red cells. The activation coefficient (α) is the ratio of in vitro erythrocyte transketolase activity (ETK) after thiamin pyrophosphate has been added minus the basal ETK before thiamin pyrophosphate has been added, to the basal ETK, expressed as a percentage. Higher α coefficients represent greater degree of thiamin deficiency. However, the α coefficient may not be the appropriate measure of thiamin deficiency in young infants [22] as, with very low exposure to thiamin in utero and post-partum, the capacity of ETK to have its activity augmented may be reduced, underestimating the frequency of biochemical thiamin deficiency [5], [19]–[21]. Indeed, at Mahosot Hospital, Vientiane, Laos basal ETK is a better discriminator between infants with and without clinical beriberi than α and may be the most relevant measure in these chronic deficiency states [22]. Overt infantile beriberi is relatively easy to diagnose but may be the tip of a much larger iceberg of deficiency. A significant proportion of infants admitted with other diseases, such as acute respiratory infection or diarrhoea, may also have clinically unapparent thiamin deficiency contributing to the illness, and thiamin treatment may improve their outcome. An autopsy study 100 years ago demonstrated that many infants with post-mortem evidence for beriberi were misdiagnosed with other conditions [23]. That clinically important but clinically unapparent thiamin deficiency may occur has been suggested in China [19], [24], [25], the Philippines [23], Australia [26], United Kingdom [27], Thailand [28], in Africa [19], [29], Jamaican children with malnutrition [30], the critically ill [31], in psychiatric patients [32] and in the disadvantaged [33] and elderly [34]. In rural southern Laos 30% of older children and adults presenting with malaria had evidence for biochemical thiamin deficiency [35]. Infectious diseases, such as typhoid and malaria, may precipitate beriberi [6], [25], [35], [36]. An increase in body temperature by 1 °C increases basal metabolic rate by 10%, increasing the utilization of thiamin [5]. Therefore, in societies with infantile beriberi, infectious diseases in infants and their mothers may contribute to clinically unapparent but important thiamin deficiency or frank beriberi. We therefore measured the ETK of infants admitted at a hospital in Vientiane to determine what proportion of sick infants without clinical beriberi had biochemical thiamin deficiency. Ethical approval was granted by the Ministry of Health, Laos and the Oxford Tropical Research Ethics Committee. Infants and their mothers were included if the mothers gave witnessed informed oral consent. This was thought to be the most appropriate mode for the Lao situation in 2002 by the two IRBs as the study was a simple observational design with minimal risk. Mothers were given an information sheet describing the study and informed oral consent was documented by the signature of someone who was not a member of the study team. We included all infants who had no overt clinical evidence of beriberi and were admitted to the paediatric service (57 beds) at Mahosot Hospital, a 400 bed primary-tertiary hospital in Vientiane, the capital of Laos. All infants were examined and treated according to hospital guidelines. We did not formally define beriberi but asked physicians to ensure that they only recruited infants whom they believed did not have beriberi. In this hospital paediatricians are very aware of thiamin deficiency and readily treat with thiamin. A 1. 8 ml venous blood sample in lithium heparin was taken from the infant. If beriberi was suspected after recruitment, parenteral thiamin (50 mg) was given. Infants were recruited only on the first admission if they were admitted more than once during the study period. Immediately after collection the lithium heparin anticoagulated blood was centrifuged and washed in phosphate buffered saline three times, with removal of the buffy coat initially and after each wash. Washed red cell samples were stored at −30 °C for a maximum of 3 months and then at −70 °C until shipment to the UK on dry ice. ETK assays were performed at two laboratories. In Oxford (patients 1–394) the assay was performed by a modification of the nicotinamide-adenine dinucleotide dependent method with ribose-5-phosphate as the substrate [36], [37] except that samples were collected into acid citrate dextrose by Krishna et al. [36]. Because of retirement of the machine in the first laboratory, the assay was switched to Cambridge, where an adaptation of the method of Vuilleumier et al. [38] was used (patients 395–778). A Cobas Fara (Roche Co. , Gipf-Oberfrick, Switzerland) was used in both laboratories and at both sites α was calculated as: ( (ETK Activated - ETK basal) /ETK basal) x100. Haemoglobin concentrations were not assayed in Oxford and therefore basal ETK could not be expressed in micromoles/min/gHb for these 394 sample triplets. The method in Oxford measured ETK activity in the presence and absence of TPP [37] whilst that in Cambridge [38] pre-incubated the sample and TPP before the assay reagents are added. The coefficients of variation for the ETK assays in Oxford and Cambridge were 9. 0–10. 0 and 5. 1–7. 7, respectively. Different upper reference ranges for α are described and we have used the more conservative α of >31% [36] as defining definite severe biochemical deficiency. Storage of ex vivo human samples for 18 months at −70 °C does not appear to affect α values [35], [36]. A case control study, also at Mahosot Hospital [22], suggested that the best discriminator between infants with and without clinical beriberi was a basal ETK<0. 59 micromoles/min/gHb. We therefore also compared infants with and without basal ETK<0. 59 micromoles/min/gHb. Data were analyzed using Stata (v10, StataCorp.). Comparisons between 2 groups (α>31 versus α≤ 31 and basal ETK<0. 59 micromoles/min/gHb versus basal ETK≥0. 59 micromoles/min/gHb) were made by the Mann-Whitney U, Student' s t, Pearson' s chi-squared, and Fisher' s exact tests, as appropriate. Because of the multiple comparisons, a Bonferroni adjusted p-value conservatively rounded down to <0. 02 was used. Significant factors from the univariate analysis were then included in two separate multivariate models to identify independent predictors of infantile beriberi using the α coefficient and basal ETK. Using a stepwise selection procedure, only variables that were significant at P<0. 05 were retained in the final model. Independent predictors for mortality were also identified. We categorized infants as aged <2 and ≥2 months as a secondary infant mortality peak at 3–4 months of age is associated with a high incidence of infantile beriberi in a population [39], [40]. We have attempted to report the study according to the STROBE guidelines [41]. Between January 7th 2003 and 2004 1,030 infants were admitted to the three pediatric wards. Their median (range) age was 4 (0–12) months and 595/1,023 (58. 2%) were male. Seven hundred and seventy-eight (75. 5%) infants were recruited; reasons for exclusion of 252 infants were: thiamin given before recruitment (34. 1%), infant discharged (19. 8%) or died (19. 0%) before assessment, insufficient blood sample volume (15. 9%), declined consent (6. 3%), readmission (3. 6%), responsible physician disagreed with recruitment (0. 8%) and inability to take a blood sample (0. 4%). The median (range) age of the 778 infants recruited was 5 (0–12) months (Table 1). Patients were recruited on the general pediatric ward (26. 6%), pediatric infectious disease ward (30. 6%) and on pediatric intensive care (42. 8%). Of 771 mothers with addresses recorded, 79. 8% and 16. 1% had homes in Vientiane City and Vientiane Province, respectively. The majority of infants (79. 2%) were breastfed but 32. 2% were both breast and bottlefed. Fever (≥37. 5 °C), dyspnoea, central cyanosis and cold fingers were present in 52. 8,39. 3,10. 4 and 17. 9 % of infants, respectively. The median (range) daily maternal cash expenditure was 19,000 (0–60,000) Lao kip, equivalent to median (range) 2. 4 (0–7. 5) $US in December 2010. Mothers ate predominantly glutinous (sticky) rice, with a median (range) of 3 (0–4) meals/day, rather than non-glutinous rice. The predominant discharge diagnoses were diarrhoea (30. 6%), pneumonia (24. 6%), sepsis (12. 1%) and prematurity (10. 6%). Parenteral thiamin was given to 19. 6% of infants after recruitment but before discharge. The overall mortality (died in hospital or taken home critically ill, not expected to survive) was 92/1,029 (8. 9%). For infants recruited to the study mortality was 43/777 (5. 5%). The median (range) α for 778 infants was 13. 7 (−13. 2 to 110. 1) % (Table 1). The median (range) basal ETK and activated ETK for the 382 infants for whom these was measured were 0. 84 (0. 20–3. 24) and 0. 95 (0. 23–3. 42) micromoles/min/gHb, respectively. The α coefficient and basal ETK were inversely correlated (Spearman' s rho = −0. 46, p = <0. 001, Fig. 1). Forty (5. 1%) infants had an α>31% and 51/382 (13. 4%) of infants had a basal ETK<0. 59 micromoles/min/gHb (Table 1); 19/51 (37. 3%) infants had basal ETK<0. 59 micromoles/min/gHb and α>31% but 3/22 (13. 6%) infants had α>31% and a basal ETK≥0. 59 micromoles/min/gHb. Of the 40 infants with α>31%, 30% received thiamin before discharge, in comparison to 19. 3% of those with α≤31% (P = 0. 09). Of the 51 infants with a basal ETK<0. 59 micromoles/min/gHb, 27. 5 % received thiamin before discharge, in contrast to 14. 6% of those with a basal ETK≥0. 59 micromoles/min/gHb (P = 0. 02), suggesting that subtle clinical evidence of thiamin deficiency may have been recognized after admission. Infant basal ETK<0. 59 micromoles/min/gHb was associated with a significantly (P<0. 02) higher frequency of infants presenting at ≥2 months of age, infant breastfeeding, death when ≥2 months old, a pneumonia discharge diagnosis, a lower frequency of prematurity discharge diagnosis, vomiting, dyspnoea, a lower pulse rate, fewer years of maternal schooling, a higher frequency of maternal Lao Sung ethnicity and maternal pork, chicken, fish or vegetables consumption since parturition (Table 1). α>31% was associated with a higher frequency of infants presenting at ≥2 months of age, of infant breastfeeding, vomiting, dyspnoea, cold fingers, maternal Lao Sung ethnicity, mothers being rice farmers and fewer years of maternal schooling. There was no apparent relationship between biochemical measurements of thiamin status and maternal food avoidance behaviour. There was a tendency for basal ETK to decline with infant age up to 6 months and then to increase (Fig. 2), whilst there was a tendency for α to rise in the first 3–6 months of age (Fig. 3). There was no obvious relationship between measures of biochemical thiamin deficiency and seasonality (data not shown). Multivariate logistic regression of the above admission variables significant (P<0. 02) on univariate analysis suggested that infant age ≥2 months (OR (95%CI) 11. 49 (3. 42–38. 65), P<0. 001) and fewer maternal years of schooling (OR (95%CI) (0. 85 (0. 78–0. 93), P< = 0. 001) were significantly independently associated with infant basal EKT<0. 59 micromoles/min/gHb. For the α coefficient, multiple logistic regression of the above admission variables significant (P<0. 02) on univariate analysis suggested that infant age ≥2 months (OR (95%CI) 4. 21 (1. 45–12. 25), P = 0. 008), the presence of cold fingers (OR (95%CI) 3. 25 (1. 61–6. 53), P = 0. 001), maternal occupation as rice farmer (OR (95%CI) 3. 30 (1. 33–8. 21), P = 0. 01) and fewer years of maternal schooling (OR (95%CI) 0. 90 (0. 82–0. 99), P = 0. 027) were significantly independently associated with α>31%. Mortality was 5. 5% but, among infants ≥2 months old, mortality was higher in those with basal ETK<0. 59 micromoles/min/gHb (3/47,6. 4%) than in those with basal ETK≥0. 59 micromoles/min/gHb (1/146,0. 7%, P = 0. 045) (relative risk = 9. 32 (95%CI 0. 99 to 87. 5) ). (Figs. 4 & 5). Mortality was significantly (P<0. 02) higher in younger infants, those who did not have a discharge diagnosis of diarrhoea or pneumonia, had a discharge diagnosis of prematurity, were not breast fed, did not receive rice water or chewed rice, had a faster pulse rate, lower body temperature, peripheral oedema, central cyanosis and respiratory distress. Overall, infants who died had lower α coefficient than those who lived. Basal ETK did not significantly differ between those who lived and died, but the sample size was considerably smaller (Table 2). There was no apparent relationship between mortality and the administration of thiamin after admission (P = 0. 8). Multiple logistic regression (n = 746) of admission variables significantly associated (<0. 02) with outcome on univariate analysis, suggested that breastfeeding (OR (95%CI) 0. 29 (0. 14–0. 62) P = 0. 001) and older age (OR (95%CI) 0. 85 (0. 74–0. 97) P = 0. 014) were associated with infant survival whilst respiratory distress (OR (95%CI) 5. 04 (2. 24–11. 36) P<0. 001) and higher pulse rate (OR (95%CI) 1. 02 (1. 00–1. 04), P = 0. 014) were independently associated with infant death. This study suggests that a substantial minority of infants (13. 4%) admitted without clinical evidence of beriberi had biochemical thiamin deficiency. Overall mortality was not associated with measures of biochemical thiamin deficiency, but the study was not powered to detect mortality differences. However, in support of the hypothesis that thiamin deficiency is an important cause of infant death in Vientiane, mortality was higher amongst those ≥2 months of age if they had low basal ETK (Fig. 4). A higher proportion of older infants, in comparison to those <2 months old, also had low basal ETK and α>31%. Largely forgotten work from the south Pacific and India [39], [40] suggested that a key clue to the importance of beriberi in a community was a secondary mortality peak at 3–4 months of age, after the high early mortality amongst neonates, as recently described from the Karen community displaced on the Thailand/Burma border [10]. As also noted for infantile beriberi, thiamin deficiency was associated with infant breastfeeding. The association between fewer years of maternal schooling and low basal ETK and α coefficient >31% suggests that aspects of maternal poverty and/or low levels of education in Vientiane may predispose their infants to thiamin deficiency. Aside from thiamin deficiency, infant mortality was associated with admission respiratory distress and non-breast feeding, suggesting that the care of these patients should be prioritized. This study has important limitations. Two laboratories had to be used for ETK determinations and haemoglobin concentrations were not assayed in Oxford and therefore basal EKTA could not be expressed for these samples. The long sample storage period is unlikely to have affected α [35], [36] and if it did, it would have tended to underestimate thiamin deficiency [42]. It is likely that infants with thiamin deficiency have other clinically important nutritional deficiencies, such as vitamin A, riboflavin and folate. Lao children have a high frequency of stunting [18]. Seventy years ago a (unrandomised) clinical trial suggesting that thiamin supplementation in east London infants increased growth [27]. There is an urgent need to examine the inter-relationships between different nutrients, infection and Lao childhood growth and development. Traditional prolonged post-partum Lao maternal food avoidances may lead to fatal wet beriberi in infants and neurological symptoms in nursing mothers [13], [18]. In the 2000 Lao National Health Survey 78% of women were recorded as observing food taboos after delivery and for a prolonged duration–a mean of 88 days [43]. Among 300 mother-infant pairs living on the outskirts of Vientiane, despite high levels of ante-natal care attendance (91%), 93% of women observed post-partum restricted diets and 96. 6% of mothers were estimated to have inadequate thiamin intake [18]. One explanation for the persistence of infantile beriberi in Laos is that post-partum food avoidances last longer in Laos than elsewhere, such as in Malaysia and Tamil Nadu, where food avoidances generally cease ∼40 days after delivery [44], [45]. There was surprisingly no apparent relationship between either biochemical measurement of thiamin status and maternal food avoidance behavior [13], [18]. Indeed, the mothers of infants with low basal ETK had more frequently taken pork than those with higher basal ETK-surprisingly as pork is an important source of thiamin. However, the dietary assessment used here was crude and more detailed assessments are needed. The finding that Lao Sung maternal ethnicity was associated with a higher frequency of α>31% and basal ETK<0. 59 micromoles/min/gHb suggests that more research is needed on ethnic and dietary risk factors for thiamin deficiency. Lao Sung ethnicity refers to mountain top people of diverse cultures and those near Vientiane may differ in dietary habits from those living in such more remote areas. That no infants with low basal ETK had a discharge diagnosis of prematurity reflects the relationship of basal ETK with age as all those with prematurity are, by definition, <2 months old. Importantly, the clinical significance of biochemical thiamin deficiency in both those apparently healthy and in those ill without clinical beriberi remains unclear. Basal ETK is influenced by factors other than thiamin status-younger red cells have higher ETK and differences between patients could reflect variation in haematopoesis and red cell survival [46]. Presumably basal ETK would be higher in those with haemolysis associated with glucose-6-phosphatase deficiency. However, there was no significant difference in basal ETK between male and female infants. In addition there may, or may not, be different red cell transketolase isoenzymes in human erythrocytes, differing in their affinity for thiamin pyrophosphate [47]. Lao people have multiple risk factors for thiamin deficiency. The consumption of polished rice, alcohol and thiaminase-containing foods such as ‘paa dek’ (fermented fish paste), thiamin antagonists such as betel nut and the hard physical labour of rural rice farming are likely to be important in adults [5], [6]. Malaria and other fevers in pregnant and post-partum women may further predispose to beriberi in their infants by further depleting the short term body stores. It has also been suggested that biochemical thiamin deficiency predisposes to infection [48] and host genetic factors may be important in susceptibility as may occur in Wernicke' s encephalopathy [49]. What do these results mean for Lao public health? There appears to be a significant but clinically unapparent burden of thiamin deficiency among sick infants admitted in Vientiane. Using the criteria of World Health Organization ([8] Table 7, α>25%), it would be described as a ‘mild’ public health problem. However, this definition may not be appropriate for infants [22] and given the ease and low cost of life-saving therapy it may be an important remedial public health issue. During the one year study 86 infants with suspected beriberi were admitted at Mahosot Hospital and between 2005 and 2008 ∼54 infants with beriberi were admitted/year with a mortality of 6%, despite thiamin therapy. At Luang Namtha provincial hospital in Northern Laos 3. 4% of all children admitted 2008–2009 had a clinical diagnosis of beriberi (P. Douangdala, S. Inthalad, G. Slesak, unpublished data). There are anecdotal reports of beriberi from other areas of Laos (C. Perks, L. Srour, H. Barennes, T. Saito pers. comm.) but, given the wide ethnic and nutritional diversity in Laos, whether all communities are afflicted remains very unclear [12]. If highly milled glutinous rice in conjunction with post-partum food avoidance are key factors we might expect infantile beriberi to be, paradoxically, a more important public health problem in the relatively affluent Lao populations with access to rice mills, rather than the more remote poor communities who rely on hand pounding rice. Aside from causing infant death, thiamin deficiency may have unappreciated longterm consequences on growth and neurological development. Among the infants on the Thailand/Burma border, McGready et al. [50] demonstrated delayed visual maturation type 1 and suggested thiamin deficiency as a possible cause. Follow up of Israeli infants who survived beriberi, due to substandard infant formula devoid of thiamin, suggests that thiamin deficiency was associated with subsequent delayed language development and epilepsy [51]. Whether epilepsy in some Lao patients is a consequence of infantile beriberi or clinically-cryptic thiamin deficiency requires investigation. If sick infants are to be supplemented, these data suggest that breast fed infants aged ≥2 months, of mothers with few years of schooling should be the priority group. In sick infants thiamin should be administered parenterally to increase tissue thiamin levels rapidly. There is evidence that gastrointestinal absorption of thiamin is saturated at doses of >5 mg [5], suggesting that oral doses above this give limited, if any, benefit. How long oral supplementation of thiamin should be continued for is unclear. If thiamin is given to breastfed infants it should also be given to their mothers, but how and for how long is unknown. Similarly, whether dietary supplementation of pregnant and nursing Lao mothers with thiamin or thiamin-rich foods would improve child and maternal health in Laos remains unclear. Although 94. 9% of Lao children were recorded as breastfed in the 2000 National Health Survey, only 23. 6% of Lao infants aged under 4 months were recorded as exclusively breast fed [43]. It will be important not to allow concerns of thiamin deficiency to reduce the prevalence of breastfeeding. Although the cost of 100 mg parenteral thiamin plus syringe/needle in Vientiane is ∼1. 5 $US, this is an appreciable amount in Laos where 74% of the population live on <2 $US/day [52]. Additional risks, such as injection abscesses and anaphylaxis, are present but rare [53] and there seems little evidence that thiamin supplementation increases the risk of cancer, although again there are very few data [54]. The high benefit and low cost of parenteral thiamin suggests that supplementation of older sick infants may be warranted. However, we would caution that the mortality difference presented here is based on only 4 deaths among 197 infants aged >2 months with basal ETK data with borderline relative risk and significance. The most efficient method to determine whether the biochemical thiamin deficiency is clinically important and whether thiamin supplementation should be provided on admission for sick infants, or those with particular syndromes, would be a clinical trial. More data are needed on the importance of rice milling/pounding and soaking practices and food avoidance behaviour. The age distribution of infant mortality in Laos and their geographical and ethnic variations may give valuable clues as to the public health importance of beriberi and inform the need for preventative measures.

Infantile beriberi, or clinical thiamin (vitamin B1) deficiency in infants, is a forgotten disease in Asia, where 100 years ago it was a major public health problem. Infants with this deficiency, commonly aged ∼ 2–3 months, present in cardiac failure but usually rapidly improve if given thiamin injections. It remains relatively common in Vientiane, Lao PDR (Laos), probably because of prolonged intra- and post-partum food avoidance behaviours. Clinical disease may be the tip of an iceberg with subclinical thiamin deficiency contributing to sickness in infants without overt clinical beriberi. We therefore recruited 778 sick infants admitted during one year at Mahosot Hospital, Vientiane, without clinical evidence of beriberi, and performed erythrocyte transketolase (ETK) assays. 13. 4 % of infants had basal ETK<0. 59 micromoles/min/gHb suggesting biochemical thiamin deficiency. Mortality was 5. 5% but, among infants ≥2 months old, mortality was higher in those with basal ETK<0. 59 micromoles/min/gHb (3/47,6. 4%) than in those with basal ETK≥0. 59 micromoles/min/gHb (1/146,0. 7%) (P = 0. 045, relative risk = 9. 32 (95%CI 0. 99 to 87. 5) ). We conclude that clinically unapparent thiamin deficiency is common among sick infants (≥2 months old) admitted to hospital in Vientiane. This may contribute to mortality and a low clinical threshold for providing thiamin to sick infants may be needed.

Abstract Introduction Methods Results Discussion

pediatrics and child health

2011

Clinically Unapparent Infantile Thiamin Deficiency in Vientiane, Laos

In microbiome studies, an important goal is to detect differential abundance of microbes across clinical conditions and treatment options. However, the microbiome compositional data (quantified by relative abundance) are highly skewed, bounded in [0,1), and often have many zeros. A two-part model is commonly used to separate zeros and positive values explicitly by two submodels: a logistic model for the probability of a specie being present in Part I, and a Beta regression model for the relative abundance conditional on the presence of the specie in Part II. However, the regression coefficients in Part II cannot provide a marginal (unconditional) interpretation of covariate effects on the microbial abundance, which is of great interest in many applications. In this paper, we propose a marginalized two-part Beta regression model which captures the zero-inflation and skewness of microbiome data and also allows investigators to examine covariate effects on the marginal (unconditional) mean. We demonstrate its practical performance using simulation studies and apply the model to a real metagenomic dataset on mouse skin microbiota. We find that under the proposed marginalized model, without loss in power, the likelihood ratio test performs better in controlling the type I error than those under conventional methods. In recent years, metagenomics studies have been growing rapidly due to the advances of next-generation sequencing (NGS) technologies [1]. Microbiota have been known to be associated with various diseases, e. g. , obesity and diabetes [2,3], Crohn’s disease [4], bacterial vaginosis [5], and cancer [6,7]. The microbial abundance is usually measured in read counts. However, such quantities are not directly comparable across samples due to the uneven total sequence counts of samples. Therefore, the read counts are often normalized to relative abundances which sum to 1 for all microbes in a sample [8]. Relative abundance can be characterized by a point mass at zero and a right-skewed continuous distribution with a positive support, the so-called “semi-continuous” or “zero-inflated continuous” data. The zero values indicate that certain microbes are absent in the sample, or the rare microbes are present but missed due to undersampling, while the continuous distribution with a positive support describes the levels of relative abundance among the present microbes. The relative abundance is often described by a two-part model [9], which separates zeros and positive values explicitly by two submodels: a logistic model for the probability of the outcome being positive in Part I and a (generalized) linear regression model for the amount of the (transformed) positive value in Part II. An important issue in such two-part models is to determine the distributional form in Part II. The nonzero relative abundance data are non-normally distributed and bounded in [0,1). Beta distribution has been used to model this outcome. A two-part Beta regression model can be thus developed [10–12]. It includes two sets of parameters, one in the logistic regression for the presence of a microbe, and the other in the Beta regression for the relative abundance conditional on the presence of the microbe. These two sets of parameters are interpreted as effects on the presence of a microbe and on the level of relative abundance given that the microbe is present, respectively. That is, there is a conditional interpretation in Part II. However, it is often of great interest to have a straightforward interpretation of covariate effects on the overall marginal (unconditional) mean. For example, [13] proposed a marginalized two-part log-normal model by parameterizing covariates effects directly in terms of the marginal mean. As conventional two-part Beta regression models do not provide an unconditional interpretation of covariate effects, we propose a marginalized two-part Beta regression model for microbiome abundance data which parameterizes covariate effects in terms of the marginal mean. The proposed model not only accounts for the zero-inflated nature of the microbiome data but also yields more interpretable effect estimates. Of note, an alternative to describe zero-inflated data is the Tobit model [14] where zero values are considered as left censored observations of the underlying true negative values (of Normal or other distributions accommodating negative values). However, the Tobit model is not appropriate for the Beta distribution which does not have a support of negative values. Consequently, the Tobit model cannot be applied directly to the relative abundance data. We begin with the conventional two-part model with a Beta component in Part II [10–12]. For a given operational taxonomic unit (OTU), let Yi denote its semi-continuous relative abundance for subject i, where 0 ≤ Yi < 1 and i = 1,2, …, n. Specifically, a two-part Beta regression model has the following form: Y i ∼ 0 with probability 1 - p i ∼ Beta (μ i ϕ, (1 - μ i) ϕ) with probability p i, where the density function of the Beta distribution is parameterized as Γ (ϕ) Γ (μ i ϕ) Γ [ (1 - μ i) ϕ ] y i μ i ϕ - 1 (1 - y i) (1 - μ i) ϕ - 1, with μi (0 < μi < 1) and ϕ (ϕ > 0) being the mean and dispersion parameters of the Beta distribution, respectively, and pi is the probability that the observation Yi is from the Beta distribution. The two-part model describes the probability pi in the logistic component and the conditional mean in the Beta component as functions of covariates, logit (p i) = log (p i 1 - p i) = X i T α, (1) logit (μ i) = logit [ E (Y i | Y i > 0) ]= log (μ i 1 - μ i) = X i T β, (2) where α and β are vectors of regression coefficients, Xi = (1, xi1, …, xip) T is the (p + 1) dimensional covariate vector (including an intercept) for the i-th subject. We assume identical covariates for both parts of the model for simplicity of notation. One can instead allow for different sets of covariates for the two parts. To obtain interpretable covariate effects on the marginal mean, we propose the following marginalized two-part Beta regression model. Let vi = E (Yi) be the marginal mean of Yi. The first part of the proposed marginalized two-part model is the same as Part I in the conventional two-part model, logit (p i) = log (p i 1 - p i) = X i T α. (3) In Part II, the marginal (unconditional) mean vi, instead of the conditional mean μi, is modeled as a function of covariates: logit (v i) = log (v i 1 - v i) = X i T γ. (4) As we can see, the marginalized two-part model not only captures zero-inflation and skewness as the conventional two-part model, but also allows us to examine covariate effects on the overall marginal mean. In the S1 Text, we can see that the likelihood of the conventional two-part model can be reparameterized to that of α, γ and ϕ in the marginalized model. However, the interpretation of covariate effects are different in the two frameworks, which will be elaborated in the next subsection. The estimation of the marginalized two-part model can be carried out in SAS Proc NLMIXED (The main code is shown in S1 Code). To obtain starting values of the estimation, a logistic model and a Beta regression model are fitted for the binary part and the positive part, respectively. Then the estimates of these two models are used as starting values for the two-part marginalized model. The convergence of the estimation is determined by a threshold value 1 × 10−8 for the relative gradient, a common convergence criterion in SAS Proc NLMIXED. This criterion is satisfied in our simulations for all replicates, and in the real data analysis for all 131 OTUs. In this section, we conduct simulation studies to evaluate the finite-sample performance of the proposed marginalized two-part model. To test the effect of the covariate on the overall marginal mean E (Yi), likelihood ratio tests (LRT) are performed and compared under the marginalized two-part (MTP) model and the conventional two-part (CTP) model. In addition, the two sample T-test and the Wilcoxon rank sum test are also compared. We assume that, in both parts, there is only one binary covariate x1, which is generated from the Bernoulli distribution with p = 0. 5. However, according to the interpretation of the covariate effects in the preceding section, the proposed model can be applied to multiple covariates. The response yi is generated below: logit (pi) =α0+α1xi1, logit (vi) =γ0+γ1xi1, f (yi) = (1−pi) 1 (yi=0) ×[ piBeta (μiϕ, (1−μi) ϕ) ]1 (yi>0), where μ i = 1+exp (- x i T α) 1+exp (- x i T γ) is the conditional mean given that yi is positive and ϕ is the dispersion parameter of the Beta distribution. In the simulation studies, 1000 samples of sizes 200 and 400 are generated. We set the parameters as α0 = 1. 5, γ0 = −2. 5, and ϕ = 1, while α1 and γ1 may have different values according to which of the two criteria are under study: the type I error or the power. First, we evaluate the type I error for testing the null hypothesis H0: the binary covariate x1 has no effect on the overall marginal mean of yi. In the MTP model, this is equivalent to testing H 0 M: γ 1 = 0 as shown in Eq (10). However, testing H 0 C: α 1 = β 1 = 0 in the CTP model is not equivalent to testing H0. Specifically, according to Eq (8), even though neither of the coefficients is zero, the binary covariate x1 may still have no effect on the marginal mean. This means that the conventional model cannot control the type I error for testing H0 when both α1 and β1 are non-zero. The results are shown in Fig 1. Type I errors are calculated under two settings: α1 = 0, γ1 = 0 and α1 = 1, γ1 = 0. For each setting, two α-levels are considered: 0. 01 and 0. 05. As we can see from Fig 1, under the first setting (α1 = 0, γ1 = 0), all the methods control the type I error reasonably well. Under the setting α1 = 1, γ1 = 0, the LRT under the MTP and the T-test control the type I error well, while the LRT under the CTP and the Wilcoxon test cannot control the type I error, especially the LRT under the CTP model. Because in this setting, testing H 0 C in the CTP model is not equivalent to testing the null hypothesis H0. The powers under two different settings, α1 = 0, γ1 = 1 and α1 = 1, γ1 = 1, are shown in Fig 2. As we can see, the LRT under the CTP and the MTP are the most powerful methods with the power close to 1 in all settings. The Wilcoxon test performs a little worse than the LRT while the T-test has the lowest power. We also estimate the coefficients in the MTP model under the setting α1 = 1, γ1 = 1. The results in Table 1 demonstrate that the biases are negligible and the coverage probabilities are acceptably close to the nominal level 0. 95 for all the model parameters. In addition, we observe small differences between the empirical standard errors and our estimates. The mean squared errors for sample size 400 are smaller than those for sample size 200. According to the simulation results, the LRT under the MTP model has the best performance: it controls the type I error reasonably well and also achieves the best power. The T-test has the similar performance in the error control while it is not as powerful as the LRT under the MTP model. The LRT under the CTP model is powerful, however, it fails to control the type I error. The Wilcoxon test has poor performances in both the error control and power than the LRT under the MTP model. To assess the robustness of the proposed method, we consider a setting where positive responses are generated from another distribution. First of all, the only covariate xi is generated from the Uniform distribution on (0,1), while the response yi has the following distribution: y i ∼ 0 with probability 1 - p i, where logit (p i) = α 0 + α 1 x i; and the overall marginal mean vi of the response is logit (v i) = γ 0 + γ 1 x i. Instead of the Beta distribution, positive responses are generated from the Binomial distribution Bin (100, μi) and then divided by 100 to make them bounded in (0,1). As in the previous simulation, we set μ i = 1+exp (- x i T α) 1+exp (- x i T γ). The probability of having exactly 0 success in 100 trials is (1 − μi) 100, which is negligible with the proper choice of the parameters α and γ. Thus almost all the zero values in this zero-inflated Binomial data are structural zeros. In this simulation study, 1000 samples of sizes 200 and 400 are generated. The parameters are set as α0 = 2, γ0 = −0. 5, while α1 and γ1 may have different values in order to calculate the type I errors and the powers. The type I errors are calculated under two settings: α1 = 0, γ1 = 0 and α1 = 1, γ1 = 0. For each setting, two α-levels are considered: 0. 01 and 0. 05. As we can see from Fig 3, under both settings, the proposed marginalized model controls the type I error reasonably well. The conventional model controls the type I error under the setting α1 = 0, γ1 = 0 while it fails under the setting α1 = 1, γ1 = 0, similar to Fig 1. As shown in Fig 4, both the marginalized model and the conventional model have power equal to 1 under all settings. From the simulation studies we can conclude that the proposed marginalized two-part Beta regression model is powerful and control the type I error well. Also, it is robust against model misspecification. In this section, the proposed marginalized two-part model and the conventional two-part model are applied to a real metagenomic dataset on mouse skin microbiota to investigate the effects of immunization on the relative abundances of 131 core OTUs [16,17]. The data are publicly available at https: //www. nature. com/articles/ncomms3462#supplementary-information. In addition to the likelihood ratio tests under CTP and MTP, the T test and the Wilcoxon rank sum test are also included for comparison. All the tests are carried out with Bonferroni’s correction. The skin dataset contains the relative abundances of the most common 131 OTUs for 261 mouse skin samples, including 78 non-immunized and 183 immunized individuals. There is a presence of a large portion of zero abundances in the skin data, ranging from 0 to 68. 97% with average 33. 03% and median 33. 72% (see S3 and S4 Figs). The positive values are highly right skewed and the logit transformations in the MTP model and the CTP model capture the skewness (See S5 Fig). Fig 5 shows the results for these four methods. As we can see, the LRT under the marginalized two-part model results in significant effects of immunization on 45 (namely, 31 + 14) OTUs. The LRT under the conventional two-part model has significant results for all these 45 OTUs, and 14 (namely, 8 + 4 + 2) additional OTUs. The T test identifies 31 of these 45 OTUs and another 7 (namely, 4 + 3) OTUs. Similar to the LRT under conventional two-part model, the Wilcoxon test identifies all these 45 OTUs and 21 (namely, 9 + 8 + 4) additional OTUs. Finally, 60 OTUs are not identified by any methods. The LRT under the CTP model and the Wilcoxon test identify more OTUs than the LRT under the MTP due to their failure to control the type I error as shown in Simulation studies (Fig 1). Actually, for those 14 OTUs identified by the CTP but not by the MTP, all of them have significant coefficients in Part I of the two-part model. Out of the 21 OTUs that are identified by the Wilcoxon test but not by the MTP, 17 have significant coefficients in Part I of the two-part model. This corresponds to the setting α1 = 1, γ1 = 0 where both the CTP and the Wilcoxon test have much higher type I errors than the MTP (See the lower panel of Fig 1). Because it is less powerful than the MTP (Fig 2), the T test identifies less OTUs than the MTP. Table 2 shows 10 most significant OTUs from the MTP model. As in [17], for OTUs which cannot be classified at the species level, the next highest classifiable taxonomic level (denoted by ‘o’, ‘f’ and ‘g’ for order, family, and genus, respectively) is displayed. We use a number in the superscript to distinguish among different OTUs with the same classification name. The detailed results of all the 45 OTUs identified by the proposed MTP model are shown in S1 Table. Moreover, for most of the 131 OTUs, the proposed marginalized two-part model fits the observed data better than the conventional two-part model. Fig 6 shows the density curves of the observed relative abundances, the predicted relative abundances using the MTP model, and the predicted relative abundances using the CTP model for two OTUs. As we can see, the MTP model fits the observed data much better than the CTP model. In this paper, we propose a marginalized two-part Beta regression model for semi-continuous microbiome compositional data. This model allows investigators to obtain covariate effects on the marginal mean of the outcome. It takes into account the compositional and zero-inflation nature of the microbiome relative abundance data. It also has an unconditional interpretation of the covariate effect on the marginal mean. Our proposed marginalized two-part model has satisfactory performance in both simulation studies and real data analysis. For count outcomes exhibiting many zeros, a zero-inflated Poisson (ZIP) regression model or a zero-inflated negative binomial (ZINB) model, is often employed to examine the relation between covariates and the response. To model the overall population mean count directly, the marginalized ZIP model and the marginalized ZINB model were proposed by [18] and [19], respectively. However, in the case of bounded count data, the ZIP is questionable while the zero-inflated binomial (ZIB) model and its extension for over-dispersion: the zero-inflated beta-binomial (ZIBB) model, are available in [20–22]. It is of interest to develop a marginalized modeling approach for ZIB or ZIBB. More recently, there has been increasing interest in analyzing correlated zero-inflated semi-continuous data. The correlation may stem from the structure of clustered data or from longitudinal data where repeated measures are correlated for the same subject. Typically, random effects are included to account for the correlations between observations [10,15,23–25]. However, similar limitation exists in these two-part random effects models, as they cannot account for covariate effects on the marginal mean. Recently, Smith et al. [26] proposed a marginalized two-part model for longitudinal semicontinuous data based on the log-skew normal distribution for positive values. In future studies, we will extend our marginalized two-part model to correlated semi-continuous data bounded by 0 and 1. Finally, it is of interest to consider different microbiomes together, taking into account the constraint that the relative abundances of all OTUs sum to 1. Scealy and Welsh [27,28] considered Kent models for such compositional data. It merits further consideration to incorporate zero values in the Kent model framework.

Semi-continuous compositional data are typically analyzed using two-part models which separately describe the probability of zero values and the distribution of positive values. The second part of the model provides a conditional interpretation of covariate effects on the positive response. However, it is of great interest in many applications to assess the covariate effect on the marginal mean of the response. For this purpose, we propose a marginalized two-part model by reparameterizing the marginal mean in Part II. We show that the proposed marginalized two-part model outperforms conventional methods by simulation studies in terms of controlling the Type I error and maximizing the power. We apply our method to a microbiota dataset, and find consistent results with our simulation studies.

Abstract Introduction Models Results Discussion

sequencing techniques microbiome microbiology simulation and modeling next-generation sequencing metagenomics probability distribution mathematics genome analysis molecular biology techniques skewness microbial genomics research and analysis methods medical microbiology molecular biology research assessment probability theory normal distribution genetics transcriptome analysis biology and life sciences physical sciences genomics research errors dna sequencing computational biology

2018

A marginalized two-part Beta regression model for microbiome compositional data

Mitochondrial DNA (mtDNA) mutations cause severe maternally inherited syndromes and the accumulation of somatic mtDNA mutations is implicated in aging and common diseases. However, the mechanisms that influence the frequency and pathogenicity of mtDNA mutations are poorly understood. To address this matter, we created a Drosophila mtDNA mutator strain expressing a proofreading-deficient form of the mitochondrial DNA polymerase. Mutator flies have a dramatically increased somatic mtDNA mutation frequency that correlates with the dosage of the proofreading-deficient polymerase. Mutator flies also exhibit mitochondrial dysfunction, shortened lifespan, a progressive locomotor deficit, and loss of dopaminergic neurons. Surprisingly, the frequency of nonsynonymous, pathogenic, and conserved-site mutations in mutator flies exceeded predictions of a neutral mutational model, indicating the existence of a positive selection mechanism that favors deleterious mtDNA variants. We propose from these findings that deleterious mtDNA mutations are overrepresented because they selectively evade quality control surveillance or because they are amplified through compensatory mitochondrial biogenesis. Mitochondria contain the electron transport chain complexes responsible for generation of most of a cell’s energy and also play crucial roles in Ca2+ buffering, metabolite synthesis, and apoptosis [1–3]. In addition to the ~1,000–2,000 nuclear genes that encode mitochondrial proteins, mitochondria contain a separate small circular genome, densely packed with 37 genes, that is essential for mitochondrial function. Mitochondrial DNA (mtDNA) mutations transmitted through the female germline are responsible for a host of incurable mitochondrial syndromes [4]. In addition, accumulation of mtDNA mutations in somatic tissues is implicated in aging [5,6] and common diseases of the elderly including cancer [7] and neurodegenerative diseases [8]. There are typically thousands of copies of the mitochondrial genome in a single cell, such that when mtDNA mutations occur, they frequently share residence with wild-type mtDNA, a condition known as heteroplasmy. High levels of heteroplasmic mutations correlate with the severity of mitochondrial diseases [9], yet we know little about the factors that influence the frequency of mtDNA mutations or the emergence of their associated phenotypes [10,11]. To explore the cellular mechanisms that influence the frequency and pathogenicity of mtDNA mutations, we are using the fruit fly Drosophila as a model system. In previous work, we found that many fundamental features associated with somatic mtDNA mutations in mammals are conserved in Drosophila, including a similar mtDNA mutation frequency, a preponderance of transition mutations, and an increased frequency of mtDNA mutations with age [12]. Bratic et al. further extended the utility of using Drosophila to study mtDNA mutations by knocking in exonuclease- and polymerase-deficient forms of DNA polymerase γ (PolG), the polymerase responsible for replicating the mitochondrial genome [13]. While these mutant strains are developmentally lethal in the larval stage as homozygotes, heterozygotes for the exonuclease-deficient PolG exhibited increased mtDNA mutation frequency. However, the physiological effects of this high mutation burden and the possibility of negative selection acting against the resulting mtDNA mutations were not explored. In our current work, we created a transgenic proofreading-deficient version of the Drosophila mtDNA polymerase (designated PolGmut) that confers a 10- to 55-fold increase in the mtDNA mutation frequency, depending on transgene dosage without exhibiting embryonic lethality. PolGmut expressing flies exhibited dosage-dependent phenotypes analogous to those of human mitochondrial diseases, including shortened lifespan, a progressive locomotor defect, and loss of dopaminergic neurons. Analysis of the frequency and distribution of mtDNA mutations in these mutator flies revealed an unexpectedly high ratio of nonsynonymous to synonymous mtDNA mutations. The mutations detected in mutator flies also tended to occur preferentially at conserved mtDNA sequences and resulted in pathogenic alterations. Together, these findings suggest that positive selection acts in favor of deleterious mitochondrial variants, either through the selective evasion of mutant-bearing mitochondria from negative selection or because cells that stochastically acquire deleterious mtDNA mutations activate compensatory mitochondrial biogenesis. Future work with this mtDNA mutator model will facilitate the study of cellular mechanisms that influence the frequency and pathogenesis of mtDNA mutations, as well as the identification of molecular factors that influence these processes. Previous work has shown that altering a conserved aspartate residue in the second exonuclease domain of PolG to alanine impairs proofreading ability and results in a dramatically elevated mtDNA mutation frequency in multiple species [13–17]. Thus, we generated a Drosophila PolG transgenic construct with an alanine substitution at the equivalent site (designated PolGmut; Fig 1A). Because overexpression of PolG in Drosophila results in mtDNA depletion [18], we created our transgene using a genomic DNA fragment containing both the endogenous PolG gene and its associated cis-regulatory transcriptional elements to avoid artifacts associated with overexpression (Fig 1A). This construct was then used to create transgenic flies using standard methodologies (see Materials and Methods). Flies bearing one or two copies of the PolGmut transgene expressed similar levels of PolG mRNA as the endogenous PolG gene, thus confirming that this transgene does not cause PolG overexpression (Fig 1B). Moreover, a single copy of the PolGmut transgene was capable of rescuing the recessive lethal phenotype caused by an overlapping set of deletions that remove the endogenous PolG gene, thus confirming that the PolGmut transgene encodes a functional mtDNA polymerase (Fig 1A and 1C). However, the rescued flies displayed a marked reduction in mtDNA copy number (Fig 1D), consistent with previously published work suggesting that Drosophila PolG bearing this proofreading alteration is not fully functional [13]. In contrast, flies expressing one or two copies of the PolGmut transgene in a strain hemizygous for the endogenous PolG gene did not display mtDNA depletion (Fig 1D). Given that mtDNA depletion could potentially confer phenotypes that are unrelated to mtDNA mutations, all of our remaining work involved the use of flies hemizygous for the endogenous PolG gene (Df2) and bearing zero, one, or two copies of the PolGmut transgene, which we refer to as 0xPolGmut, 1xPolGmut, and 2xPolGmut, respectively. To test whether the PolGmut transgene conferred an increased mtDNA mutation frequency, we performed Duplex Sequencing (DS) on mtDNA isolated from individual heads of 1-day-old transgenic flies. DS is a high-accuracy next-generation sequencing approach capable of detecting a single mutation in >107 wild-type bases [19]. We addressed the possible influence of genetic background by comparing PolGmut transgenic flies to control non-transgenic flies hemizygous for the endogenous PolG. Furthermore, fly strains were outcrossed to the same WT strain prior to sequence analysis to ensure that all of the genotypes being compared inherit their mtDNA from the same parental strain, as well as to replace mitochondrial genomes that had potentially accumulated mtDNA mutations over multiple generations of replication by the mutator polymerase (S1 Fig). We first tested whether the Polgmut transgene introduces greater replication errors by measuring the frequency of unique mutations (see materials and methods for description of unique mutations) in 1-day-old flies. 0xPolGmut (control) flies had a mutation frequency comparable to the frequency previously reported for WT flies (3. 4x10-6 ± 8. 9x10-7). By contrast 1xPolGmut flies exhibited significant increases in the point mutation, insertion and deletion frequency relative to controls (Fig 2A, S4 Fig), and the frequency of these mutations increased further in 2xPolGmut flies (Fig 2A, S3 Fig, S4 Fig). Because 2xPolgmut flies inherit one copy of the PolGmut transgene maternally (S1 Fig), the increased mtDNA mutation frequency in 2x relative to 1xPolGmut animals may derive from an increased somatic mtDNA mutation frequency, as well as mutations that arise in the female germline. Consistent with previous work involving exonuclease-deficient mtDNA polymerases, mutator flies exhibited an increased prevalence of predominantly G: C to A: T transition mutations (Fig 2D) [14,20,21]. We next explored the influence of age on mtDNA mutation frequency in mutator flies by sequencing mtDNA from flies aged 25 and 50 days. The highest mtDNA mutation frequencies were observed in 50-day-old 2xPolGmut flies, in which the point mutation frequency of unique mutations was elevated ~55-fold relative to age-matched controls (Fig 2A). Although there was a trend towards increased mutation frequency with age for all mutation types detected, only 2xPolGmut flies exhibited a significant age-associated increase in point mutations relative to young flies of the same genotype (Fig 2A). Like young mutator flies, aged mutator flies also exhibited a prevalence of G: C to A: T transition mutations (S2 Fig). Additionally, when we combined data from control 0xPolGmut animals of all ages to increase the total number of mutations detected, G: C to A: T transition mutations were also the most frequent mutation type detected, consistent with our previously published work [12]. To search for evidence of clonal expansion, we also quantified the total mutation frequency, which in contrast to the unique mutation frequency, includes multiple occurrences of the same mutation. Control flies harbor a low total mutation burden (1-day-old flies had just 4. 7x10-6 ± 1. 8x10-6 mutations/nucleotide, or ~1 point mutation per 14 mtDNA molecules; Fig 2B, S1 Table) that is similar to that of the unique mutation frequency, suggesting little clonal expansion occurs in WT flies [12]. The total mutation frequency increased in a dose-dependent manner in mutator flies: 1-day-old 1xPolGmut flies had a mutation frequency of 3. 0x10-5 ± 0. 4x10-5 (~one point mutation per 2. 2 molecules of mtDNA Fig 2B, S1 Table) and 1-day-old 2xPolgmut flies had a mutation frequency of 8. 8x10-5 ± 0. 9x10-5 (~1. 3 point mutations/mtDNA molecule; Fig 2B, S1 Table). Although the total mutation frequency in mutator flies exceeded that of the unique mutation frequency, the increase was small and we discovered very few clonally-expanded mutation sites in flies of any genotype (Fig 2C). Furthermore, very few mutations rose above 1% heteroplasmy even in old 2xPolGmut flies (Figs 2C and 3A). These findings suggest that clonal expansion of mtDNA mutations is either restricted to individual cells, or that the short lifespan of Drosophila is incompatible with extensive clonal expansion of mtDNA mutations. Our previous work to measure the mtDNA mutation frequency in Drosophila involved the Random Mutation Capture method, which only allowed us to analyze three small parts of the mitochondrial genome, thus precluding detailed analysis of the distribution of mtDNA mutations [12]. By contrast, DS enabled us to characterize the frequency of mutations across the entire mitochondrial coding sequence. Only the non-coding control region [ChrM: 14917–19524] was refractory to DS owing to its high A: T content (~95%), which prevents efficient sequence capture and accurate reassembly. We found that mutations were distributed relatively uniformly between tRNA, rRNA, and protein-coding genes with no apparent mutational hotspots or mutational deserts (Fig 3A and 3B, S3 Fig). The mild variation in mutation frequency detected between genes is likely explained by differences in GC content and the G: C to A: T mutation bias of the polymerase (Fig 3C). Mutator mice display premature aging phenotypes, including a reduced lifespan, kyphosis, and hair loss [15]. Therefore, we tested whether mutator flies also exhibit signs of premature aging by examining lifespan and locomotor activity. Control flies had a median lifespan of 75 days, whereas the 1xPolGmut flies had a modest reduction in lifespan, displaying a median lifespan of 64 days (Fig 4A). 2xPolGmut flies showed a further reduction in lifespan, with a median lifespan of 53 days. Mutator flies also displayed a defect in locomotor performance using a simple test of climbing ability. Normal flies exhibit negative geotaxis, climbing to the top of a vial after being tapped to the bottom, and this behavior declines naturally as flies age. The presence of the PolGmut transgene did not influence climbing ability in young flies, but conferred a dose-dependent decline in climbing behavior in older flies relative to controls (Fig 4B). The decline in climbing ability preceded the onset of decreased viability in 2xPolGmut flies and ultimately culminated in a complete failure in climbing at ages >25 days. Approximately 10% of 2xPolGmut flies also exhibited a rhythmic seizure phenotype beginning approximately 24 hours prior to death (S1 Video). This phenotype was not observed in any other genotype. A number of previous observations indicate that dopaminergic neurons are particularly sensitive to mitochondrial stress [22–25]. In particular, mitochondrial toxins and mutations affecting mitochondrial quality control pathway components result in the selective death of dopaminergic neurons in humans and animal models [22]. Moreover, a recent study has shown that mice expressing a proofreading-deficient form of DNA polymerase γ exhibit selective dopaminergic neuron death when crossed to parkin mutant mice [26]. These observations prompted us to test whether an increased load of mtDNA mutations leads to degeneration of dopaminergic neurons in Drosophila. To perform this analysis, we dissected whole brains and immunostained with antiserum to tyrosine hydroxylase to quantify the number of dopaminergic (DA) neurons in 50-day-old flies. The number of DA neurons in the protocerebral posterior lateral 1 (PPL1) cluster was significantly reduced in 2xPolGmut flies (Fig 4C, S5 Fig), suggesting that a high mtDNA mutational load triggers the loss of a subset of dopaminergic neurons consistent with these prior reports [27]. To test whether the premature aging phenotypes of mutator flies are caused by mitochondrial dysfunction, we monitored several mitochondrial functional parameters. Since PolG mutator mice show reduced Complex IV activity and assembly [26], we assayed Complex IV activity in mutator flies and found it to be significantly reduced in a dose-dependent fashion relative to age-matched controls (Fig 4D). Consistent with this deficit, mutator flies also had a reduced abundance of ATP (Fig 4E). 2xPolGmut flies also had elevated levels of the autophagy marker Ref (2) p (the Drosophila homolog of p62), possibly suggesting that there is a buildup of autophagic intermediates in response to an upregulation of autophagy (Fig 4F and 4G). However, the abundance of the mitochondrial unfolded protein stress markers HSP60 and mitochondrial HSP70 were unchanged in mutator flies (S6 Fig), indicating that the increased load of mtDNA mutations does not result in sufficient protein misfolding to trigger activation of the mitochondrial unfolded protein stress response. We also observed an abnormal downturned wing posture in 2xPolGmut mutator flies at 35 days of age, similar to that seen in Drosophila PINK1 and parkin mutants [28]. However, in contrast to PINK1 and parkin mutants, both of which exhibit apoptotic muscle degeneration, there was no gross evidence of muscle degeneration or apoptosis in mutator flies. Ultrastructural examination of flight muscle tissue also failed to detect alterations in mitochondrial morphology or integrity (S7 Fig). Together, these results suggest that flies harboring high mutation loads suffer from non-structural muscle abnormalities, thus potentially making these flies a suitable model for the study of mitochondrial myopathies. Studies in cultured cells have indicated the existence of pathways that can be manipulated to decrease the frequency of a deleterious heteroplasmic mutation [29–32]. However, there is little evidence that these pathways are normally operative in the somatic tissues of an intact animal model. To address this matter, we subjected mutator flies to a variety of analyses aimed at the detection of selective forces acting against harmful mutations. To account for the clonality of mutations, we used the frequency of total mutations in all of our remaining analyses. To diminish the influence of sampling bias and increase the number of mutations detected per animal, we re-sequenced 1-day-old 1xPolGmut flies using a technical advance in reagent preparation for the Duplex Sequencing protocol that was developed during the course of our study, thus providing us with a high-quality dataset with very high sequencing depth. The use of 1xPolGmut flies for this analysis also ensures that the mutations detected from sequencing are acquired in somatic tissues (S1 Fig). Because the third codon position is often degenerate, we hypothesized that negative selection acting against deleterious variants would result in a lower frequency of mutations at the first two codon positions relative to the third codon position. Our results were in complete reverse to our expectations: we detected higher mutation frequencies at the first and second codon positions (Fig 5A). However, a confound in this analysis concerns the high frequency of mutations at G: C base pairs, and the relative deficiency of G: C base pairs in the third codon position (S8 Fig). Notably, the AT-rich Drosophila mitochondrial genome, like many other insect species, primarily consists of A: T bases at four-fold degenerate (i. e. , synonymous) sites, where the frequency of A: T base pairs is 94% [33,34]. By contrast, G: C base pairs are predominantly located at nonsynonymous (NS) sites. To circumvent this confound, we compared the mutation frequency at NS sites and at four-fold synonymous (S) sites separately for A: T and G: C base pairs. Because mutations arising at four-fold degenerate sites do not alter the encoded amino acid, such mutations should be present at higher frequency relative to those at NS sites in the context of negative selection acting against deleterious variants. We detected no significant difference in the mutation frequency between NS sites and S sites at A: T positions. However, at G: C positions the mutation frequency was higher at NS sites relative to S sites, in complete opposition to the expectations of a negative selection model (Fig 5B). The unexpected finding that deleterious mtDNA variants were overrepresented in Drosophila prompted us to examine this matter further. Specifically, we performed Monte Carlo simulations of random mutagenesis, such that we could compare our findings from sequencing mutator flies to a neutral mutational model derived from simulations. Because the mutation frequency at G: C sites greatly exceeds that at A: T sites, we performed simulations to precisely mirror the mutational biases detected in mutator flies. Each round of simulation selected the proportion of each type of mutation detected within protein-coding regions of 1xPolGmut flies and computationally redistributed these mutations randomly throughout the coding portions of the mtDNA. Moreover, because the probability of detecting a mutation at any given site in the mitochondrial genome is directly proportional to the sequencing depth at that site, we also weighted the probability of detecting a nucleotide alteration at each site according to sequencing depth at that site in our simulations (for further details, see the Materials and Methods section). Simulations were repeated 10,000 times to create a distribution of neutral outcomes, thus providing a framework to compare findings from 1xPolGmut flies. We performed three analyses designed to detect whether negative selection reduces the frequency of pathogenic mtDNA mutations. First, we tested the hypothesis that mutations at evolutionarily conserved mtDNA sites are more susceptible to negative selection than those at sites of low conservation. Second, we tested the hypothesis that mutations resulting in nonsynonymous (NS) alterations are more prone to negative selection than synonymous (S) mutations. Third, we tested the hypothesis that mutations resulting in deleterious amino acid alterations are more susceptible to negative selection than those that result in conservative amino acid alterations. Our hypotheses predict that mtDNA mutations arising at highly conserved sites, particularly those mutations that result in deleterious NS amino acid alterations, are eliminated through negative selection. Consequently, these deleterious mutations should be underrepresented in mutator flies relative to a distribution of randomly generated mutations. To test the prediction that mutations at conserved sites are underrepresented in mutator flies, we used the PhyloP algorithm, which calculates mtDNA positional conservation scores from pairwise comparisons between 27 insect species [35]. The PhyloP algorithm assigns a logarithmic score to each nucleotide position indicating the degree of evolutionary conservation at that site; a score of zero indicates neutral evolution, whereas negative and positive scores suggest accelerated evolution and increased conservation, respectively. We then compared the average PhyloP scores from the mutations identified in mutator flies to a distribution of PhyloP scores created from Monte Carlo simulations of random mutagenesis. Consistent with our comparison of NS and four-fold S sites, we found that mutations at sites with high PhyloP scores are overrepresented, again suggesting that deleterious variants are overrepresented in mutator flies (Fig 5C). We next tested whether NS mutations were underrepresented in mutator flies by calculating the NS/S ratio and comparing this ratio to a distribution of NS/S ratios obtained from Monte Carlo simulation of random mutagenesis (neutrality). Our hypothesis that negative selection preferentially eliminates nonsynonymous mutations predicts a reduction in the occurrence of nonsynonymous mutations and thus the observed NS/S ratio should be lower than expected from neutrality. In contrast to our hypothesis, the NS/S ratio detected in mutator flies is significantly elevated relative to neutrality, indicating that nonsynonymous mutations are overrepresented in mutator flies (Fig 5D). These findings are inconsistent with the hypothesis that negative selection acts against nonsynonymous mutations. As a final test of the model that negative selection acts against deleterious mtDNA variants, we used the MutPred algorithm [36] to compare the pathogenicity of nonsynonymous mutations found in mutator flies to a distribution of MutPred scores created from Monte Carlo simulations of random mutagenesis. The MutPred algorithm uses the structural and functional properties of a protein to predict the functional consequence of a nonsynonymous amino acid substitution, and previous work has established the validity of the MutPred algorithm to predict the consequences of mtDNA mutations [37–39]. MutPred assigns scores ranging from 0 to 1 to quantify the pathogenicity of a particular variant, with higher scores indicating a greater likelihood of pathogenicity. If negative selection acts to preferentially remove the most deleterious mutations, mutator flies should accumulate variants with low MutPred scores. In contrast to this prediction, the detected mutations have an average MutPred score that is significantly higher than expected from a neutral mutational model (Fig 5E). To confirm findings from simulations, we added the combined data from our previously acquired low-depth sequencing of 1x and 2xPolGmut flies in an effort to increase the total number of mutations. We then reran our simulations using this data and applied the same three metrics to ask if there is selection against harmful mutations. Results of this analysis again revealed that deleterious mutations are overrepresented in mutator flies (S10 Fig). Our previous work revealed a strand asymmetry in the occurrence of C: G to T: A mutations [12], and this phenomenon could potentially influence our distributions of simulated mutations. To eliminate the potential confound of a mutational strand asymmetry, we repeated our analysis using the coding sequence of the Mitochondrial Cytochrome c oxidase subunit I (COX1) gene, which does not exhibit strand asymmetry in the mutation spectrum (S9A Fig). While we did not detect a significant difference in the evolutionary conservation of mutations arising in COX1 (S9B Fig) relative to neutrality, we discovered that these mutations displayed an elevated NS/S ratio (S9C Fig) and an overrepresentation of mutations with high MutPred Scores (S9D Fig), indicating that our findings are not a consequence of mutational strand asymmetry. Taken together, our analyses do not support the model that selection acts against deleterious mtDNA mutations; instead, our findings indicate that pathogenic mtDNA variants are overrepresented in mutator flies. One potential explanation for the overabundance of deleterious mutations in mutator flies is that nucleotide context influences the mutation frequency and, by chance, results in a higher mutation frequency at functionally important sites. Such an occurrence was observed in mutator mice where the high abundance of C>T transitions within the ‘TCA’ context was suggested to explain the high pathogenicity of mtDNA mutations in this organism [40]. To explore this model, we examined the influence of trinucleotide context of the nucleotides directly 5’ and 3’ to the mutation bearing sites on mutation frequency. We first began by computing the number of mutations observed in all 96 trinucleotide contexts. Like mutator mice, we found that C>T mutations within the TCA context represented one of the most abundant trinucleotide mutation categories. However, unlike mutator mice, the TCA context was not the most abundant mutation category and the abundance of mutations within the TCA context was similar to the frequency of other C>T mutation contexts (Fig 6A). Also, despite the high frequency of G: C>A: T mutations, trinucleotide contexts associated with T>C transitions, such as the ATT trinucleotide context, were similarly abundant to those associated with G: C>A: T mutations (Fig 6A). A limitation of comparing the total abundance of mutations within a particular trinucleotide context is that this value will be influenced by the prevalence of that context in the genome. In addition, our ability to detect mutations at any given site in the genome will depend on sequencing depth at that site. Thus, to test whether the trinucleotide context influences mutation frequency, we calculated the trinucleotide mutation frequency by normalizing the raw number of mutations observed in each trinucleotide context to their prevalence in the genome and to the sequencing depth at these sites (see Materials and Methods). Analysis of the trinucleotide mutation frequency substantially decreased the heterogeneity across the entire C>T trinucleotide context (Fig 6B). Normalizing our data in this fashion also revealed that the high abundance of particular T>C mutation contexts reflects the composition of the Drosophila mitochondrial genome rather than an influence of sequence context on mutation frequency (Fig 6B). Given the similarity of the mutation frequency within any given trinucleotide context, we conclude that the trinucleotide context does not play a major role in the overrepresentation of deleterious mutations. The progressive accumulation of deleterious mtDNA mutations in somatic tissues is implicated in aging and common diseases of the elderly. Although the frequency of these mutations correlates with the severity of the symptoms that they cause [9], little is known about the pathways that influence the frequency and pathogenicity of mtDNA mutations. In previous work, we showed that many of the features associated with mtDNA mutations are similar in flies and vertebrates, including a low frequency of mutations and a preponderance of transition mutations [12,13]. Our current work offers further evidence in support of these previous findings by using a high accuracy next-generation sequencing approach that allowed us to sensitively detect mtDNA mutations over a broader region of the mitochondrial genome [19]. We have also extended the utility of Drosophila as a model system for studying mtDNA mutations by creating a Drosophila mtDNA mutator strain that exhibits a dramatically increased mtDNA mutation frequency and displays features associated with mitochondrial diseases and premature aging including mitochondrial dysfunction, reduced lifespan, a progressive locomotor deficit, and loss of dopaminergic neurons. Surprisingly, deleterious mtDNA mutations were overrepresented in mutator flies, suggesting the existence of a novel selective mechanism underlying this phenomenon. Our current work provides a foundation to explore the factors responsible for the overabundance of harmful mtDNA mutations and the pathways that influence the pathogenicity of these mutations. Our findings warrant comparison to another recently described Drosophila mtDNA mutator strain that was created by introducing the same exonuclease mutation used in our current study into the endogenous PolG locus [13]. While heterozygotes for this PolG knock-in allele displayed a similar increase in the mtDNA mutation frequency to flies that bear a single copy of our mutator transgene, homozygotes for the knock-in PolG allele did not survive beyond the larval stage of development. By contrast, a single copy of our PolGmut transgene rescued the recessive lethal phenotype of an overlapping set of deletions that completely remove the endogenous PolG gene. Given that our mutator transgene appears to express endogenous levels of PolG transcript, we are at a loss to explain this discrepancy. Moreover, heterozygotes for the PolG knock-in allele displayed no obvious effect on lifespan [41], whereas our mutator lines displayed a dose dependent decrease in lifespan. The explanation for this difference is also unknown but may reflect differences in the mtDNA genetic background in these two studies, which we have found in unpublished work can influence lifespan. Further study of the knock-in PolG mutant revealed that the mtDNA mutation frequency progressively increased in subsequent generations, indicating that purifying selection is unable to fully keep pace with an increased mtDNA mutation rate [13]. Although we have not directly explored the influence of our mutator transgene in the female germline, we find that mutator stocks lose viability within several generations without periodic outcrossing, consistent with the model that the accumulation of mtDNA mutations is responsible for this loss of viability. The finding that somatic mtDNA mutations accumulate with age at an accelerated rate relative to mutations in the nuclear genome has led to the suggestion that aging may be a consequence of accumulated mtDNA mutations [9]. While the phenotypes of homozygous PolG mutator mice were initially offered as support for this hypothesis, it was later found that heterozygous mutator mice live a normal lifespan despite having a mtDNA point mutation frequency that greatly exceeds that of elderly WT mice [42,43]. Our findings also indicate that a high mtDNA point mutation load can negatively affect longevity. However, our data do not support the model that the shortened lifespan of mutator flies results from the progressive accumulation of mtDNA mutations. Although there is a trend towards increased mtDNA point mutation frequency with age in mutator flies, this age-dependent increase in mtDNA point mutations is small in comparison to the mutation frequency in young mutator flies. For example, the mutation frequency in 50-day-old 1xPolGmut is approximately 19-fold higher than age-matched controls, but fully 89% of the mutations detected in 50-day-old 1xPolGmut are acquired by the time these flies reach 1 day of age. These findings suggest that a high mtDNA mutational load throughout life simply increases the probability of death later in life. The delayed phenotypic effects of mtDNA mutations may reflect the long half-lives of mitochondrial proteins [29,44–46], such that the consequences of a mtDNA mutation would require substantial time to develop. Additionally, our finding that most of the somatic mtDNA mutations in mutator flies are acquired during development, coupled with the fact that at least most of these mutations presumably result from replication errors, indicates that there is likely relatively little mtDNA replication in adult flies. This conclusion is further bolstered by the finding that key components of the mtDNA replication apparatus, including PolG, the Twinkle helicase, and the mitochondrial biogenesis factor Spargel, are primarily expressed in the female ovary, and early in development when most tissue growth occurs [47,48]. The importance of early arising mtDNA mutations to aging phenotypes may be conserved in vertebrates. Young mutator mice also display a dramatic increase in mtDNA mutation frequency relative to controls, but the mtDNA mutation frequency increases little with age, consistent with the model that these mutations occur primarily during development [15]. Previous work testing whether negative selection acts against deleterious mtDNA mutations has led to conflicting findings. Studies in cultured cells have demonstrated that pharmacological and genetic perturbations that activate mitophagic pathways can select against certain severe heteroplasmic mtDNA mutations [29–31]. Work in the nematode C. elegans has also shown that autophagy is required for the elimination of radiation-induced mtDNA damage [49], and that inactivation of the mitophagy-promoting factor Parkin results in increased abundance of point mutations in a mitochondrial mutator background and increased abundance of a deleterious mtDNA deletion when the mitochondrial unfolded protein stress pathway is activated [50–52]. Overexpression of autophagy-promoting factors in Drosophila also reduced the frequency of a heteroplasmic deletion created by expression of a mitochondrially-targeted restriction endonuclease [53]. Studies in mice expressing a proofreading-defective mtDNA polymerase also indicate that mitochondrial turnover is increased relative to controls [54]. However, reducing the activity of PINK1 and Parkin in an otherwise WT C. elegans genetic background did not significantly influence the frequency of point mutations or a large mtDNA deletion, suggesting that this pathway does not ordinarily select against deleterious mtDNA mutations [50–52]. Moreover, mutator mice do not exhibit an altered NS/S mutation ratio relative to WT mice [20,26,55], and mutator mice lacking the mitophagy-promoting factor Parkin do not exhibit an increased mtDNA mutation frequency or an altered NS/S mutation ratio relative to mutator mice [26]. Previous work in Drosophila also suggests that negative selection does not act against a heteroplasmic mtDNA mutation in somatic tissues in the absence of extreme measures to induce autophagy [53,56], and our current findings are fully consistent with this observation. In vitro studies indicate that PINK1 and Parkin selectively target depolarized mitochondria for lysosomal degradation [57]. Thus, one possible explanation for the apparent absence of negative selection opposing the accumulation of deleterious mtDNA mutations is that the mitochondria that bear these mutations are not sufficiently depolarized to trigger activation of the PINK1-Parkin pathway in vivo. Even in the event of a mtDNA mutation sufficient to trigger severe depolarization, ATP synthase (Complex V) is capable of coupling the hydrolysis of ATP to maintain partial membrane potential [31,58]. Alternatively, the fusion of mitochondria bearing mutations with healthy mitochondria containing WT genomes may allow deleterious mtDNA mutations to evade negative selection through genetic complementation. Indeed, mitochondrial stress frequently elicits mitochondrial fusion as a compensatory response [59]. While such compensatory mechanisms may prove useful when mtDNA mutations are present at low abundance, the phenotypes associated with high mutational loads in worms, flies, mice and humans indicate that these potential compensation pathways are incapable of fully preventing the deleterious consequences of a high mtDNA mutational load. While we detect no evidence of negative selection acting against deleterious mtDNA mutations, a simple absence of negative selection does not fully explain our findings. If no selective forces were acting on mtDNA, the frequency of deleterious mtDNA mutations in mutator flies should match predictions from our simulations of neutrality. However, we found that deleterious mtDNA mutations were consistently overrepresented in mutator flies. One possible explanation for this finding is that there is positive selection for deleterious mtDNA mutations, which has previously been reported in vertebrates [60,61]. Because many of the mtDNA mutations subjected to positive selection in vertebrates reside in or near the sequences that control mtDNA replication, it has been proposed that these mutations confer a replicative advantage, thus accounting for their overrepresentation [61]. However, a recent report has found strong evidence of positive selection acting on mutations that reside in protein coding sequences of human mtDNA with many of these variants appearing to result in deleterious effects on protein function [60]. One potential explanation for the excess accumulation of deleterious mutations in mtDNA coding sequences is offered by the “survival of the slowest model” [62]. In brief, this model posits that the mitochondrial quality control apparatus selectively targets oxidatively damaged mitochondria for degradation. According to this model, mitochondria that bear defective genomes are less metabolically active than fully functional WT mitochondria, and therefore less prone to damage from reactive oxygen species. This in turn makes these mutant-bearing mitochondria less prone to targeted degradation by quality control surveillance relative to fully functional mitochondria. There are also at least two alternative models to explain the overabundance of deleterious mtDNA mutations in mutator flies. One possible explanation is that the exonuclease-deficient polymerase induces mutations in a sequence context-dependent fashion and that these contexts are enriched at critical residues. Although our analyses do not offer support for this model, we only examined the nucleotides immediately neighboring mutation sites, leaving open the possibility that other features of sequence context explain the overabundance of deleterious mutations in mutator flies. A second alternative model to explain the overabundance of deleterious mtDNA mutations in mutator flies is that the subset of cells that acquire deleterious mtDNA mutations may compensate for the presence of these defective genomes by inducing mitochondrial biogenesis. If the mitochondrial biogenesis machinery is incapable of distinguishing between mitochondria with defective and WT genomes, this phenomenon would inadvertently result in a higher replication frequency of deleterious mutations relative to benign mutations because replication would be selectively induced in cells that bear deleterious mutations. Although we do not detect overt evidence of increased mitochondrial biogenesis in mutator flies (Fig 1D), our model posits that mitochondrial biogenesis would only be induced in a subset of cells, and this modest level of induction may not be readily detectable on a macroscopic level. Two recent studies in C. elegans offer potential support for this model by showing that worm strains bearing a heteroplasmic mtDNA deletion maintain tight regulation of WT mtDNA abundance, but that the abundance of the deletion can vary dramatically between individuals in the population [50,51]. These findings suggest that mutant genomes can “hitchhike” to high frequency as a consequence of a compensatory mitochondrial biogenesis response to decreased mitochondrial activity. However, a difference between these studies and our current work is that the mitochondrial unfolded protein stress pathway appears be an active participant, if not the driver, of the high mtDNA deletion frequency in C. elegans [50,51], whereas we detected no evidence of mitochondrial unfolded protein stress pathway activation in mutator flies (S6 Fig). Future experiments will be required to more fully investigate the potential role of the mitochondrial unfolded protein stress pathway and other candidate pathways that may influence the frequency of deleterious mtDNA mutations in mutator flies. All experiments were performed with flies raised and maintained at 25°C on standard cornmeal-molasses food unless otherwise stated. The w1118 isogenic fly strain, Df (2L) BSC252 strain, and Df (2L) FDD-0428643 strain were obtained from the Bloomington Drosophila Stock Center. The Drosophila POLG genomic region was PCR amplified from genomic DNA with primers (5’- TAAATCAATGTGACCGCCGC and 5’- TGTCCTTGCCTTGGGAACTG) and cloned into TOPO vector (Invitrogen). The D263A mutation was introduced by PCR using primers (5’- CAATGTCTCCTACGCAAGGGCGCGACTGAAG and 5’- CTTCAGTCGCGCCCTTGCGTAGGAGACATTG). Kanamycin resistance selection cassette (loxp-kanamycin-loxp) was PCR amplified and cloned into C-terminus of dPOLG-D263A using the PmeI site on the TOPO vector. PCR primers containing 70 bp homology arms were used to generate linearized fragment for recombineering (5’-TACGGAGGAGTGTGTGGTCGCAGGTTGGACTTCAGTTGCCTTAAAGGATGTTTCCTTTATTAAAACGAGGATGCAGTTCCACCTGATCAG and 5’- TGCCTTGGGAACTGGGAAACGTATCGGCAACAGGATGCTTTAAATGCAAGGTTATTTAAAAACATAGTGATGTACAAGAAAGCTGGGTCG). The recombineering and Kanamycin excision process were performed using a published procedure [63,64]. To increase transgenesis efficiency, a 40 kb P[acman] clone with dPOLGD263A was generated from modified 100 kb P[acman] constructs. In brief, two 500 bp homology arms flanking the POLG 40 kb fragment were PCR amplified using primers (left arm: 5’- AGGCGCGCCTGTATTGCCTCAGCCGGTTG and 5’-CGCGGATCCTCGCTGTGTCGATAAGGAAC; right arm: 5’-CGCGGATCCGTTCGATTTGGTCAACCTGC and 5’- AACTTAATTAAGAGTCCAATGGGATTCCACA) and cloned into attB-P[acman]-ApR vector [63]. The 40 kb fragment with dPOLGD263A was then introduced into attB-P[acman]-ApR by recombineering. A list of the genes situated on this 40 kb fragment, along with their presumed functions is shown in S3 Table. To increase the copy number and allow for transgenesis, DNA from confirmed colonies was extracted and transformed into EPI300 cells. To create the PolGmut transgenic fly, modified 40 kb P[acman] constructs were integrated specifically into the 92F1 and 28E7 regions of the Drosophila genome using a φ31-mediated transformation protocol (BestGene Inc. , Rainbow Transgenic Flies, Inc.). One- to two-day-old flies were collected into vials in groups of up to 20 flies. Flies were transferred every two to three days onto fresh food and the number of deaths was recorded. Lifespans were repeated at least three times with a minimum of 350 flies per genotype. The seizure phenotype of moribund mutator flies was noted during the lifespan analysis. All survivorship data were calculated using R software. The R package “survival” was used to generate Kaplan-Meier survival curves. Genotypes were compared using the log-rank test to determine significance. Climbing was assayed using a modified protocol for the previously published rapid iterative negative geotaxis (RING) assay [65,66]. Briefly, clean vials containing up to 20 files were placed into the RING apparatus. Flies were manually tapped to the bottom of the vials and their climbing behavior recorded using a digital video camera. This procedure was repeated for a total of three trials per vial and a minimum of 60 flies per genotype at each time point. Still images were captured using iMovie software (iMovie' 11 v9. 0. 8. Apple Inc.) 3. 0s after the flies were tapped down. The vertical height attained by each fly was scored using Fiji software (an open-source image processing software) [67]. Genotypes were compared to age-matched controls using the Wilcoxon rank-sum test. Four adult flies were homogenized in 200 μl of lysis buffer (100 mM Tris, 4 mM EDTA) followed by flash freezing in liquid nitrogen. Samples were boiled for 3 min before centrifugation at 8000xg for 5 min. Supernatant was then diluted 50-fold with lysis buffer for ATP quantification. ATP levels were measured using a commercial ATP Determination Kit (Invitrogen) as previously described [68]. Cytochrome c oxidase was measured as previously described [69]. Homogenates were prepared from 4 male flies homogenized with a hand-held rotor (VWR) in PBS with 0. 1% Triton X-100 and protease inhibitor cocktail (Roche). Absorbance was measured with a SpectraMax M2 plate reader (Molecular Devices). Activities were normalized to total protein, and quantified using DC protein assay (Bio-Rad). To generate protein extracts for Western blot analyses, four male flies were collected and frozen in liquid nitrogen. Flies were thawed and manually homogenized using a pestle in 100μL of lysis buffer (50 mM Tris-HCl (pH 7. 4), 150 mM NaCl, 1% NP-40,10% glycerol, 10 mM NaF, 1 mM Na3VO4,100 μg/ml PMSF, Sigma protease inhibitor cocktail (Sigma #P8340) ). Cell debris was removed from the lysate through centrifugation at 13,000rpm for 15 minutes at 4°C. The protein lysate was boiled in SDS-PAGE sample buffer with 2% beta-mercaptoethanol for 10 minutes, and the resulting proteins subject to Western blot analysis. For Western blot analyses, extracts were separated by SDS-PAGE on 4–20% Bis-Tris gels (GenScript #M42012) and transferred onto PVDF membranes overnight. Membranes were blocked in iBind Flex Solution Kit (SLF2020) and Western blots were performed using the iBind Flex Western Device (SLF2000) according to the manufacturer’s protocol. Immunodetections were performed using the following antibodies: 1: 8000 mouse anti-Actin (#MAB1501, Chemicon/Bioscience Research Reagents), 1: 200 rabbit anti-Ref (2) P (Ab178440, Abcam), 1: 500 rabbit anti-HSP60 (#4870S, Cell Signaling Technology), 1: 1000 rabbit anti-GRP 75/mt-Hsp70 (sc-13967, Santa Cruz Biotechnology, Inc.). The secondary antibody anti-mouse HRP (BioRad) was used at 1: 1000 for actin. The secondary antibody anti-rabbit HRP (BioRad) was used at 1: 2000 for Ref (2) P, and at 1: 500 for HSP60 and GRP 75/mt-Hsp70. Signal was detected using Thermo Scientific electrochemoluminescence reagents. Quantification of western blot images was performed using Fiji software [67]. Western blot data were normalized using log-transformation to stabilize variance before means were compared using Student t-test. Each experiment was repeated with at least three biological replicates. TEM was performed as previously described with minor modifications [70]. Briefly, indirect flight muscles were dissected from 50-day-old control and 2xPolGmut flies and placed in fixative containing 2. 5% glutaraldehyde, and 2% paraformaldehyde in 0. 1 M sodium cacodylate buffer, pH 7. 4, and incubated overnight at 4 °C. Fixed tissues were then postfixed in 1% OsO4, dehydrated in an ethanol series, and embedded using Epon. Samples were subjected to ultra-thin sectioning at 70 nm and stained with 6% uranyl acetate and a Reynolds lead citrate solution before TEM examination. Grids were viewed using a JEOL JEM 1400 transmission electron microscope. Adult brain dissection, fixation, immunohistochemistry, and imaging were performed as described previously [71]. DA neurons were labeled with anti-TH antiserum (1: 50, Immunostar). Serial optical sections were taken at 1-μm intervals and the confocal image stacks were analyzed using Imaris software (Bitplane Inc). The number of TH-positive neurons within each of the major DA neuron clusters was determined by visual inspection of individual confocal Z-series images. One- to two-day-old male flies were collected and transferred into fresh vials every 2–3 days. Once flies reached the appropriate age for sequencing, heads were harvested using a razor blade, flash frozen in liquid nitrogen, and stored at -80°C. Total DNA was isolated from individual fly heads using the QIAamp DNA Micro isolation kit following the manufacturer’s instructions. DNA yield for a single head typically ranged between 20-30ng of total DNA. Total DNA was prepared for DS using a previously described protocol [72] with several modifications. Briefly, ~20 ng of total DNA was sonicated in 60 μL of nuclease-free ddH2O using a Covaris AFA system with a duty cycle of 10%, intensity of 5, cycles/burst 100, time 20 seconds x 5, temperature of 4°C. After sonication, each sample was subjected to end-repair and 3’-dA-tailing using the NEBNext Ultra End-repair/dA-tailing kit (New England Biolabs) according to the vendor’s instructions. Each sample was then ligated with 2 μL of 15 μM DS adapters, prepared as described [72] using the NEBNext Ultra Ligation kit (New England Biolabs) according to the manufacturer’s instructions. Each sample was then cleaned of excess adapters using AgenCourt AmpureXP magnetic beads and PCR amplified, as previously described [72]. After library construction, mtDNA was enriched for sequencing by targeted DNA capture using IDT xGen Lockdown probes (Integrated DNA Technologies) specific for non-repetitive and non-low complexity regions of the Drosophila mitochondrial genome, as designated by RepeatMasker (http: //www. repeatmasker. org), using the manufacturer’s instructions. Probe sequences are found in S2 Table. Duplex Sequencing adapters used in collecting data for analyzing mutation selection were chemically synthesized as a collaborative effort with Integrated DNA Technologies to develop a prototype synthesis method. The captured DNA samples were sequenced on an Illumina NextSeq500 using 150bp paired-end sequencing. The resulting reads were aligned against the Drosophila genome (BDGP Release 6 + ISO1 MT/dm6) using the Burrows-Wheeler Aligner and Samtools [73] coupled with a custom software workflow described previously [72]. Reads not uniquely mapping to the mitochondrial genome were excluded from further analysis. Reads mapping to a large repetitive region [ChrM: 5961. . 5983] were excluded from our analyses to avoid artifacts caused by misalignment. The breakpoints of the repetitive region were determined using the RepeatMasker Web Server v. 4. 0. 6 [74]. Sequence data has been uploaded to the Sequence Read Archive (SRA) repository, and can be accessed at SRA accession PRJNA495611. A heteroplasmy cutoff of 70% was applied to filter polymorphisms from the reference genome. After processing, we called unique somatic mutations by counting every mutation only once at each position of the genome. Total mutation frequency counts all mutations detected, including multiple occurrences at the same site. Spectrum data, mutation frequency by codon position, mutation context by GC content, and four-fold degenerate site analyses were performed using scripts developed in Python v. 2. 7. Parsing of the mitochondrial genome as well as GC content analyses were performed using Biopython [75]. All measures of mutation frequency are calculated as a fraction: mutation frequency = [total # mutations / total sequenced bases] of the indicated mutation type. For calculations of trinucleotide mutation frequency, the denominator of this equation is calculated for each of the 96 trinucleotide contexts by tabulating the sequencing depth at each nucleotide in the protein-coding sequence and then grouping according to its 3’ and 5’ flanking nucleotides. Scripts to analyze trinucleotide sequence context as well as trinucleotide mutation frequency were developed in Python v. 3. 4, and modified from a previously published duplex sequencing workflow [76]. Circular plots of the distribution of mutations were generated using the R package, ‘circlize. ’ Statistics were performed and graphs were generated in RStudio v1. 1. 383, Microsoft Excel for Mac v. 16. 10, and StataCorp Stata v. 12. 1. To search for evidence of selection, a bioinformatics workflow was developed in Python v. 2. 7 (https: //github. com/csamstag/mito-mutations) to analyze mtDNA mutation data obtained from duplex sequencing, and to run simulations of mutagenesis mimicking the mutation spectra obtained from sequencing mutator flies. Data for simulations were obtained by aggregating the results from sequencing four 1-day-old 1xPolgmut flies. Point mutations were identified as described above, except that multiple mutations of a given type at the same site were also included in our analyses. Monte Carlo simulations were performed to generate a distribution of random mutations for statistical comparison to experimental findings. To control for the observed mutational biases in our data, we designed our scripts to match parameters observed from sequencing mutator flies (i. e. , the same number of G: C to A: T mutations, G: C to T: A mutations, etc.). To account for variation in sequencing depth, random mutations were generated using a probability function that was weighted according to the total read depth at each position. Each round of simulation mirrored the spectrum and number of mutations observed in the protein-coding regions of mutator flies. Simulations were repeated 10,000 times, and each simulation was analyzed using three metrics designed to detect selective forces. To examine the relationship between mutation frequency and sequence conservation, we obtained the PhyloP scores for each position in the Drosophila melanogaster mitochondrial genome from the UCSC Genome Browser. These values were used to calculate the average PhyloP score of the mutations detected from sequencing mutator flies. Similarly, we calculated the average PhyloP values of the mutations identified from simulations performed as described above. Empirical p-values represent the fraction of time a simulation displayed an average PhyloP score greater than or equal to the average PhyloP score observed in mutator flies. To test whether selection influences the frequency of NS variants, we compared the average NS/S ratio obtained from sequencing mutator flies to the NS/S ratios obtained from simulations. We performed simulations as described above and binned those mutations that map to coding sequences according to whether they induce NS or S alterations. We then calculated the average NS/S ratio for each simulation. The empirical p-value reflects the fraction of simulations in which the NS/S ratio was greater than or equal to the average NS/S ratio observed in 1xPolgmut flies. A similar approach was used to compare the NS/S ratio in the COX1 gene from sequencing mutator flies to the distribution of NS/S ratios from simulations of mutagenesis of the COX1 gene. Simulated mutagenesis of the COX1 gene was performed as described above, including adjustments for sequencing bias, and sequencing depth. To test whether selection influences the frequency of pathogenic mutations, we used MutPred [36] software to calculate pathogenicity scores for all NS variants detected in mutator flies, as well as all NS mutations generated from simulations. The average MutPred score from mutator flies was then compared to a distribution of MutPred scores from simulations. The empirical p-value was determined as the fraction of simulations in which the average MutPred score was greater than or equal to the average MutPred score observed in mutator flies. A similar approach was used to analyze the pathogenicity of NS mutations occurring within the COX1 gene. Adjustments were made to simulations to account for mutational bias and sequencing depth within the COX1 gene.

The energy needs of an animal cell are supplied by tiny organelles known as mitochondria. Each of the many mitochondria in a cell has a set of blueprints for making more mitochondria, known as mitochondrial DNA (mtDNA). As animals age, their mtDNA acquires irreversible defects called mutations, which accumulate and may cause aging. Cells can selectively destroy malfunctioning mitochondria, so we hypothesized that mitochondria with harmful mutations would be selectively destroyed. To test our theory, we created a fruit fly strain with a high mtDNA mutation rate. Our hypothesis predicts that, because mitochondria bearing harmful mtDNA mutations would be destroyed, we should detect primarily less harmful mutations in our strain. However, the mtDNA mutations we detected were more harmful than expected by chance. We suggest two possible explanations: First, mitochondria with harmful mtDNA mutations may be degraded less often because they generate little energy and are not damaged by toxic byproducts of energy production. Second, cells may compensate for harmful mtDNA mutations by stimulating mitochondria to multiply, creating more healthy mitochondria but also more mitochondria with harmful mtDNA mutations. Future studies will distinguish between these models and further advance our understanding of aging and aging related disease.

Abstract Introduction Results Discussion Materials and methods

invertebrates deletion mutation mitochondrial dna animals invertebrate genomics animal models mutation substitution mutation drosophila melanogaster model organisms experimental organism systems forms of dna mitochondria dna bioenergetics cellular structures and organelles drosophila research and analysis methods genomics animal studies insects animal genomics arthropoda biochemistry point mutation eukaryota cell biology nucleic acids gene identification and analysis genetics mutation detection biology and life sciences energy-producing organelles organisms

2018

Deleterious mitochondrial DNA point mutations are overrepresented in Drosophila expressing a proofreading-defective DNA polymerase γ

Schistosoma mansoni and S. haematobium are co-endemic in many areas in Africa. Yet, little is known about the micro-geographical distribution of these two infections or associated disease within such foci. Such knowledge could give important insights into the drivers of infection and disease and as such better tailor schistosomiasis control and elimination efforts. In a co-endemic farming community in northern Senegal (346 children (0–19 y) and 253 adults (20–85 y); n = 599 in total), we studied the spatial distribution of S. mansoni and S. haematobium single and mixed infections (by microscopy), S. mansoni-specific hepatic fibrosis, S. haematobium-specific urinary tract morbidity (by ultrasound) and water contact behavior (by questionnaire). The Kulldorff' s scan statistic was used to detect spatial clusters of infection and morbidity, adjusted for the spatial distribution of gender and age. Schistosoma mansoni and S. haematobium infection densities clustered in different sections of the community (p = 0. 002 and p = 0. 023, respectively), possibly related to heterogeneities in the use of different water contact sites. While the distribution of urinary tract morbidity was homogeneous, a strong geospatial cluster was found for severe hepatic fibrosis (p = 0. 001). Particularly those people living adjacent to the most frequently used water contact site were more at risk for more advanced morbidity (RR = 6. 3; p = 0. 043). Schistosoma infection and associated disease showed important micro-geographical heterogeneities with divergent patterns for S. mansoni and S. haematobium in this Senegalese community. Further in depth investigations are needed to confirm and explain our observations. The present study indicates that local geospatial patterns should be taken into account in both research and control of schistosomiasis. The observed extreme focality of schistosomiasis even at community level, suggests that current strategies may not suffice to move from morbidity control to elimination of schistosomiasis, and calls for less uniform measures at a finer scale. Schistosomiasis is amongst the most common human parasitic diseases with over 230 million people affected worldwide [1]. More than 90% of them live in sub-Saharan Africa [2]. The two major species are Schistosoma mansoni and S. haematobium, which are co-endemic in many regions [3]. However, little is known about the geographical distribution of both species within such co-endemic regions. Knowledge on micro-geographical variations of single and mixed Schistosoma infections and associated disease could provide important insights into the drivers of infection and disease and as such better tailor schistosomiasis control and elimination efforts. Recent progress in geographic information systems (GIS) has facilitated a better understanding of geospatial dimensions of schistosomiasis on the large scale. On continental and national scales, climatic (e. g. temperature and rainfall) and physical factors (e. g. vegetation, large water bodies, altitude) have been identified as major determinants of the heterogeneous geographical distribution of Schistosoma infection (either S. mansoni or S. haematobium, e. g. [4]–[13]). On subnational levels, distance to water contact sites, land use and the distribution of infected snails have been reported to contribute to these heterogeneities (e. g. [14]–[20]). Few studies have however exploited these techniques to address the geospatial dimensions of schistosomiasis on the micro-scale, i. e. within communities or among households [21]–[29]. Most of these considered spatial patterns of only one Schistosoma species even though S. mansoni and S. haematobium often occur together [3]. Moreover, micro-geographical clustering of Schistosoma infection has never been studied in relation to Schistosoma-specific morbidity. In the present study, we set out to investigate the spatial patterns of S. mansoni and S. haematobium infection and morbidity in a co-endemic community on the bank of Lake Guiers in the north of Senegal [30], [31]. During the past decades, many communities around Lake Guiers (‘Lac de Guiers’) in the north of Senegal have become co-endemic for S. mansoni and S. haematobium [32]–[35]. Schistosoma mansoni was introduced in Richard-Toll in 1988 upon construction of the Diama dam and rapidly spread throughout the region [36], [37]. By 1994, virtually the whole Lake Guiers area had become exposed to this species [38]. Today, both S. mansoni and S. haematobium are wide-spread in the communities around the lake, and the situation is still dynamic. This study was part of a larger investigation on the epidemiology of schistosomiasis and innate immune responses (SCHISTOINIR: www. york. ac. uk/res/schistoinir) for which approval was obtained from the review board of the Institute of Tropical Medicine, the ethical committee of the Antwerp University Hospital and ‘Le Comité National d' Ethique de la Recherche en Santé’ in Dakar. Informed and written consent was obtained from all participants prior to inclusion into the study. For minors, informed and written consent was obtained from the legal guardian. Participants with severe pathology that needed further treatment were referred to the appropriate health authority. After the study, all community members were offered praziquantel (40 mg/kg) and mebendazole (500 mg) to treat and prevent schistosomiasis and soil-transmitted helminthiasis, respectively [30], according to WHO guidelines [39]. This cross-sectional study was conducted from July until November 2009 in Diokhor Tack (16°11′24″N 15°52′48″W), the largest community on the Nouk Pomo peninsula in Lake Guiers. Details on the study area have been described elsewhere [30]. In short, it is an isolated, compact and homogeneous Wolof community of Muslim faith with a surface of ∼0. 25 km2. Cultivation is the main means of subsistence and the farmlands that surround the village are irrigated with water from the lake. Although the water from Lake Guiers is piped to the capital city of Dakar, 250 km away [40], the people living nearby do not have access to safe water. Water contact takes place in the lake or in specific sites in canals that are connected to the lake in the west (Figure 1). There were no periodic anthelminthic treatment programs prior to our study and the community does not have a health facility. The nearest ‘health post’ is ∼12 km away. All community members that gave informed consent (or their legal guardians) were included in the study. Participants were registered and recruited from door to door for the parasitological and ultrasound surveys. The community consisted of 71 households, 68 of which participated in this study (Figure 1). This corresponded to a total study population of 599 individuals. For microscopic diagnosis of Schistosoma infection, two feces and two urine samples were collected from each participant on consecutive days. For each feces sample, two Kato-Katz slides of 25 mg fecal material each were prepared, and urine samples were filtered and processed according to standard procedures, as previously described [30], [31]. In analogy with earlier micro-geographic studies [21]–[23], S. mansoni and S. haematobium infection densities were expressed as the number of eggs detected per gram of feces (epg) or per 10 ml of urine (ep10ml), respectively, including both negative (0 epg or 0 ep10ml) and positive individuals [41]. Single infection was defined as passing eggs of only one species, and mixed infection as passing eggs of both S. mansoni and S. haematobium, irrespective of the route of egg elimination [30]. Schistosoma-specific morbidity was determined by ultrasound, as previously described [31]. Pathologic lesions associated with S. haematobium or S. mansoni infection were recorded according to the Niamey guidelines [42]. Individuals with signs of hepatic morbidity that were not specific to S. mansoni (e. g. hepatitis, cirrhosis or fatty liver) were excluded [42]. To assess the presence or absence of S. mansoni-specific hepatic fibrosis, the liver image pattern was determined [42]. Liver image patterns of C (“periportal fibrosis possible”) to F (“very advanced periportal fibrosis”) were categorized as S. mansoni-specific hepatic fibrosis [31]. Individuals with liver image pattern A (“no sign of periportal fibrosis”) or B (“incipient periportal fibrosis not excluded”) were categorized as controls [42]. To assess the presence or absence of S. haematobium-specific urinary tract morbidity, the urinary bladder score was determined [42]. A score of ≥1 was considered as S. haematobium-specific bladder morbidity in accordance with previous studies [31], [43], [44]. The severity of morbidity was represented by the liver image pattern score for S. mansoni- and by the upper urinary tract score for S. haematobium-specific morbidity [42]. Finally, individual questionnaires were used to explore water contact behavior in a random subsample of people older than 5 years of age. Water contact sites as well as the center of each household were located using a hand-held differential global positioning system with an accuracy of 3 m (Garmin Etrex H). Household locations in latitude and longitude were then linked to the collected individual infection, morbidity and questionnaire data (multiple observations per location). These data were imported into SaTScan 9. 1. 1 (Software for the spatial and space-time scan statistics, developed by M. Kulldorff, Harvard Medical School, Boston and Information Management Services Inc. , Silver Spring, Maryland, USA. Available at www. satscan. org) according to the software' s user guide [45]. ArcMap 9. 3 (ESRI, Redlands, California, USA) was used to project the geographic coordinates and statistically significant clusters (see below) on to the Universal Transverse Mercator zone 28N (1984 datum). The widely used Kulldorff' s scan statistic in SaTScan™ tests whether events such as disease cases are distributed randomly in space and, if not, identifies the approximate location of significant geospatial clusters [46]. The test uses a moving circular window that varies up to a predefined size. Each window is a potential cluster. For each window, a likelihood ratio test is applied based on the observed and expected number of cases inside and outside the window to test the null hypothesis of absolute spatial randomness against the alternative hypothesis that there is an elevated risk within the window as compared to outside. The window with the maximum likelihood is the ‘most likely cluster’. The p-value of the maximum likelihood ratio test statistic was obtained after 999 Monte Carlo replications. A maximum window size of 50% of the study population was chosen upon sensitivity analysis using maximum sizes from 10 to 50%. Only statistically significant (p<0. 05) most likely clusters were reported, and standard settings (i. e. non-overlapping secondary clusters) were used throughout all analyses. In case the most likely (significant) cluster contained only one household, an additional check was performed to increase the robustness of cluster detection. The standard analysis was repeated while allowing for overlapping secondary clusters (using the “criteria for reporting secondary clusters” option “no restriction = most likely cluster for each grid point” [45]), and the secondary cluster (including the first household) was reported, if it remained significant. Additionally, clusters with p<0. 06 were displayed to indicate households that tended to have increased risks. Infection densities of S. mansoni and S. haematobium showed skewed distributions, and were therefore normalized by log (base 10) -transformation after adding half of the detection limit to allow for zeros. The detection limit for S. mansoni infection was 10 epg and that for S. haematobium infection 0. 5 ep10ml. The spatial distribution of log-transformed infection densities was assessed using normal models [47]. Geometric mean (GM) infection densities in- and outside spatial clusters were computed to quantify significant spatial heterogeneities. Subsequently, Bernoulli models [46] were run to investigate the distribution of single S. mansoni, single S. haematobium and mixed infections, comparing spatial distributions of people with: The spatial distribution of the prevalence of hepatic fibrosis and urinary tract morbidity was tested using binary variables in separate Bernoulli models. Ordinal models were used to assess the distribution of the severity of S. mansoni- and S. haematobium-specific morbidity [48]. Relative risks (RR) comparing people in- and outside clusters, as well as prevalences in- and outside clusters were calculated to quantify significant spatial heterogeneities based on Bernoulli and ordinal models. Gender and age are important risk factors for both Schistosoma infection [30], and morbidity [31]. To investigate whether these demographic factors 1) caused clustering of infection and morbidity, and/or 2) impacted on the size and exact locations of statistically significant clusters, the abovementioned analyses were adjusted using multiple datasets [49]. Six datasets were prepared, containing either males or females from 0 to 9,10 to 19, or ≥20 years old (Table 1). SaTScan™ incorporated all datasets into a single log likelihood function. This function is defined as the sum of the individual log likelihoods for those data sets for which the observed case count is more than the expected. Since this adjustment was only possible for the Bernoulli and normal models, separate Bernoulli models were run for the ordinal morbidity model showing significant spatial heterogeneities in the unadjusted analysis. Finally, Bernoulli models were used to compare the geospatial distribution of people reporting to frequent a particular water contact site versus the distribution of people who did not report to frequent that site. Complete parasitological data were obtained from a total of 599 individuals from 68 households. The median household size was 8 people (range 1–20). The total study population consisted of 302 males and 297 females with a median age of 15 (range 0–85) years. Ultrasound and questionnaire data were obtained from random subsamples of 291 individuals (64 households), and 277 individuals (63 households), respectively. The prevalence of overall S. mansoni infection was 55% (328/599) and that of S. haematobium 44% (261/599). Mixed infections were observed in 32% of the population (189/599). The prevalence of S. mansoni-specific hepatic fibrosis was 31% (90/291). Most cases had liver image pattern C (71/90), 9/90 had pattern D, while advanced periportal fibrosis was observed in 10/90 cases (nine with liver image pattern E and one with F). The prevalence of S. haematobium-specific urinary tract morbidity was 80% (233/291). Positive upper urinary tract scores (range 3–12) were observed in 6% of the study population (18/291). Distributions of single and mixed Schistosoma infections, S. mansoni-associated hepatic fibrosis and S. haematobium-associated bladder morbidity according to gender and age are summarized in Table 1. Figure 2A depicts the heterogeneous geospatial distribution of S. mansoni and S. haematobium infection densities (p = 0. 001 for both unadjusted analyses). While the size of the S. haematobium infection density cluster increased upon correction for the spatial distribution of gender and age, both S. mansoni and S. haematobium clusters remained statistically significant (Figure 2B; p = 0. 002 and p = 0. 023, respectively). Participants with mixed and those with single S. haematobium infections were randomly distributed (p = 0. 16 and p = 0. 080, respectively), while those with single S. mansoni infections tended to cluster geographically (Figure 3A; RR = 1. 7; p = 0. 053). Figure 3B indicates that the clustering of single S. mansoni was independent of the spatial distribution of gender and age (p≤0. 050), although the cluster size and exact location were slightly altered upon adjustment. Figure 4A indicates that people with hepatic fibrosis (RR = 1. 9; p = 0. 054) and urinary tract morbidity (RR = 1. 2; p = 0. 053) tended to cluster in the same area. Adjusted analysis however indicated that these heterogeneous patterns were dependent on the distribution of gender and age (p = 0. 087 for hepatic fibrosis and p = 0. 071 for urinary tract morbidity in the adjusted analysis). In order to assess the distribution of morbidity by severity, ordinal analyses were performed for liver image pattern and upper urinary tract scores. Figure 4B shows that this resulted in one significant cluster (p = 0. 001) in which the RR increased with the severity of hepatic fibrosis: the RR for a healthy liver image pattern A was 0. 3, that for B 1. 3, for C 1. 4, for D 2. 7 and for E 4. 3. Moreover, the only person with pattern F in the community lived in this cluster. Bernoulli models were used to investigate whether this cluster of severe hepatic fibrosis was independent of the distribution of gender and age. Since more severe hepatic fibrosis was only observed in adults, these analyses were restricted to ≥20-year-olds. Unadjusted Bernoulli models revealed that image patterns D–F (as opposed to A–C) clustered in the households within the hepatic fibrosis cluster that were closest to water contact site IV (RR = 6. 3; p = 0. 043; data not shown). The combined distribution of patterns E and F (as opposed to A–D) was homogeneous (p = 0. 20). In the adjusted model, the cluster of pattern D–F remained statistically significant (Figure 4B; p = 0. 031). The risk of severe urinary tract morbidity was homogeneously distributed (p = 0. 38 in the ordinal analysis). Water contact activities were concentrated at site IV with 62% of the interviewees (172/277) reporting to frequent this site. Numbers of observations in the other sites were limited. Nonetheless, spatial analysis of the questionnaire data revealed significant heterogeneities in the self-reported use of the different water contact sites (Figure 5). People from two adjacent households in the northeast were more likely to frequent site II than those from the rest of the community (3/4 in- versus 5/273 outside the cluster; p = 0. 005). People living in the center of the community were more likely to frequent site III than others (4/17 versus 2/260; p = 0. 022). Those from the southwest were more likely to frequent site V (8/53 versus 1/224; p = 0. 001). Use of site I and IV did not appear to be linked to a particular group of households (p = 0. 31 and p = 0. 16 respectively). The present micro-geographical study revealed significant clusters of S. mansoni and S. haematobium infection density in different sections of one community in a co-endemic area, possibly related to heterogeneities in the use of different water contact sites. While the distribution of urinary tract morbidity was homogeneous, a strong geospatial cluster was found for severe hepatic fibrosis. Particularly those people living adjacent to the most frequently used water contact site were more at risk for advanced morbidity than those living farther away. These findings confirm the well-known focality of schistosomiasis [50]. Even within one community, one cannot assume the risk of schistosomiasis to be homogenous. More remarkably even, the two Schistosoma species clustered in different sections of the community; Schistosoma mansoni infections clustered in the north while S. haematobium clustered in the south. A series of recent GIS studies showed significant micro-geographical heterogeneities in S. haematobium infection within a mono-endemic Kenyan community [21], [22]. Those in S. mansoni mono-endemic communities showed conflicting results with heterogeneous spatial patterns in some studies and homogeneous patterns in others [24]–[29]. To our knowledge, only Farooq et al have so far investigated the spatial distribution of both infections in a co-endemic community in Egypt in the 1960s. They reported higher infection levels of S. mansoni in small children in one section of the community and higher levels of S. haematobium in another section [51]. This is in agreement with the divergent distributions of S. mansoni and S. haematobium infection densities observed in the present study. Several interrelated factors may underlie these observations, and are discussed below. First of all, the micro-geographic distributions of the intermediate snail host of S. mansoni and S. haematobium, which belong to the genus Biomphalaria and Bulinus, respectively, may be divergent as well. Unfortunately, it was logistically impossible to collect snails in the present study. Yet, it is known that these snail species prefer different niches and that their distribution is influenced by chemical, physical, and biological factors [52]–[57]. Indeed, Woolhouse and Chandiwana demonstrated that the snail hosts of S. mansoni and S. haematobium occupy different locations in one single habitat in a co-endemic focus [53]. Ecological factors may thus have favored S. mansoni transmission in the north and S. haematobium transmission in the south. At the human host level, behavioral factors may have played a role in the observed spatial pattern of infection. Although based on a small number of observations, our questionnaire data indeed indicated that people from the north and center were more likely to frequent the northern sites than other community members, whereas those from the southwest were more likely to use the southernmost site. It thus seems that the first group maintained S. mansoni transmission in the north and the second S. haematobium transmission in the south. This corresponds to the study of Woolhouse and Chandiwana reporting 1) a similar geospatial segregation of S. mansoni and S. haematobium infection in the snail host population between transmission sites, and 2) a very focal man-to-snail transmission, within a distance of 40 m. Interestingly, they proposed that these divergent patterns most likely reflected differences in the distribution of defecation from that of urination, favoring S. mansoni and S. haematobium transmission, respectively [53]. In contrast to water contact behavior, age and gender of the human host were shown to have a negligible impact on the divergent pattern of S. mansoni and S. haematobium infection. Spatial clustering in the different sections of the community remained significant upon correction for age and gender. Other factors that may have contributed to the spatial pattern include genetic differences in susceptibility to infection [58]–[60]. Indeed, extended families tended to live together in this community (L. Meurs, personal observation). Also, people from the same section/extended family are more likely to have similar behavioral patterns [51], [61]–[63]. In contrast to the spatial distribution of S. haematobium infection, the distribution of S. haematobium-associated urinary tract morbidity was homogeneous. This was unexpected as S. haematobium infection has consistently been reported as an independent risk factor for urinary tract morbidity [43], [44], [64]–[71]. The fact that S. haematobium was only introduced in this region approximately 6 years prior to this study [72], may explain the relatively low severity of urinary tract morbidity in this community and the consequent absence of a spatial pattern. The severity of urinary tract morbidity is expected to progress over time with cumulative exposure to S. haematobium eggs [73]. On the other hand, a strong geospatial cluster was found for severe S. mansoni-specific hepatic fibrosis which overlapped with that of S. mansoni infection density. At first sight, this seems to be in contrast with previous studies by this group showing that current S. mansoni infection is not associated with hepatic fibrosis [31], which usually develops after 5–15 years of exposure [3]. However, a closer look at the overlapping clusters showed that teenagers had the highest infection densities and contributed most to the S. mansoni infection density cluster (data not shown). Adults on the other hand had more advanced morbidity, and contributed most to the severe hepatic fibrosis cluster. This suggests that these adults were in fact the teenagers with the highest S. mansoni infection densities earlier in life. Moreover, the clustering of severe hepatic fibrosis in adults seemed to be associated with the distance to the water. Those living within ∼100 m of the major water contact site (Figure 4B) were at least six times more likely to develop advanced hepatic fibrosis (liver image pattern D–F) than those living further away. However, other factors cannot be excluded such as genetic predisposition [74], diet or nutritional status [75], or co-infections, which may have put those living in close vicinity of the water at a higher risk of developing hepatic morbidity than the rest of the community. To our knowledge, only Booth et al have so far investigated micro-geographical variations in Schistosoma-associated morbidity. They found an association between splenomegaly and the combined exposure to S. mansoni and Plasmodium falciparum but did not explicitly investigate spatial clustering [76]. To our knowledge this is the first study to quantify micro-geographical infection patterns of S. mansoni and S. haematobium in a co-endemic community, and the first to relate these to patterns of Schistosoma-specific morbidity. Apart from the strengths, it is also important to address some limitations of our study. First, the study was cross-sectional and the results were merely descriptive in nature. The present study was a first attempt to describe patterns of schistosome infection and morbidity on a micro-scale, and it was not designed to explain the underlying mechanisms of potential micro-geographical clustering. Based on the limited data that were available, we generated a number of hypotheses, but other risk factors, including environmental, malacological, genetic, immunological and socio-economic factors, should be included in future studies. In addition, more spatial as well as spatio-temporal studies [21]–[23] are necessary to confirm our observations in other geographical areas and to explain them. Another limitation was that there is as yet no standard technique available to investigate spatial patterns. The emergence of various statistical methods has greatly boosted geospatial studies on schistosomiasis and increased our understanding of this disease. On the other hand, the large variety of methods has also hampered the comparison between the different micro-geographical studies that have be conducted so far and standardization is recommended. Current WHO schistosomiasis control strategies aim to prevent morbidity in later life through regular mass drug administration (MDA) to at risk populations in so-called homogeneous ecological zones [77], [78]. However, the strong micro-geographical clustering of infection and morbidity observed in the present study suggests that less uniform strategies should be developed to better tailor control efforts at the local level. A more targeted approach will be even more relevant in view of resolution WHA65. 21 on the elimination of schistosomiasis, recently adopted by the WHO [77]. It is expected that MDA alone cannot break the Schistosoma life cycle and that complementary interventions will have to be put in place [79]. Micro-geographical studies will help to get much needed insights into local transmission dynamics of S. mansoni and S. haematobium and hence to develop sustainable control and elimination strategies [80].

In the developing world, over 230 million people are infected with parasitic Schistosoma worms. Schistosoma mansoni and S. haematobium are the most abundant species in Africa, affecting the liver and urinary tract, respectively. Both parasites are spread through infested freshwater. Although it is known that the disease occurs focally within countries or regions, little is known on its geographic spread on a smaller scale. Here, we examined 599 people from a community in northern Senegal for S. mansoni and S. haematobium infections and related abnormalities of the liver and urinary tract. We recorded where they lived and where they had water contact and visualized this information in geographical maps. The study showed that each Schistosoma species clustered in a different section of the community, and that liver abnormalities were more severe near the mostly used water contact site. So far, this is the first study to investigate the geographical spread of both species in a single community, and the first to map schistosomal disease on such a small scale. Further studies are needed to confirm and explain these results. They could contribute to a better understanding of schistosomiasis and have important consequences for the control and elimination of this disease.

Abstract Introduction Methods Results Discussion

2013

Micro-Geographical Heterogeneity in Schistosoma mansoni and S. haematobium Infection and Morbidity in a Co-Endemic Community in Northern Senegal

Viral persistence is associated with hierarchical antiviral CD8 T cell exhaustion with increased programmed death-1 (PD-1) expression. In HCV persistence, HCV-specific CD8 T cells from the liver (the site of viral replication) display increased PD-1 expression and a profound functional impairment that is not reversed by PD-1 blockade alone. Here, we report that the inhibitory receptor cytotoxic T lymphocyte-associated antigen-4 (CTLA-4) is preferentially upregulated in PD-1+ T cells from the liver but not blood of chronically HCV-infected patients. PD-1/CTLA-4 co-expression in intrahepatic T cells was associated with a profound HCV-specific effector dysfunction that was synergistically reversed by combined PD-1/CTLA-4 blockade in vitro, but not by blocking PD-1 or CTLA-4 alone. A similar effect was observed in circulating HCV-specific CD8 T cells with increased PD-1/CTLA-4 co-expression during acute hepatitis C. The functional response to combined blockade was directly associated with CTLA-4 expression, lost with CD28-depletion and CD4-independent (including CD4+FoxP3+ Tregs). We conclude that PD-1 and CTLA-4 pathways both contribute to virus-specific T cell exhaustion at the site of viral replication by a redundant mechanism that requires combined PD-1/CTLA-4 blockade to reverse. These findings provide new insights into the mechanisms of virus-specific T cell dysfunction, and suggest that the synergistic effect by combined inhibitory receptor blockade might have a therapeutic application against chronic viral infection in vivo, provided that it does not induce autoimmunity. Virus-specific CD8 T cells become progressively exhausted during chronic viral infection due to increased level or duration of antigenic stimulation without sufficient CD4 help[1]. Among the CD28 family of costimulatory molecules, programmed death-1 (PD-1) is an immune inhibitory receptor that is highly expressed on both exhausted and activated T cells[2]. Interactions between PD-1 and its ligands PD-L1/PD-L2 can inhibit antigen-specific T cell proliferation and effector function[2], [3]. Importantly, blockade of PD-1 signaling can restore function to exhausted virus-specific CD8 T cells with reduced viral load in mice with chronic lymphocytic choriomeningitis virus (LCMV) infection in vivo[4], thereby raising the possibility that immune exhaustion can be reversed with potentially therapeutic antiviral effects. A role for PD-1 pathway in viral persistence and antiviral T cell exhaustion has been shown in various chronic viral infections including hepatitis B virus (HBV), human immunodeficiency virus (HIV), simian immunodeficienty virus (SIV) and hepatitis C virus (HCV) [2], [5], [6], [7], [8]. In particular, HCV is a highly persistent human pathogen that infects the liver and causes significant morbidity and mortality due to chronic liver disease[9]. Patients with chronic HCV infection harbor dysfunctional antiviral T cells with increased PD-1 expression in circulating blood, and PD-1 blockade can restore their antigen-specific effector function in vitro [10], [11], [12], [13], [14]. However, HCV-specific CD8 T cells in the liver (the site of HCV infection) display markedly increased PD-1 expression compared to peripheral blood [10], [13], [15] and a profound functional impairment that is refractory to PD-1 blockade alone[13]. Similarly, highly activated circulating HCV-specific CD8 T cells in acute evolving hepatitis C show markedly increased PD-1 expression with a deep functional impairment that is unresponsive to PD-1 blockade. These results suggested the existence of additional inhibitory mechanisms that contribute to virus-specific CD8 T cell exhaustion in HCV-infected patients, particularly in PD-1high cells. Since intrahepatic PD-1+ CD8 T cells also express increased levels of immune inhibitory receptor cytotoxic T lymphocyte-associated antigen-4 (CTLA-4) [13], we asked if CTLA-4 might contribute to virus-specific T cell dysfunction in HCV-infected patients. We show that CTLA-4 is preferentially co-expressed in PD-1high CD8 T cells (particularly HCV-specific CD8 T cells) in peripheral blood during acute hepatitis C and in the liver during chronic HCV infection. PD-1/CTLA-4 co-expression was associated with marked antigen-specific effector T cell dysfunction that was dramatically and synergistically reversed by combined PD-1/CTLA-4 blockade in vitro. The response to combined PD-1/CTLA-4 blockade was directly associated with CTLA-4 expression, independent of CD4 T cells including FoxP3+ Tregs and dependent on CD28 expression. Collectively, these findings suggest that both CTLA-4 and PD-1 pathways contribute to HCV-specific T cell exhaustion in a distinct manner in HCV infection and that combined inhibition of CTLA-4 and PD-1 pathways may have potential therapeutic application in reversing immune exhaustion. In chronically HCV-infected patients (C), CTLA-4 expression was greater in CD8 T cells from liver infiltrating lymphocytes (LIL) compared to peripheral blood lymphocytes (PBL) (p<0. 0001) (Figure 1A). This compartmental difference was further amplified in HCV-specific CD8 T cells both in MFI (p = 0. 023) and percentages (p = 0. 005), but not in CD8 T cells specific for influenza virus (Flu), cytomegalovirus (CMV) or Epstein-Barr virus (EBV) epitopes (p>0. 3) (Figure 1B). These expression patterns mirrored PD-1 expression, which is also upregulated in intrahepatic HCV-specific CD8 T cells[10], [12], [13], [16]. Of note, CTLA-4 expression was low in CD8 T cells from liver explants of two HCV-seronegative patients with nonalcoholic steatohepatitis and alcoholic liver disease (Figure S1). In peripheral blood, CTLA-4 expression levels were uniformly low in total, HCV-specific and non-HCV-specific CD8 T cells during chronic or resolved (R) HCV infection; this contrasts with PD-1 expression, which is elevated in circulating HCV-specific CD8 T cells from chronic compared to resolved patients[13]. CTLA-4 expression was also upregulated in total and HCV-specific (but not non-HCV-specific) CD8 T cells from the peripheral blood of patients with acute hepatitis C (Figure 1A/B). Figure 1C shows the representative PD-1 and CTLA-4 expression for peripheral blood and intrahepatic CD8 T cells specific for HCV and non-HCV epitopes detected by HLA-A2/peptide tetramers. Of note, CTLA-4 was preferentially expressed in PD-1-high, but not PD-1-intermediate or PD-1-negative CD8 (Figure 1D) and CD4 T cells (data not shown) with a strong association between PD-1 MFI and CTLA-4 expression in percentage (R = 0. 81, p<0. 0001) (Figure 1E) as well as MFI (R = 0. 69, p<0. 0001, data not shown). Taken together, CTLA-4 expression is induced early with PD-1 in HCV-specific CD8 T cells during acute hepatitis C but becomes compartmentalized to the liver with chronic infection. Intrahepatic CD8 T cells were further examined for expression levels of additional CD28 family receptors. As shown (Figure 2A), CD28 was highly expressed in PD-1+CTLA-4+ subset, compared to PD-1+CTLA-4− or PD-1−CTLA-4− subsets (median 62% vs 49% vs 33%, p<0. 0001). By contrast, ICOS and BTLA expression levels were generally low, although a slight increase in ICOS expression was observed in PD-1+CTLA-4+ subset compared to others (median 2. 4% vs 0. 5% vs 0. 1%, p = 0. 049). Thus, intrahepatic CD8 T cells may be subject to inhibitory signals from PD-1 and CTLA-4 as well as a positive signal from CD28 with little contribution from ICOS or BTLA. CTLA-4 expression was also increased in CD4 T cells from the liver compared to peripheral blood (p<0. 0001) (Figure S3A). Since CD4+FoxP3+ Tregs are upregulated in HCV-infected patients and they co-express CTLA-4[17], we asked whether FoxP3+ Tregs accounted for increased CTLA-4 expression in intrahepatic CD4 T cells. However, FoxP3+ Treg frequencies did not differ between the liver and blood compartments (p = 0. 209) (Figure S3B/C); furthermore, FoxP3−CTLA-4+ CD4 T cells were enriched by 2–3 fold in the liver compared to blood (Figure S3D). Thus, intrahepatic CD4 and CD8 T cells in HCV-infected patients display increased CTLA-4 expression without increased FoxP3+ Treg frequency. We previously showed that PD-1 blockade failed to restore function to highly PD-1+ HCV-specific CD8 T cells from HCV-infected liver, although it enhanced the functionality of PD-1 intermediate CD8 T cell from blood[13]. As CTLA-4 was preferentially expressed in PD-1high CD8 T cells (Figure 1D), we asked if CTLA-4 blockade might reverse their dysfunction, either alone or combined with αPD-L1. To this end, intrahepatic and peripheral lymphocytes from HCV-infected patients were cultured with 15mer peptides spanning the entire HCV NS3 or Flu matrix protein in the presence of blocking αPD-L1 and/or αCTLA-4 or isotype control antibodies. On day 7, the cultures were examined for antigen-specific IFN-γ and TNF-α expression by intracellular cytokine staining. As shown in Figure 3A, there was little to no HCV-specific CD4 or CD8 T cell cytokine expression in intrahepatic lymphocytes cultured with αPD-L1 or isotype control antibodies, although low level responses were occasionally seen with αCTLA-4 alone. Remarkably, combined blockade with αPD-L1 and αCTLA-4 resulted in a marked enhancement of intrahepatic HCV NS3-specific CD8 and CD4 T cell cytokine production from 4/6 patients. The comparison of total NS3-specific cytokine responses showed a significant difference between combined PD-1/CTLA-4 blockade and single PD-1 blockade (2. 3% vs 0%, p = 0. 0037 by Mann Whitney U). However, in blood, HCV-specific CD8 T cell cytokine response was augmented by αPD-L1 without further enhancement by αCTLA-4. Thus, combined PD-1/CTLA-4 blockade resulted in marked HCV-specific cytokine response in LIL but not PBL in HCV-infected patients: LIL 4/6 patients (67%) vs PBL 0/8 patients (0%), p = 0. 015 by Fisher' s Exact. The effect of inhibitory receptor blockade is shown in representative flow cytometry plots for intrahepatic CD8 (Figure 3B); notably, the intrahepatic Flu-specific cytokine response was readily detectable without blockade and not enhanced by PD-1/CTLA-4 blockade (Figure 3B, right panels). The effect of combined PD-1/CTLA-4 blockade on HCV-specific CD8 T cell function was more directly examined in HLA A2+ patients using HLA-A2/peptide tetramers. As shown in Figure 4A, αPD-L1 alone did not enhance intrahepatic HCV NS3 1073-specific CD8 T cell expansion compared to isotype in patient C57, and a 2-fold enrichment was observed with αCTLA-4. However, combined PD-1/CTLA-4 blockade induced a dramatic increase in the HCV NS3 1073 tetramer+ CD8 T cell frequency compared to isotype control or single blockades with aPD-L1 or aCTLA-4. In peripheral blood, αPD-L1 enhanced HCV-specific CD8 T cell expansion as expected while αCTLA-4 provided only an additive effect (Figure 4B). As for antigen-specific effector function, the frequency of tetramer+ CD8 T cells with HCV-specific IFN-γ production, CD107a mobilization and perforin expression increased with αCTLA-4 and combined αPD-L1/αCTLA-4 but not with αPD-L1 alone; in peripheral blood, the functional enhancement also occurred with αPD-L1 alone. Overall, the combined PD-1/CTLA-4 blockade enhanced the expansion and effector function of liver-derived HCV tetramer+ CD8 T cells compared to isotype control in 2/3 patients (Figure 4C). By contrast, intrahepatic CMV-specific CD8 T cells displayed little to no PD-1 or CTLA-4 expression ex vivo and were highly functional without further enhancement by PD-1/CTLA-4 blockade (Figure 4D). Collectively, these results show that combined PD-1/CTLA-4 blockade can restore proliferative capacity and effector function to deeply exhausted intrahepatic HCV-specific CD8 T cells in a synergistic manner. The effect of PD-1/CTLA-4 blockade was examined in 2 HLA-A2+ patients (A29, A35) with acute hepatitis C characterized by markedly elevated serum alanine aminotransferase (sALT) activities and viral titers. As shown (Figure 5A/B), circulating HCV-specific CD8 T cells displayed increased PD-1 (A29: 95%; A35: 92%) and CTLA-4 (A29: 28%; A35: 14%) expression. As we previously reported[13], HCV-specific CD8 T cells expanded poorly in vitro when stimulated in the presence of αPD-L1 alone. With αCTLA-4 alone, small increases in HCV tetramer+ CD8 T cell frequencies were observed in both patients, with proliferation directly measured by CFSE dilution in A35 (Figure 5B). However, a marked enhancement of HCV-specific CD8 T cell expansion occurred with combined PD-1/CTLA-4 blockade, mirroring the scenario with intrahepatic T cells. Notably, HCV-specific CD8 T cell dysfunction during acute infection did not persist after spontaneous (A29) or treatment-induced (A35) viral clearance (data not shown), suggesting that PD-1 and CTLA-4 inhibitory pathways can downregulate immune function upon active antigenic encounter, but without necessarily defining the ultimate virological outcome. The HCV-specific cytokine response to combined PD-1/CTLA-4 blockade was tightly correlated with CTLA-4 expression in CD8 T cells directly ex vivo (R = 0. 83, p = 0. 0026) (Figure 6A); a similar positive trend was noted with PD-1 expression, although this did not reach a statistical significance (R = 0. 36, p = 0. 20). Notably, the HCV-specific cytokine response during combined PD-1/CTLA-4 blockade did not correlate with the Foxp3+ Treg frequency directly ex vivo. Moreover, the functional restoration by PD-1/CTLA-4 blockade persisted after CD4-depletion that resulted in complete loss of CD4+FoxP3+ Tregs (Figure 6C); these data indicate that the functional response to inhibitory receptor blockade is independent of CD4+FoxP3+ Tregs. Among tetramer+ CD8 T cells, increased PD-1 expression associated with poor augmentation in antigen-specific expansion with the addition of αPD-L1 (data not shown), as previously reported[13]. However, the addition of αCTLA-4 to αPD-L1 resulted in marked expansion of tetramer+ CD8 T cells in direct correlation with CTLA-4 expression ex vivo (Figure 6B), suggesting that αCTLA-4 acts directly on effector CD8 T cells expressing CTLA-4. Since CD28 is over-expressed in PD-1+CTLA-4+ CD8 T cells (Figure 2A) and mediates positive costimulatory signaling for T cell activation [18], [19], we asked if the functional response to PD-1/CTLA-4 blockade is mediated by CD28+ T cells. To this end, the effect of PD-1/CTLA-4 blockade on the HCV-specific CD8 T cell IFN-γ response was examined in CD4-depleted lymphocyte subsets with and without CD28-depletion in 3 patients (2 chronic HCV patients with liver-derived lymphocytes, 1 acute HCV patients in peripheral lymphocytes). As shown (Figure 7A), the HCV-specific CD8 T cell IFN-γ response was markedly enhanced by combined PD-1/CTLA-4 blockade even in CD4-depleted cells. However, this response was lost with CD28-depletion in all 3 patients, suggesting that the functional response to combined blockade in CD28-dependent. The role of CD28 in the functional response to PD-1/CTLA-4 blockade was further examined in an HLA-A2+ patient with acute hepatitis C using an HLA-A2/peptide tetramer. As shown in Figure 7B, circulating HCV 1073 tetramer+ CD8 T cells in this patient showed 28% CD28 expression in addition to increased PD-1 (97%) and CTLA-4 (21%) expression ex vivo. Further subset analysis of HCV 1073 tetramer+ CD8 T cells based on PD-1/CTLA-4 expression showed that CD28 was preferentially expressed in PD-1+CTLA-4+ (50%) compared to PD-1+CTLA-4− (19. 5%) or PD-1−CTLA-4− (12. 2%) subsets (far right histogram on Figure 7B), similar to intrahepatic CD8 T cells (Figure 2A). As shown in the upper panel (Figure 7C), combined PD-1/CTLA-4 blockade markedly enhanced HCV-specific CD8 T cell proliferation in CD4-depleted lymphocytes. With CD28-depletion, this proliferative response was largely lost, even though HCV tetramer+ CD8 T cells remained detectable (Figure 7C). Taken together, our results show that both PD-1 and CTLA-4 can co-inhibit HCV-specific CD8 T cell function and that this effect can be reversed by combined PD-1/CTLA-4 blockade. We further show that this functional effect is mediated by CD28+ CD8 T cells, independently from CD4+ FoxP3+ Tregs. CTLA-4 is an immune inhibitory receptor within the CD28 family of costimulatory molecules[18], [20], [21]. Induced in activated T cells and constitutively expressed in FoxP3+ Tregs[22], [23], CTLA-4 shares its ligands B7-1 and B7-2 with CD28 but binds them with differential kinetics[19], [24], [25]. CTLA-4 inhibits T cell activation by engaging specific signaling pathways and by outcompeting the positive costimulatory receptor CD28[20], [26]. A critical immune regulatory role for CTLA-4 is evident from the massive and fatal lymphoproliferation that occurs in CTLA-4-deficient mice[27], [28]. Antibody-mediated blockade of CTLA-4 signaling can augment antigen-specific CD8 T cell responses in a CD4-independent manner, promoting anti-tumor and autoimmune effects[21], [29]. CTLA-4 also contributes to immune regulation and pathology in animal models of bacterial or parasitic infection[30], [31]. CTLA-4 may play a more variable role in viral infections. For example, in LCMV-infected mice, the disruption of CTLA-4 signaling fails to modify the course of infection or antiviral T cell responses in vivo[32], [33], unlike PD-1 blockade[4]. By contrast, increased CTLA-4 expression on HIV-specific CD4 (but not CD8) T cells is strongly associated with disease progression and reversible immune dysfunction in HIV-infected patients[34]. Little is known about the relevance of CTLA-4 in HCV infection. We show here that CTLA-4 (together with PD-1) contributes to HCV-specific T cell dysfunction in HCV-infected liver that can be dramatically reversed by combined CTLA-4 and PD-1 blockade. There are several notable findings in our study. For example, CTLA-4 was upregulated in deeply exhausted, HCV-specific PD-1high CD8 T cells at the site of viral replication (i. e. liver). The increased CTLA-4 and PD-1 expression was functionally relevant, since intrahepatic HCV-specific CD8 T cells regained their function with combined PD-1/CTLA-4 blockade, although not with single-blockade of PD-1 or CTLA-4. Overall, combined PD-1/CTLA-4 blockade strongly enhanced the HCV-specific CD8 T cell response in vitro in 6/9 patients from the liver (4/6 HLA-A2−; 2/3 HLA-A2+), compared to the peripheral blood response which was enhanced in none of the 11 chronically infected patients (p = 0. 0016). The combined blockade enhanced antigen-specific T cell cytokine production (e. g. IFN-γ and TNF-α) and cytolytic potential (e. g. perforin expression and CD107a degranulation) as well as their expansion. Thus, PD-1/CTLA-4 blockade promoted a polyfunctional HCV-specific T cell response, perhaps acting beyond the inhibition of cellular apoptosis which is increased in highly PD-1+ CD8 T cells in the liver or during acute hepatitis C[35], [36]. Interestingly, while combined blockade enhanced both HCV-specific T cell IFN-γ and TNF-α production (Figure 3), the increase was particularly evident for TNF-α, suggesting that PD-1/CTLA-4 blockade may promote a cytokine profile that differs from a preferential (albeit weak) IFN-γ rather than TNF-α production by dysfunctional HCV-specific CD8 T cells in chronic hepatitis C[11]. The combined PD-1/CTLA-4 blockade also enhanced intrahepatic HCV-specific CD4 T cell function, an important consideration given the relevance of CD4 T cells in immune regulation[37], [38]. Although HCV persistence is associated with increased levels of circulating CD4+FoxP3+ Tregs that constitutively express CTLA-4[17], [39], [40], [41], the level of functional restoration in CD8 T cells by PD-1/CTLA-4 blockade correlated directly with the frequency of CTLA-4+ CD8 T cells but not FoxP3+CD4+ Tregs ex vivo. Furthermore, while the functional response to combined blockade differed between the liver and blood, FoxP3+ Treg frequencies did not differ between the two compartments ex vivo or following in vitro culture with PD-1/CTLA-4 blockade (data not shown). Finally, the response to combined blockade persisted after CD4 T cells (including CD4+FoxP3+ Tregs) were depleted. These results suggest that PD-1/CTLA-4 blockade targets effector T cells directly in a manner independent of CD4 T cells including FoxP3+ Tregs[42]. PD-1 and CTLA-4 inhibit T cell activation through distinct mechanisms that converge on Akt: PD-1 inhibits CD28-mediated activation of phosphatidylinositol 3-kinase (PI3K) and CTLA-4 activates the type II serine/threonine phosphatase PP2A, both leading to the inhibition of Akt phosphorylation[43]. By competing with CD28 for B7-1 and B7-2[24], [44], CTLA-4 can reduce CD28-mediated PI3K activation, further enhancing the negative signaling through PD-1. Since PD-L1 also interacts with B7-1[45], both αPD-L1 and αCTLA-4 can increase the accessibility of B7-1 to CD28. In our study, HCV-specific CD8 T cells co-expressing PD-1 and CTLA-4 (e. g. from HCV-infected liver) were deeply exhausted and resistant to PD-1 blockade alone, whereas combined PD-1/CTLA-4 blockade had a synergistic effect in restoring their function. However, the two negative regulators may act redundantly to inhibit T cell function in this setting, such that both must be blocked to regain normal functions. The PD-1+CTLA-4+ phenotype with a functional response to PD-1/CTLA-4 blockade also occurred in circulating HCV-specific CD8 T cells during acute hepatitis C. By contrast, circulating HCV-specific CD8 T cells from chronic HCV patients (e. g. with intermediate PD-1 expression without CTLA-4 co-expression) were functionally augmented by PD-1 blockade alone. These findings suggest that the CTLA-4 and PD-1 pathways are induced early in HCV infection to co-regulate HCV-specific CD8 T cell function in a redundant manner that differs between tissue compartments over the course of infection. Notably, CD28 was highly expressed in intrahepatic PD-1+CTLA-4+ CD8 T cells compared to CTLA-4− CD8 T cells. CD28 expression may be induced to counter the inhibitory signals mediated by PD-1 and CTLA-4. Conversely, CTLA-4 may be induced in CD28+PD-1+CD8 T cells to downregulate the active inflammation at the site of viral replication. In either case, increased CD28 expression may enable greater functional enhancement upon PD-1/CTLA-4 blockade due to unhampered positive signaling through CD28. Indeed, the functional response to PD-1/CTLA-4 blockade was abolished in HCV-specific CD8 T cells by CD28-depletion in our study. Along these lines, direct CD28 costimulation enhanced HCV-specific CD8 T cell IFN-γ response in HCV-monoinfected patients but not in HIV/HCV-coinfected patients with reduced CD28 expression on CD8 T cells in one study[46]. Since loss of CD28 expression is a marker of T cell senescence and functionality[47], detection of CD28 expression in PD-1+CTLA-4+ CD8 T cells provides an additional marker for reversible functional exhaustion. Collectively, these costimulatory receptors may define a dynamic and complex functional hierarchy for antigen-specific CD8 T cells at various stages and types of viral infections that may respond to distinct therapeutic modulation. If this is correct, combined blockade could have potential therapeutic implications in chronic viral infection, provided it does not trigger autoimmunity. There are distinct differences between our study and those in human HIV or murine LCMV infections[4], [34]. In peripheral blood of HIV-infected patients, increased CTLA-4 expression with a functional response to CTLA-4 blockade was limited to HIV-specific CD4 but not CD8 T cells. This difference may be explained by the reduced CTLA-4 and CD28 expression in HIV-specific CD8 T cells[34], since the functional response to combined PD-1/CTLA-4 blockade depended on both CTLA-4 and CD28 expression in our study. Alternatively, HIV-specific CD8 T cells might exhibit compartmental differences (e. g. between blood and tissue compartments) similar to HCV. In LCMV-infected mice, LCMV-specific CD8 T cells displayed increased CTLA-4 and PD-1 expression, but responded only to PD-1 but not CTLA-4 blockade in vivo. Furthermore, HIV-specific and LCMV-specific T cells were functionally augmented by αPD-L1, without a synergistic response to combined αPD-L1/αCTLA-4[4], [34]. Thus, PD-1 may play a more universal role in antiviral T cell exhaustion whereas the effect of CTLA-4 may differ between viral infections, T cell subsets and even anatomical locations. In conclusion, both CTLA-4 and PD-1 contributes to HCV-specific T cell exhaustion in a redundant manner in human HCV infection, particularly in HCV-infected liver; this intrahepatic virus-specific T cell dysfunction can be synergistically reversed by combined PD-1/CTLA-4 blockade in vitro in a CD4-independent and CD28-dependent manner. These findings provide new insights to the mechanisms that regulate virus-specific T cell dysfunction and suggest that immune exhaustion at the site of antigen expression may be reversed by combined inhibitory receptor blockade. All subjects were recruited with informed consent approved by the Institutional Review Boards. All investigations have been conducted according to the principles expressed in the Declaration of Helsinki. Patients were recruited at the Philadelphia Veterans Affairs Medical Center (PVAMC) and the Hospital of the University of Pennsylvania. A total of 47 patients with chronic hepatitis C without HIV coinfection (group C) were examined, including 33 cirrhotic patients undergoing liver transplantation and 14 patients with chronic stable HCV infection. Control groups included 10 healthy HCV-seronegative subjects (group N), 4 HCV-seropositive but RNA-negative patients with spontaneous resolution of HCV infection without prior antiviral therapy (group R) and 6 patients with acute hepatitis C (group A) diagnosed by acute serum alanine amino-transferase (sALT) elevation with documented HCV-seroconversion and/or viremic fluctuations greater than 10-fold without prior liver disease as described previously[48]. The patient characteristics are shown in Table 1. All fluorescent monoclonal antibodies (mAbs) were purchased from BD Bioscience (San Jose, CA) except for: (i) αFoxP3 and αCD28 from eBioscience (San Diego, CA); (ii) FITC-labeled αPD-1 (αCD279; clone EH12. 2H7) from BioLegend (San Diego, CA); and, (iii) PE-labeled αPD-1 from the Dana Farber Cancer Institute (Boston, MA). Of note, PD-1 and CTLA-4 expression in all subjects was examined using FITC-labeled αPD-1 (clone M1H4, BD) and PE-labeled αCTLA-4 (αCD152; clone BNI3, BD). In selected subjects, the patterns of PD-1 (low/intermediate/high) and CTLA-4 expression in CD8 T cells were compared using FITC-labeled αPD-1 from BioLegend or PE-labeled αPD-1 from the Dana Farber Caner Institute combined with APC-labeled αCTLA-4 (BD). Dead cells were excluded with 7-AAD. For functional blockade, αPD-L1 mAb (clone 29E. 2A3. C6) from the Dana Farber Cancer Institute[13], [49] and αCTLA-4 mAb (clone BNI3; BD) [34] were used. The HCV-specific T cell response was measured using a pool of 105 overlapping 15mer peptides spanning the entire NS3 protein derived from HCV genotype 1a[13], [39], [48], [50]. Similarly, the T cell response to influenza virus was examined using 49 overlapping 15mer peptides spanning the conserved matrix M1 protein (residues 1–252) based on the human A/PR/8/34 (H1N1) virus[51]. For HLA-A2+ subjects, the following peptides corresponding to optimal CD8 epitopes were synthesized for antigenic stimulation and tetramer synthesis as described previously [13]: (i) HCV NS3 1073 (CINGVCWTV), NS3 1406 (KLVALGINAV) and NS5B 2594 (ALYDVVSKL); (ii) influenza matrix (GILGFVFTL); (iii) EBV BMLF1 (GLCTLVAML); and, (iv) CMV pp65 (NLVPMVATV). Cells were stained with fluorescent antibodies according to the manufacturer' s instructions; events were acquired with a FACSCalibur or FACSCanto (Becton Dickinson, San Jose, CA) and analyzed with FlowJo software (Tree Star Inc. , San Carlos, CA). Compensations were established using single color controls. As CTLA-4 is more readily detected in the cytoplasm due to rapid internalization[19], [34], [52], CTLA-4 expression was assessed by intracellular staining following permeabilization[34]. Cutoffs for CTLA-4 expression was defined by isotype control where 99. 9% of the events were negative. Figure S2 further confirms the preferential CTLA-4 expression on PD-1high CD8 T cells with corresponding isotype and unstained controls. PBL were isolated by standard Ficoll-Histopaque (Sigma Chemical Co. , St Louis, MO) density centrifugation[39], [53]. LIL were isolated from 20–50 gm of fresh liver explant tissue that was transported in complete media and processed within 24 hours of explant (usually 1–3 hours) as described previously[13]; briefly, this procedure incorporated careful dicing of liver into 5 mm3 pieces, incubation of the liver slurry at 37°C with 1 mg/ml collagenase (Type 1a; Roche Molecular) and 1 µg/ml DNase (Sigma Aldrich) for 30 minutes, further mechanical dissociation using the Seward Stomacher 400 Lab Blender (Brinkman Instruments, Westbury, NY), filtration through a 70 µm nylon filter and Ficoll-Histopaque density centrifugation. Control experiments showed that collagenase digestion for 30 minutes did not alter PD-1, CTLA-4, or CD28 expression (data not shown). PBL and LIL (2×106 cells/ml/well) were stimulated on day 0 with overlapping HCV NS3 or influenza matrix 15mer peptides (2 µM) in complete media in the presence of isotype control antibodies, αPD-L1[13], αCTLA-4[34] or both αPD-L1 and αCTLA-4 (10 µg/ml for each mAb). Cell cultures were stimulated with rIL-2 (100 IU/ml) on day 4 and examined by flow cytometry on day 7 with CD107a, intracellular cytokine or perforin staining as previously described[13], [54]. For intracellular cytokine staining, expanded PBL and LIL cultures were stimulated for 6 hours with HCV or Flu peptides in the presence of brefeldin A (10 µg/ml) before surface staining, permeabilization and intracellular staining with αIFN-γ and αTNF-α. Antigen-specific CD107a mobilization was quantified by adding FITC-labeled αCD107a before peptide stimulation. In selected HLA-A2+ subjects with available cells, PBL and LIL were stimulated with HLA-A2 restricted antigenic peptides (10 µg/ml) with the blocking conditions described above. In select experiments, antigen-specific IFN-γ+ T cell response was quantified by IFN-γ ELISPOT assay in which cultured lymphocytes were stimulated for 45 additional hours with antigenic peptides or control media (200,000 cells/well in triplicates) as previously described[39], [50], [53]. HCV-specific IFN-γ+ T cell frequency was calculated by subtracting the mean IFN-γ spot forming units (SFU) in control wells from the mean SFU in antigen-stimulated wells and expressed as IFN-γ SFU/106 cells. Lymphocytes were labeled with 5 mM CFSE (Molecular Probes, Eugene, OR) as described previously[13], [17] before 7 days of culture with antigenic peptides (10 µg/ml) in the presence of isotype control or blocking antibodies as described above. Cell cultures were stimulated with rIL-2 (100 IU/ml) on day 4 and examined by flow cytometry on day 7 for antigen-specific T cell expansion. CD4 T cells were depleted from PBL and/or LIL using CD4 Dynabeads (Invitrogen, Oslo, Norway) as previously described[39]. CD28+ T cells were depleted by sequentially staining with αCD28-PE (clone CD28. 2, BD Pharmingen) and anti-PE Microbeads before separation by AutoMACS (Miltenyi Biotec Inc) as previously described [17]. The efficiency of CD4 and CD28 depletion was >97% (data not shown). Clinical and immunological parameters were compared using the Mann-Whitney U-test, the paired t-test and the Kruskal-Wallis test. Frequency differences were compared by Fisher' s Exact test or the Chi-square test as appropriate. Correlations were tested for significance by the Spearman rank correlation test. P values below 0. 05 were considered significant.

Hepatitis C virus (HCV) is an important human pathogen with a high rate of persistence associated with chronic liver disease that can progress to cirrhosis and hepatocellular carcinoma. Chronic HCV infection occurs in the setting of impaired antiviral T cells that over-express an inhibitory receptor PD-1 (programmed death-1 receptor). Recent studies showed that in vitro inhibition of the PD-1 pathway via an inhibitory antibody can reverse the functional impairment in HCV-specific CD8 T cells from blood but not the liver (the site of viral infection and disease progression). In this study, we show that a second co-inhibitory receptor, CTLA-4, is upregulated in HCV-specific CD8 T cells from the liver and that combined PD-1/CTLA-4 blockade (but not single blockade of PD-1 or CTLA-4) can synergistically enhance their function. This functional enhancement was CD28-dependent but CD4-independent. This effect also differed between viruses, tissue compartments (liver vs. periphery) and clinical status (acute vs. chronic). We conclude that PD-1, CTLA-4, and CD28 expression profiles define a novel hierarchy in HCV-specific CD8 T cell exhaustion than can be synergistically reversed by combined inhibitory receptor blockade. These findings have potential immunotherapeutic applications, provided that no autoimmunity is induced.

Abstract Introduction Results Discussion Materials and Methods

virology/persistence and latency gastroenterology and hepatology/hepatology virology/new therapies, including antivirals and immunotherapy virology/immune evasion immunology/immunity to infections

2009

Synergistic Reversal of Intrahepatic HCV-Specific CD8 T Cell Exhaustion by Combined PD-1/CTLA-4 Blockade

Bracoviruses are symbiotic viruses associated with tens of thousands of species of parasitic wasps that develop within the body of lepidopteran hosts and that collectively parasitize caterpillars of virtually every lepidopteran species. Viral particles are produced in the wasp ovaries and injected into host larvae with the wasp eggs. Once in the host body, the viral DNA circles enclosed in the particles integrate into lepidopteran host cell DNA. Here we show that bracovirus DNA sequences have been inserted repeatedly into lepidopteran genomes, indicating this viral DNA can also enter germline cells. The original mode of Horizontal Gene Transfer (HGT) unveiled here is based on the integrative properties of an endogenous virus that has evolved as a gene transfer agent within parasitic wasp genomes for ≈100 million years. Among the bracovirus genes thus transferred, a phylogenetic analysis indicated that those encoding C-type-lectins most likely originated from the wasp gene set, showing that a bracovirus-mediated gene flux exists between the 2 insect orders Hymenoptera and Lepidoptera. Furthermore, the acquisition of bracovirus sequences that can be expressed by Lepidoptera has resulted in the domestication of several genes that could result in adaptive advantages for the host. Indeed, functional analyses suggest that two of the acquired genes could have a protective role against a common pathogen in the field, baculovirus. From these results, we hypothesize that bracovirus-mediated HGT has played an important role in the evolutionary arms race between Lepidoptera and their pathogens. Unlike bacteria, which have obtained a notable proportion of their genes through the acquisition of sequences from distantly related organisms, eukaryotes are generally thought to evolve mainly through the modification of existing genetic information [1]. However evidence of horizontal gene transfer (HGT) in eukaryotes is accumulating and is recognized as an important factor in their evolution and acquisition of novel traits [2–5]. The majority of events reported concerns transposable elements, DNA sequences capable of excising or copying themselves from one genomic locus to integrate into another locus [6]. Genome sequencing has revealed that eukaryotes have also acquired DNA from symbionts and parasites, probably because the intimacy of these relationships favours DNA exchange. For example, numerous insect and nematode genomes contain sequences originating from Wolbachia [7,8] an endocellular bacteria widespread in insect populations infecting, in particular, host germ line cells [9]. Recently, a systematic investigation of HGT events in three available lepidopteran genomes (Bombyx mori, Danaus plexippus and Heliconius melpomene) revealed multiple ancient HGT events from bacteria and fungi to these lepidopteran genomes [10]. Here we present an original mode of HGT between two insect orders based on the integrative properties of a virus (bracovirus) that has evolved within parasitic wasp genomes for ≈100 million years and that is used to facilitate the development of their progeny in caterpillars by inhibiting host immune defenses. In one case, we could demonstrate the direction of the transfer based on the presence of a sequence important for the virus life cycle. This is a rare example where the likely mechanism of HGT can be established in an animal system. Moreover we present functional analyses suggesting that some of the transferred genes have been recycled by Lepidoptera to protect them against a common viral pathogen. Bracoviruses play a central role in parasite-host interactions involving parasitic wasps and their caterpillar hosts. Bracoviruses are injected by parasitic wasps into their hosts along with wasp eggs. These wasps develop during their larval stage within the body of their lepidopteran hosts. Tens of thousands of species of wasps belonging to the braconid family and parasitizing a large diversity of lepidopteran species are each associated with a specific bracovirus [11]. All these associations originated from a single integration event of a nudivirus genome in a common ancestor of the wasps [12]. Since this integration ≈100 MYA, the genes involved in virus particle production have been dispersed in the wasp genome, they are no longer packaged in the particles that contain genes encoding virulence factors. Moreover the endogenous chromosomally transmitted virus has evolved depending on its contribution to parasitism success, resulting in a specific set of virulence genes packaged in the particles in the different wasp lineages [13]. These viruses are now essential for successful development of the wasp larvae within lepidopteran hosts [13–15]. Viral replication and particle production occur exclusively in the wasp ovaries from endogenous viral elements present in the wasp genome. The particles, that contain dsDNA circles harbouring the virulence genes, constitute the major component of the fluid injected with the eggs into the parasitized caterpillar host during wasp oviposition. Once in the host body the particles enter lepidopteran host cells and the host cellular machinery expresses these virulence genes. Viral products ensure wasp larvae survival in the lepidopteran body by interfering with caterpillar host immune responses and development [16,17]. The dsDNA circles packaged in the particles are produced from chromosomally transmitted proviral segments stably integrated in the wasp genome [18–21]. The typical eukaryotic organization of the genes transferred by the particles [22] and their lack of similarity with viral genes suggest they originate from the wasp genome, which could be demonstrated for a few of them by phylogenetic analyses [23]. However many genes have diverged in their sequence from insect genes, to the extent that they are currently no more closely related to wasp genes than to mammalian genes [24,25]. Many other bracovirus genes have unknown origins and display no similarities to genes in data banks except with other bracovirus sequences. For example, Cotesia congregata bracovirus (CcBV) encodes 26 bracovirus specific gene families (named BV1 to BV26) [18]. We previously reported that some viral circles were found to be reintegrated in the genome of different geographic strains of the wasp Cotesia sesamiae [13,26]. The occurrence of circle integrations back into wasp genomes probably reflects a broad integration ability of circles since it was recently shown that integration into the DNA of parasitized lepidopteran host cells is a part of the bracovirus life cycle. Indeed it was shown that Microplitis demolitor bracovirus circle integration into lepidopteran Pseudoplusia includens DNA occurs by a specific mechanism involving a conserved viral site named Host Integration Motif [27]. During integration the circles are opened specifically at this site, resulting in integrated forms readily distinguishable from that of the proviral form [27]. The analysis of Cotesia sesamiae bracovirus (CsBV) reintegrated circles suggests that the same mechanism was involved in their integration back into the wasp genome [13,26]. Parasitized caterpillars represent most of the time an evolutionary dead-end since parasitoid wasps inhibit metamorphosis [28] and the host usually does not survive parasitism [29]. However it is conceivable that some hosts might successfully defend themselves against the parasite by interrupting wasp oviposition, eliminating the eggs or killing the larvae, resulting in the reproduction of Lepidoptera that have been infected by bracoviruses. Parasitoid wasps could also target non-host species and fail to interfere with their development [30]. We can speculate that caterpillar escape from the fatal issue after virus injection could allow bracovirus particle entry into germ cells and in rare cases stable integration of circles into lepidopteran genomes. To determine if bracovirus sequences could indeed be integrated into lepidopteran genomes, we compared the DNA sequences packaged in the particles of Cotesia congregata bracovirus (CcBV), the genome of which is almost completely characterized [18], to a series of genomes from non-host Lepidoptera of the parasitoid wasp C. congregata and from Manduca sexta, a regular host. Here, in contrast to a previous bioinformatic analysis listing a series of bracovirus insertions, most of them relatively short [31], we searched for large nucleotide stretches (more than 500 bp long) that could encode potentially domesticated genes by lepidopteran species and evaluated the evolutionary meaning of these integrations by functional analysis of two of the transferred genes. Similarity searches allowed the identification of bracovirus DNA insertions in the genomes of the monarch (Danaus plexippus), the silkworm (Bombyx mori), the beet armyworm (Spodoptera exigua) and the fall armyworm (Spodoptera frugiperda) but not in the genome of tobacco hornworm (M. sexta), the regular host of Cotesia congregata. All these insertions were characterized by the presence of large stretches of nucleotide sequences strikingly similar to those of bracoviruses (close to 90% identities at the nucleotide level) flanked by lepidopteran-specific sequences. Insertions include genes but also in some cases parts of bracovirus circles, the organization of which has been conserved, indicating the direction of HGT was from bracovirus to Lepidoptera. Moreover, in one insertion a regulatory signal involved in dsDNA circle production in the wasp has been retained, constituting an unambiguous signature of the bracoviral origin of the sequence since bracovirus replication is non autonomous and occurs exclusively in the wasp ovaries. Altogether our data indicate that bracoviruses have been a recurrent source of genes for Lepidoptera. Moreover functional analyses provide convergent results suggesting that some of these genes might contribute to protect larvae against baculoviruses, deadly pathogens that threaten them in the field, which would explain why they have been maintained in lepidopteran genomes. Unexpected levels of similarities were observed between sequences of several lepidopteran genomes and bracoviruses. The level of similarity is in the range of that found for homologous genes coding for highly conserved proteins such as histone H4, almost invariant from plants to animals. However this similarity is unlikely to be due to conservative selection since the encoded genes are conserved only in a limited number of phylogenetically closely related lepidopteran species. In this study we report the presence of these bracovirus-related sequences in several lepidopteran genomes and discuss the possible mechanisms involved in their acquisition. Compared to a previous report describing bracovirus DNA insertions in the monarch and silkworm genomes [31] we provide here an in depth analysis of the structure of the bracoviral and lepidopteran flanking sequences. We show that monarch insertions are fixed in the species, that their presence in the lineage is ancient and that they have undergone rearrangements since their integration. By measuring selection pressures using genomes from individuals of 80 monarch and 8 related species we show that the selection acting on these genes is mainly conservative, which suggests the domesticated Ben genes could play a role in monarch physiology. In addition we report for the first time HGT and domestication of bracovirus sequences in Lepidoptera of the Spodoptera lineage. Moreover we present functional analysis on 2 unrelated genes suggesting the transferred genes could protect the Lepidoptera against a viral pathogen. High similarities observed could be due in theory either to DNA sequence transfer from bracovirus to lepidopteran genomes or vice versa. Wasp larvae containing a bracovirus as an endogenous virus have an intimate relationship with Lepidoptera since they develop within the body of their hosts, for this reason it is possible that acquisition of lepidopteran genes by bracoviruses can occur. Accordingly horizontal transfer of a Mariner like transposable element (MLE) shared by a parasitoid wasp and its host was previously reported. In this case, the direction of the transfer was supposed to be from Lepidoptera to Hymenoptera based on the presence of this transposon in closely related species of the lepidopteran host and its absence in a closely related parasitoid species [46]. Another horizontal transfer of a transposable element (Helitron) was reported between Copidosoma floridanum an endoparasitoid wasp (not associated with a bracovirus) and the Lepidoptera Trichoplusia ni suggesting that parasitism might favor horizontal transfer of TEs [47] but the direction of the transfer was not determined in that study. Similarly, Thomas et al. , (2010) also found evidence of horizontal transfer of Helitrons in bracoviruses and Bombyx mori [48]. One of the insertions described here is particularly informative regarding the direction of the transfer because it contains a regulatory sequence typical of bracoviruses (see Fig 5). The sequences named Direct Repeat Junction (DRJ) that terminate all bracovirus proviral segments are conserved among BVs [18]. These direct repeats are involved in dsDNA circle production [49]. During viral replication, large molecules are amplified that serve as precursors for the production of individual circles, produced by a recombination between the DRJs [19]. As a result, a single DRJ (resulting from the recombination) is present on a circle. This recombination process was confirmed recently by inactivation of two Tyrosine recombinase genes (vlf1 and int-1) using RNA interference, which resulted in impairment of circle formation [50]. The presence of a DRJ in the BV2-5 insertion in the S. exigua genome constitutes an unambiguous signature of its bracoviral origin since these regulatory elements are specific of the bracovirus life cycle. This clearly demonstrates that the BV2-5 sequence originated from the bracovirus and was acquired by the lepidopteran genome. The direction of the other horizontal transfers, although not as clearly proven, also appears to be more likely from bracovirus to Lepidoptera genomes, because only a limited number of closely related lepidopteran species harbour these sequences. Moreover bracovirus life cycle features suggest they are involved in horizontal transfer. Indeed bracovirus circles have been shown to enter cells of all tissues tested [51,52] and to integrate into the DNA of lepidopteran host cells as a part of the wasp life cycle [27]. Several components of the virus particles belong to the integrase family (VLF1, INT1, INT2) [53] and thus potentially mediate integration. During parasitism, bracoviruses do not replicate in host tissues and therefore integration into host cell DNA may allow persistence of bracovirus DNA in lepidopteran larvae that continue to develop [27]. It was previously shown that a side effect of this integration mechanism was to allow circle integration events back into germline cells of the wasp [26]. This was indicated by the analysis of bracovirus sequences in Cotesia sesamiae genome. Strikingly, segments homologous to CcBV circle 10 were found in two different genomic locations in C. sesamiae strains of Kenya [26]. Sequence comparison of circular and reintegrated viral forms [13,26] indicated that circle integration likely involved the same mechanism as the one described for the integration of bracovirus circles into lepidopteran host genomic DNA during parasitism, using specific sites on the circle (the Host Integration Motifs) [27]. The occurrence of circle integration into lepidopteran host germline DNA resulting in sequence transfer between bracovirus and Lepidoptera is likely another consequence of this viral integration mechanism. Although we did not find integration of complete circles such as those described in the wasp genome, the largest Ben9 encoding region in Danaus plexippus corresponds to more than half of C25 sequence and has retained in this Lepidoptera the bracovirus organisation with two genes (RnaseT2 and Ben9) separated by non-coding sequences [18]. We hypothesize that bracovirus insertions correspond to remnants of circles integrated in Lepidoptera genomes that have been subject to many rearrangements since their integration. Indeed it is likely that after circle integration bracovirus sequences are lost, unless they provide a selective advantage to the insect. Therefore, identification of complete circles in genomes, corresponding to recently integrated sequences, not fixed in the species, might require more diverse template sources than the very limited number of individuals used for lepidopteran genome sequencing. The insertions described in this paper are most probably all ancient. For example, Ben9 was already present in the common ancestor of the Danaina subtribe 5 MYA [35]. Moreover evidence that rearrangements have occurred is provided by the comparison of the two Ben9 gene containing regions, one having conserved a larger part of the bracovirus non-coding sequence than the other. The insertion in the B. mori genome has also been obviously rearranged since a stretch of lepidopteran specific DNA separates bracovirus sequences in two parts. BV2-5, Se-BLL2 and SF2. 5 insertions in Spodoptera spp correspond mostly to single genes, which could represent an ultimate stage of domestication, most of the sequence of the circle having been lost. It is also possible that a broader mechanism than virus-mediated integration, such as DNA repair, which is involved for example in transgenic mice production [54], might have resulted in the insertion of fragments of bracovirus circles in Lepidoptera. However it should be noted that in all cases described in this study the insertions correspond exclusively to sequences from bracovirus circles: we did not find any stretch of wasp sequence flanking or separating bracovirus DNA sequences in the lepidopteran genomes. Thus although integration of wasp DNA could be possible in theory, given that the wasp larvae develop within lepidopteran hosts, we did not search for, nor find evidence of wasp DNA (non-viral) integration in this study. The presence of bracovirus sequences in lepidopteran genomes is apparently a paradox given that infected larvae are considered as an evolutionary dead-end (see Fig 10). For example, CcBV has been shown to induce alteration of host developmental programming resulting in inhibition of metamorphosis, even when experimentally injected in a lower amount than during wasp oviposition [28]. Accordingly, we found no evidence for HGT of CcBV genes in M. sexta, a common host of C. congregata but instead genes having similarities with other polyDNAviruses [33]. Some host species might be less susceptible to the effect of bracoviruses on development or could have developed resistance mechanisms, and therefore “live to tell the tale” after parasitism and injection of particles (Fig 10). However we propose that the main route of bracovirus gene acquisitions by Lepidoptera could be through parasitoid wasp stinging of non-host species (Fig 10). In the field, the host range of the wasp C. congregata corresponds to several species of sphingidae, but in laboratory conditions it was shown to oviposit in non-host species such as the noctuidae Trichoplusia ni [30]. Such behaviour might offer the opportunity for bracovirus DNA to integrate into genomes of lepidopteran lineages that do not belong to the host range of bracovirus-associated wasps (such as species of the monarch lineage for example) and to “escape” bracovirus induced host development arrest. In this context the cellular machinery of Lepidoptera appear to be sufficiently conserved to express a bracovirus gene normally adapted for expression in a different lepidopteran family. Indeed, the conservation of Ben9 and BV2-5 intron splicing in two different lepidopteran families (sphingid/nymphalids and sphingid/noctuids respectively, Fig 1, Fig 2), illustrates that these genes can be “ready to be expressed” even in non-target species. Although not recorded in the field to our knowledge, oviposition in non-hosts may happen since parasitoids, such as Cotesia species, attacking aggressive caterpillars do not have the time to intensively examine the potential host before oviposition. In rearing conditions, they sometimes lay their eggs into other adult wasps, which shows the lack of specificity in their choice [55]. Conservation of bracovirus genes in lepidopteran genomes is likely associated with an increase in insect fitness due to the expression of the viral genes. This hypothesis is sustained by functional studies with the SeBLL2 and BV2-5 proteins from S. exigua showing they have an impact on baculovirus infection. These results suggest that host domestication of these bracoviral genes might increase insect protection to this natural pathogen playing a role in regulating population dynamics in the field [56,57]. We have found that the interference of recombinant BV2-5 with the cellular cytoskeleton dynamics has a strong impact on the baculovirus producing this protein suggesting BV2-5 could confer larval protection against baculovirus infection. This hypothesis is corroborated by the fact that an S. exigua BV2-5 bearing strain is less susceptible to baculovirus infection than the European population bearing the BV2-5 truncated form (S6 Fig). However the genetic background between the two strains is probably different and other approaches such as the use of CRISPR/Cas9 technology to produce S. exigua lines by knocking out of BV2-5 or restoring the functional BV2-5 will be required to unambiguously demonstrate the protective function of this protein, after baculovirus infection. The alteration of a fundamental cellular component such as cytoskeleton dynamics probably also induces a cost. S. exigua is a Palearctic species, which was introduced in America in 1876 probably from Europe [58]. The fact that Lepidoptera now collected in Europe encode a truncated form of BV2-5 suggests that a recent mutation has spread in this population. It is tempting to speculate that BV2-5-mediated baculovirus protection might induce a cost leading for example to increased susceptibility to other pathogens such as bacteria or parasitoids. The frequency of one or the other form of BV2-5 might depend on the abundance and local selective pressure exerted by pathogens and/or parasites and the cost might also explain why BV2-5 has been lost in S. frugiperda while it was detected in S. litura a more recent species in the Spodoptera lineage. In any case BV2-5 coding sequence is more conserved than the other part of the bracovirus insertion suggesting the gene as Ben genes in the monarch is generally under conservative selection and not neutrally transmitted. In addition to BV2-5, we have observed that another gene of bracovirus origin Se-BLL2 can also confer certain level of protection in experimental conditions against both viral forms of baculovirus, occluded derived virions (responsible of the primary infection) and budded viruses (responsible of the systemic infection of larvae). C-type lectins are carbohydrate-binding proteins playing a range of functions in multiple organisms [59]. In general, PDV lectins are able to specifically recognize carbohydrates on the surface of the endoparasitoid eggs and, thus, inhibit the recognition of the eggs by the lepidopteran host recognition proteins [60]. Although little is known about the response of C-type lectins to viral invaders, crustacean lectins have been reported to be related to the antiviral defense [61,62]. In lepidopterans, the only example of antiviral response involving C-type lectins was reported by Chai et al. [63]. Our experiments have shown that BLL2 action is interfering with the initial viral entrance into the Sf21 cells (Fig 8B and 8C). According to these results, it is likely that antiviral action of Se-BLL2 is due to its interaction with viral or host cell membrane glycoproteins involved in viral binding and entrance. Nevertheless, additional studies will be needed to define the exact mode of action of BLL proteins as well as their possible role in the host interaction with viral and non-viral pathogens and parasitoids. In any case it should be noted that the acquired genes do not confer a complete protection against baculovirus infection and our study confirm that S. exigua larvae are indeed susceptible to baculovirus infection (Fig 9B). According to the literature the susceptibility of Spodoptera spp depend on many factors such as the larval stage [64], the type of plant hosting the insects [65], the geographical origin of the insects, and even on the midgut microbiota composition [66]. Many individuals ingesting a sublethal dose of OBs can survive with a covert infection (larvae harbouring baculovirus but not displaying the disease symptoms) the incidence of which can be over 50% in the field for Spodoptera exigua [56]. Little is known on the molecular aspects of this phenomenon but BV2-5 effect on cytoskeleton dynamics could possibly contribute to this latency. Taken together a large number of factors can modulate insect susceptibility and given the high incidence of baculovirus infection in the field being even only less susceptible can have a great impact at the population fitness. In the context of a host-pathogen arms race any new trait that confers an advantage to any of the competitors is susceptible to be incorporated into the gene pool. Altogether our results strongly suggest that two acquired genes can confer an advantage against viral infection although the comprehensive analysis of the molecular function of the identified proteins is awaited and we cannot completely exclude at this stage that they could have other functions. Ben genes also probably have a role for the Lepidoptera since they have been maintained in different lineages and we have shown that in the monarch they are mostly under conservative selection. We have described in this report several insertions of bracovirus DNA sequences in a series of lepidopteran genomes. In mammals a few examples have been described of integrated retrovirus receptor genes conferring a specific protection against new infections by related viruses using the same cell entry mechanism [67,68]. Recently, this concept of genes acquired and domesticated by hosts to protect against related virus infections has been shown to operate also for a Bornavirus (negative strand RNA virus) [69]. Virus resistance conferred by expression of viral genes in plants has also been described. Indeed, transgenic plants expressing viral gene constructs can exhibit resistance to infection by the virus [70,71]. Here, we extend this concept of an organism using pathogen genetic resources as a protection against other pathogens, to insects. Indeed, we show that domestication of different bracovirus genes most likely confers protection to Lepidoptera against baculoviruses, a common pathogen in the field. What is very original compared to previous reported cases is the use of viral sequences as a protection against a distantly related virus. Indeed, most of the viral sequences inserted into host genomes that were hypothesized to confer a protection are effective against closely related viruses. The protection mechanisms are based on the expression of defective proteins of viral origin that are able to interact with those of the pathogenic virus and thus interfere with cell entry [72], replication [73] or interfere by producing small RNAs inducing destruction of virus transcripts having highly similar sequences [74,75]. Since baculovirus infection of the host could be lethal for the parasitoid [76], it might be speculated that the function of some of the bracoviral genes domesticated by Lepidoptera was already to protect the parasitized larvae against baculovirus infection. This might provide an explanation for both the unusual ability to interfere with distantly related virus infections and the fact that the bracovirus genes have conserved the same structure after their integration into Lepidoptera genomes. A specific bracovirus circle integration mechanism into lepidopteran host DNA operating during parasitism and resulting occasionally in circle reintegration into wasp genome has been previously characterized [27]. This mechanism is likely involved in HGTs between Hymenoptera and Lepidoptera, although it is also possible that some of the sequences might have been integrated through DNA repair. Once integrated into lepidopteran genomes, bracovirus genes are readily domesticated by Lepidoptera since they are already adapted for expression in lepidopteran tissues during parasitism. Indeed the majority of the CcBV genes expressed during parasitism were shown to possess an insect structure with an arthropod transcription start site, at least one intron and polyadenylation signals [22] and we showed here that the splicing machinery of different Lepidoptera families can produce the same mRNAs from a bracovirus gene containing introns. Altogether the ability of bracoviruses to mediate integration, the fact that bracovirus gene structure is adapted to expression in Lepidoptera and that bracovirus circles have acquired different gene sets depending on the wasp lineage suggest we are only seeing the tip of the iceberg and that numerous cases of integration and domestication of bracovirus sequences will be identified with the exponential rise of genomic data provided by new generation sequencing. Thus this phenomenon is not merely a curiosity but has most likely played an important role in the arms race between Lepidoptera and their pathogens. Sequences sharing high similarity with Bracovirus sequences in lepidopteran genomes were identified using the 35 CcBV circles [18] as queries in megablast analysis (NCBI) against whole genome shotgun contig data banks (wgs at NCBI) restricted to lepidopteran genome sequences. Unlike the bioinformatic study which recently reported numerous short insertions of bracovirus sequences in lepidopteran genomes [31] we focused here on bracovirus-like sequences more than 1 kb long and encoding at least one gene. A blast analysis between proviral integrated circle sequences and the different contigs identified was used to determine the precise location of high scoring pairs (HSP) reported in Fig 1. HSP and annotated sequences were visualised using DNAPlotter [77]. Homologous transcribed sequences were then searched for in the Transcriptome Shotgun Assembly (TSA) database at NCBI, for Spodoptera and Bombyx sequences (no data was found in TSA for Danaus plexippus). It should be noted that although all the identified insertions are very closely related to CcBV since the donor circle can be identified in many cases, this does not necessarily indicate that C. congregata was the donor species since only a handful of bracovirus packaged genomes have been sequenced to date among those associated with the estimated 1200 species of the Cotesia genus [78]. Specimen of Danaina subtribe were kindly provided by David Smith. They were collected randomly in the field in the late 1990s, killed in ethyl ethanoate vapour immediately before storage in 95% ethanol [34]. They were stored for ≈6 years at -20°C then for 9 years at room temperature (D. Smith, personal communication). Adult D. plexippus were collected in Australia (voucher number, vn: 396), D. genutia in Thailand (vn: 430), D. chrysippus chrysippus in Oman (vn: 262), T. septentrionis septentrionis in Malay Peninsula (vn: 216). More recently, D. plexippus plexippus caterpillars were sampled in August 2012 in Valcartier, Canada (46°56’52”N, 71°29’50”W). Four colonies of S. exigua, derived from different geographic locations, were continuously reared on artificial diet at 25± 3°C with 70±5% relative humidity and a photoperiod of 16h light: 8h dark. The FRA strain was supplied by M. Lopez-Ferber, (INRA St. Christol les Alés, France) [79]. The ALM strain was established from successive collections from southern Spain [80]. XEN-R strain was obtained from cotton fields in Pattville, AL. (USA) and was later selected for resistance to Bacillus thuringiensis [81,82]. SUI population was provided by Andermatt Biocontrol AG (Grossdietwil, Switzerland). The DNA from the Mexican population was provided by P. Caballero (Universidad Publica de Navarra). Finally, DNA representing S. exigua from Japan was obtained from the cell line Se301 originally derived from insects collected in Japan (Hara, et al. 1995). After grinding frozen samples in liquid nitrogen, DNAs from D. plexippus larvae were extracted by C. Béliveau using QiaAMP DNA mini kit (Qiagen) and were sent by M. Cusson (Québec, Canada). For specimens of Danaina subtribe, 20 mg of tissues were dried at 37°C to eliminate ethanol, frozen in liquid nitrogen and ground with a pellet and a mortar previously refrigerated at -80°C. DNA was then extracted using the QiAmp DNA Mini Kit (Qiagen) following the supplier’s instructions. To compensate for partial degradation of DNA from old samples, primers were designed for amplification of short fragments. A 35-cycle PCR (94°C for 60 s; 50°C for 60 s; 72°C for 60 s) was performed with 10 pmol of each primer, 0. 2 mM dNTP (MP Biochemicals), 1. 5 mM MgCl2,0. 5 unit Goldstar (Eurogentec) and 20 ng of genomic DNA. PCR products (8 μl) were run on 1. 5% agarose gels. The EF1α gene was used as a control of DNA sample quality. Successful amplifications of Ben9 insertions (Fig 2) were obtained: for D. plexippus using primers Ben9 12F and Ben9 12R (176 bp fragment, S1 Table); for D. chrysippus and D. genutia using primers Ben9 13F and Ben9 13R (173 bp fragment); for and T. septentrionis using Ben9 14F and Ben9 14R (123 bp fragment); amplifications of Ben4 insertion (S1 Fig) was obtained using Ben4 2F and Ben4 2R (223 bp fragment). It should be noted that the PCR analyses did not allow us to discriminate between the two Ben9 insertions (Fig 1 (A) and 1 (B) ). Total RNA was isolated from 5th instar S. exigua larvae using RNAzol reagent (Molecular research centre, INC) as described in the manufacturer’s protocol. One μg of each RNA was DNase treated (Invitrogen) and reverse transcribed into cDNA with oligo (dT) and hexamer primers using Super Script II Reverse Transcriptase from Invitrogen. D. plexippus RNA was prepared using TRIzol reagent (Invitrogen) and sent by M. Cusson (Québec, Canada), it was further treated by rDNase (NucleoSpin RNA, Macherey-Nagel) to eliminate residual DNA until no PCR amplification of a control gene could be detected from the sample. For D. plexippus, a total of 1μg of RNA was reverse transcribed into cDNA with oligo (dT) primers using Super Script II reverse transcriptase or Omniscript RT kit (Qiagen). PCR amplifications from cDNA of the different genes were performed using standard protocols and specific primers (S1 Table). For the BV2-5 alleles, initial sequences were obtained from the transcriptome of S. exigua larvae exposed to different types of pathogens [40]. Two primers flanking the coding sequence were designed and used to amplify this sequence from cDNAs originating from different populations and cells. RT-PCR-amplified alleles of BV2-5 were directly sequenced or cloned into pGEM-T Easy vector (Promega) and sequenced using standard primers. At least two independent clones were sequenced for each insect population. Selection pressures operating on D. plexippus Ben4 and Ben9 genes were measured using Illumina data available at NCBI (SRA data bank) and corresponding to 80 individuals from different wild populations and 8 individuals from other Danaus species (D. erippus, D. chrysippus, D. eresimus, D. gilippus) [30] in order to identify molecular signatures that might be associated with particular selection pressures. The reads from D. plexippus samples were mapped onto the reference genome of D. plexippus [30] with Bowtie2 (2. 2. 4) [83] using default parameters. The Ben regions were then extracted with Samtools [84]. A consensus sequence was obtained for each sample (minimum coverage of 5 reads and minimum frequency of a variant for the individual to be considered as heterozygote = 0. 25). Sequences with a missing base ratio above 50% due to heterogeneity in the sequencing were discarded. Reads corresponding to the two Ben9 copies could not be separated due to high similarity, therefore Ben9 was analyzed as a single gene. We performed de novo assembly of the reads from the other species using Velvet 1. 2. 07 [85] and a k-mer length of 31 and the Ben genes were identified by BLAST. All the sequences were aligned using MAFFT [86], the sequences with a STOP codon were corrected in order to use them for the selection analyses. The global dN/dS (ω) ratios was measured using two different methods, AnalyseCodonData implemented in HyPhy [87] and Codeml from PAML [88] with a means ω ratio for all branches (model = 0) and one ω value for all sites (NSsites = 0). Then, the ω ratio was measured for each site along the alignment, in order to identify regions with a particular selection signature, using the SLAC method implemented in HyPhy (significance thresold: p-value ≤ 0. 1) and Codeml from PAML [88] with a means ω ratio for all branches (model = 0) and selection model (NSsites = 2), which classify the sites into three classes (ω = 0 neutral, 0<ω< 1 under negative selection and ω>1 under positive selection). These analyses were performed using the phylogeny based on genome wide SNP data [30]. The phylogenetic tree shown in Fig 3 was built based on the common region of Ben4 and Ben9 using PHYML and substitution model HKY85 and 1000 bootstraps. A universal genome walking kit (Clontech) was used for the sequencing of the whole integrated BV2-5 and Se-BLL2 contigs. For this purpose, three S. exigua genomic DNA libraries [89] were subjected to primary and secondary PCRs using the general primers provided by the kit and specific primers designed to amplify 5’ and 3’ flanking regions of the BV2-5 and Se-BLL2 open reading frames (ORFs) (S1 Table). The amplified fragments were purified, cloned into the pGEM-T Easy vector and sequenced. The presence and abundance of mRNA of Se-BV2-5 and Se-BLL2 in different larval tissues were analyzed by quantitative reverse transcription PCR (qPCR). Briefly, total RNAs from fat body, midgut and hemocytes were isolated from untreated 5th instar larvae using the RNAzol reagent (Molecular research center, INC) as described in the manufacturer’s protocol. A total of 1 μg RNA was reverse transcribed into cDNA with oligo- (dT) primer using SuperScript II reverse transcriptase (Invitrogen). cDNAs were used to determine the level of transcripts for each gene by qPCR. Reactions were carried out using an ABI Prism 7700 thermocycler from Applied Biosystems. SYBR green Ex Taq master mix (Clontech) was employed in a total volume of 20 μl. Specific primers for each gene were designed by Primer Express Software (Applied Biosystems) (S1 Table). For each gene, at least three biological replicates were employed. Data are presented as fold change using the method of 2-ΔΔCt and normalized to the internal control gene, ATP synthase. The effect of baculovirus infection on the expression pattern of the lectins in the larval midgut was determined. Third instar larvae were orally infected with S. exigua nucleopolyhedrovirus (SeMNPV). Each larva was fed with 104 occlusion bodies (OBs). Midguts of treated and untreated larvae were collected 72 h after treatment [66]. Total RNA from the treated and control larvae were collected as described above. cDNA was reverse-transcribed and the presence and abundance of the mRNA was determined using qPCR as described above. The full ORF of the two main allelic forms of BV2-5 (complete and truncated) were amplified by PCR from cDNA obtained from S. exigua FRA and Xen-R larvae, respectively. They were cloned into pFBD-pH vector, downstream of the p10 promoter to generate pFBD-pH-BV2-5 (for the complete form) and pFBD-pH-BV2-5t (for the truncated form) vectors. pFBD-pH refers to the dual vector pFBD (Clontech) containing the AcMNPV polyhedrin gene downstream of the PH promoter. In order to generate recombinant baculoviruses, Escherichia coli strain DH10Bac that contains the AcMNPV ΔCC bacmid [90] and the pMON7124 helper plasmid [91] was transformed with pFBD-pH-BV2-5, pFBD-pH-BV2-5t, or pFBD-pH plasmids according to a standard procedure described for the Bac-to-Bac system (Invitrogen). Recombinant bacmids were selected based on white-blue screening of DH10Bac colonies and the positive clones were confirmed by PCR. Bacmid DNAs were isolated from bacterial cells according to standard procedure and used to transfect S. frugiperda ovary-derived cell line Sf21 using Insect Gene Juice Transfection Reagent (Novagen). Four to six days post transfection, the recombinant ΔCC-pH, ΔCC-pH-BV2-5 and ΔCC-pH-BV2-5t bacmid-derived viruses were collected and multiplied to produce high-titer stocks for further experiments. Another type of construct was generated to study the cellular localization of BV2-5 by expressing this protein fused to GFP. Primers containing BglII and EcoRI were designed to amplify the BV2-5 gene from the pFBD-pH_BV2-5. The obtained fragment was sub-cloned into pGEM-T Easy, double digested with BglII and EcoRI, and cloned in p166AcV5-Se8-GFP [92] in order to obtain the fusion gene BV2-5_GFP in the plasmid p166AcV5-Se8-BV2-5GFP. Subsequently, the GFP gene and the recombinant BV2-5GFP gene were amplified using specific primers that contained the NotI and PstI restriction sites (Forward BV2-5GFP: 5’ TTGCGGCCGCATGTTGCCTATTACC3’; Forward GFP: 5’ CTGCGGCCGCATGGGCAAAGGAGAAGAACTTT3’; Reverse: 5’AGCTGCAGTTACGACCAGCCGCCGCTGGCATCT3’). Both genes were cloned under the ph promoter of pFBD to generate pFBD-GFP and pFBD-BV2-5GFP, which were used to transpose into the AcMNPV bacmid as previously described (S4 Fig). Cellular localization of the BV2-5 protein was determined using the recombinant baculovirus expressing BV2-5 fused to GFP (AcMNPV-BV2-5GFP). Previously to the confocal analysis, Sf21 cells were maintained in Grace’s medium (Invitrogen) supplemented with 10% fetal bovine serum and 0. 5% penicillin/streptomycin at 27°C. A first set of cells was infected with AcMNPV-BV2-5 GFP and a second set with AcMNPV-GFP at a multiplicity of infection (MOI) of 10. A third set of cells was infected with AcMNPV-GFP and treated with 5 μM latrunculin A (Sigma Aldrich) 12 hours post infection (hpi). A fourth group of cells was maintained without any treatment as a negative control. Seventy-two hpi, cells were pelleted by centrifugation for 2 min at 3000xg and fixed with 4% paraformaldehyde (PFA) for 20 min. Then, the cells were washed twice with PBS and permeabilized for 10 min with 0. 2% Triton X-100 in PBS-BSA 10%. After another step of PBS washing, cellular actin was stained overnight at 4°C with phalloidin-tetramethylrhodamine B isocyanate TRITC (Sigma Aldrich). Finally, the cells were washed, stained with DAPI (4’, 6’-diamidino-2-phenylindole) to visualize the nucleus of cells and fixed by dakocytomation fluorescent mounting medium (Dakocytomation). Mounted cells were observed under confocal microscope (FV1000, OLYMPUS). Effect of BV2-5 on baculovirus multiplication in cell culture was determined by a one-step growth curve assay. Sf21 cells were infected with the different recombinant baculoviruses at an MOI of 2. After infection, cells were washed and incubated in fresh medium. At different time points, an aliquot of medium was harvested and the viral titer (amount of budded viruses) in each sample was determined by qPCR. For that purpose, viral DNAs were extracted using Prepman reagent (Applied Biosystems) following the manufacturer protocol and were quantified by comparing the obtained Ct values against a standard curve of known viral concentration. Three independent replicates were performed for each sample. Recombinant Se-BLL2 was expressed and produced in an Escherichia coli expression system and purified with affinity chromatography using the HiTrap Chelating HP column (GE Healthcare). In order to test the effect of BLL2 on baculovirus infection, the purified protein was added in different concentrations (50 μg/mL, 10 μg/mL, and 1μg/mL) to AcMNPV-GFP virions in presence of 10mM CaCl2 and the mixture was incubated for 2 hours. After that, the mixture of virus-lectin was then used to infect Sf21 cells (MOI of 0. 5). Similarly, other sets of cells were infected with AcMNPV-GFP or AcMNPV-GFP incubated with 10 mM CaCl2 as controls. Thirty-six hours post infection, percentage of cells showing GFP was determined for the different treatments in order to compare the virus entry to cells. In addition, an aliquot of the medium was harvested at different time points, and the virus titer was determined for each sample and time point by qPCR as described above. S. exigua third instar larvae were infected with Spodoptera exigua multiple nucleopolyhedrovirus (SeMNPV) by the drop-feeding method. Occlusion bodies (OBs) (5x105) from SeMNPV were added to a solution containing sucrose and phenol red colorant (10% and 0. 05%, respectively) in presence or absence of purified Se-BLL2 (0. 15mg/mL). The larvae were allowed to drink from the virus and control solution in Petri dishes and then transferred individually to the assay plates. Mortality was then recorded every 12 h until death or pupation of all the larvae. Sixteen larvae were used for each treatment and three independent replicates were performed. Mortality was expressed as the percentage of dead larvae. The time to death was assessed by comparing the mortality curves using the Kaplan Meier method (GraphPad Prism 5). The statistical significance was determined using the log-rank analysis (Mantel-cox test). Insect bioassays comparing susceptibility to SeMNPV in two different populations (SUI and MEX) of S. exigua were performed as described above at a final dose of 103 OBs/larva. The putative ORFs were determined with the EditSeq program from DNASTAR and the homologs in other insect species were obtained using BLAST comparison at NCBI (http: //www. ncbi. nlm. nhi. gov), Silkbase (http: //silkworm. genomics. org. cn), Manduca sexta genome project (http: //agripestbase. org/manduca) and LepidoDB (http: //www6. inra. fr/lepidodb). The predicted amino acid sequences were aligned using the ClustalX software [93] and visualized in GenDoc program [94]. Evolutionary distance was calculated for aligned sequences by Maximum-likelihood method and the phylogenetic trees were conducted with the MEGA5 program [95]. Reliability of an inferred tree was determined using bootstrap test (1000 replicates). For a clearer view of the branches, bootstrap values are reported over 100. For the Lectins comparison, the names and accession numbers of proteins compared were as follows. Sf lectin 3–1 (Sf1H08856-3-1) and Sf lectin 5–1 (Sf2H07501-5-1): Spodoptera frugiperda proteins obtained from Spodobase (http: //bioweb. ensam. inra. fr/spodobase); littoralis_C2971: S. littoralis lectin-like protein; Sl_lectin: lectin-like protein from S. litura; Se-BLL1-6: S. exigua bracovirus-like lectins (KP406769-74). CsMBV CTL CrPDV HP, CvBV L, CpPDV lectin, CrBV lectin, CcV3, CvBV 2L: C-type-lectins from bracoviruses of Cotesia species, (AGO14401. 1), (BAC55179. 1), (AEE09593. 1), (AAS10157. 1), (AAO74641. 1), (CCQ71085. 1), (AEE09562. 1), Gi-CTLD, Gi-LRP, Gf CTLD2, Gf CTLD3, Gf CTLD4: C-type-lectins from bracoviruses of Glyptapanteles species (ABK56997. 1), (ABK56993. 1), (ACE75074. 1), (ACE75072. 1), (ACE75071. 1). Nv HLPB, Mr LBP, Mr HLBP1, Mr HLPB, Mr_HLBP C-type-lectins from Hymenoptera, Nasonia vitripennis, Megachile rotundata Megachile rotundata, Microplitis demolitor (XP_001599898. 2), (XP_003708137. 1), (XP_003701025. 1), (XP_003706756. 1), (XP_003704952. 1) (XP_008555202. 1), Se-LL1,2 and 3: Spodoptera exigua lepidopteran-like lectins 1,2 and 3 (KP406775-77), Bm CTL19: B. mori C-type lectin 19 (NP_001165396. 1), Bm CTL21: B. mori C-type lectin 21 (NP_001037056. 1), Ms IML4: Manduca sexta immunolectin 4 (AAV41237. 2), Ms_IML 4 2: M. sexta immunolectin 3 (AAV41236. 1), Ms IML2: M. sexta immunolectin 2 (AAF91316. 3), Ha CTL8: Helicoverpa armigera C-type lectin 8 (AFI47453. 1), Ha LCT6: H. armigera C-type lectin 6 (AFI47451. 1), Ha CL2: H. armigera C-type lectin 2 (ACI32834. 1), Of_IML: Ostrinia furnacalis immunolectin (ABZ81710. 1), Lo IML1: Lonomia oblique immunolectin 1 (AAV91436. 1), Lo L3: Lonomia oblique lectin 3 (AAV91450. 1), Ap_CTL: Antheraea pernyi C-type lectin (AGN70857. 1), Mc_CTL: Mamestra configurata C-type lectin (AEA76325. 1), Pr_CTL: Pieris rapae C-type lectin (AEO52696. 1), As_CTL: Anopheles stephensi C-type lectin galactose binding (ACP43727. 1), Ae_CTL: Aedis aegypti C-type lectin (ABF18196. 1), Md_UP: Musca domestica uncharacterized protein LOC101901048 (XP_005189940. 1), Dv_GJ: Drosophila virilis GJ17272 (XP_002051932. 1), Dm BCTL: Drosophila melanogaster C-type lectin 27kD, isoform B (NP_001260046. 1), De_GG: Drosophila erecta GG24353 (XP_001968708. 1), Dm CTL: Drosophila melanogaster C-type lectin 27kD, isoform A (NP_608858. 3), Dy_GE: Drosophila yakuba GE14680 (XP_002087961. 1), Dw_GK: Drosophila willistoni GK23915 (XP_002064562,1), Dmoj GI: Drosophila mojavensis GI15343 (XP_002001743,1). For the BV2-5 comparision, the names and accession numbers are: S. exigua_BV2-5: S. exigua BV2-5 (KP406767); S. littoralis BV2-5: S. littoralis BV2-5; S. litura BV2: S. litura BV2-5 (GBBY01010418. 1); CcBV_BV2-5: C. congregata bracovirus hypothetical protein 3 segment 25 (CCQ71080. 1); ; GI_HP1 and GI_HP2: Glyptapanteles indiensis hypothetical proteins L1_00460 and L1_00290 (ABK57032. 1 and ABK57015. 1); GF_CHP1 and GF_CHP2: Glyptapanteles flavicoxis hypothetical proteins (ACE75094. 1 and ACE75115. 1). S. frugiperda genomic bacs at NCBI, Genbank acc: FP340419. 1 and FP340412. 1 for BV2-5 and Se-BLL2, respectively.

Eukaryotes are generally thought to evolve mainly through the modification of existing genetic information. However, evidence of horizontal gene transfer (HGT) in eukaryotes-the accidental acquisition of a novel gene from another species, allowing acquisition of novel traits—is now recognized as an important factor in their evolution. We show here that in several lineages, lepidopteran genomes have acquired genes from a bracovirus that is symbiotically used by parasitic wasps to inhibit caterpillar host immune defences. Integration of parts of the viral genome into host caterpillar DNA strongly suggests that integration can sporadically occur in the germline, leading to the production of lepidopteran lineages that harbor bracovirus sequences. Moreover, some of the transferred bracovirus genes reported here originate from the wasp genome, demonstrating that a gene flux exists between the two insect orders Hymenoptera and Lepidoptera that diverged ≈300 MYA. As bracovirus gene organisation has evolved to allow expression in Lepidoptera, these transferred genes can be readily domesticated. Additionally, we present functional analyses suggesting that some of the acquired genes confer to caterpillars a protection toward baculovirus, a very common pathogen in the field. This phenomenon may have implications for understanding how caterpillars acquire resistance against baculoviruses used in biological control.

Abstract Introduction Discussion Materials and Methods

2015

Recurrent Domestication by Lepidoptera of Genes from Their Parasites Mediated by Bracoviruses

INDETERMINATE DOMAIN (IDD) / BIRD proteins are a highly conserved plant-specific family of transcription factors which play multiple roles in plant development and physiology. Here, we show that mutation in IDD4/IMPERIAL EAGLE increases resistance to the hemi-biotrophic pathogen Pseudomonas syringae, indicating that IDD4 may act as a repressor of basal immune response and PAMP-triggered immunity. Furthermore, the idd4 mutant exhibits enhanced plant-growth indicating IDD4 as suppressor of growth and development. Transcriptome comparison of idd4 mutants and IDD4ox lines aligned to genome-wide IDD4 DNA-binding studies revealed major target genes related to defense and developmental-biological processes. IDD4 is a phospho-protein that interacts and becomes phosphorylated on two conserved sites by the MAP kinase MPK6. DNA-binding studies of IDD4 after flg22 treatment and with IDD4 phosphosite mutants show enhanced binding affinity to ID1 motif-containing promoters and its function as a transcriptional regulator. In contrast to the IDD4-phospho-dead mutant, the IDD4 phospho-mimicking mutant shows altered susceptibility to PstDC3000, salicylic acid levels and transcriptome reprogramming. In summary, we found that IDD4 regulates various hormonal pathways thereby coordinating growth and development with basal immunity. Plants and animals use pattern recognition receptors (PRRs) to rapidly activate defense signaling pathways and immune responses upon pathogen attack [1,2]. PRR receptors and their associated signaling components possess a wide range of similarities in mammals, plants and invertebrates [3,4]. Plant immunity relies on the recognition of pathogen-derived molecules in order to activate pattern-triggered immunity (PTI) and effector-triggered immunity (ETI) [5]. PTI is initiated after the perception of pathogen-associated molecular patterns (PAMPs) of highly conserved pathogen components. FLAGELLIN22 (flg22), a 22 amino acid peptide from within bacterial flagellin protein, represents one important PAMP to trigger PTI in plants and is perceived by the plasma membrane-localized receptor FLAGELLIN-INSENSITIVE2 (FLS2) which associates with BRI1-ASSOCIATED RECEPTOR KINASE (BAK1) in order to rapidly stimulate two mitogen-activated protein kinase (MAPK) cascades. MAPK cascades consist of three sequentially activated kinase modules composed of a MAPK kinase kinase, a MAPK kinase and eventually a MAPK, thereby linking upstream signals to downstream targets. In Arabidopsis as well as throughout the plant kingdom, the MAPK orthologues of MPK3, MPK4 and MPK6 represent the final step in the two flg22-activated MAP kinase cascades and transmit signals to respective target proteins by phosphorylation [6,7]. MPK3, MPK4 and MPK6 are required for full activation of defence genes [8]. In particular, MPK3/MPK6 contributes to bacterial and fungal resistance [9] as well as to a multitude of developmental processes, including the regulation of plant architecture, seed, root [10] and stomatal formation [11]. Furthermore, the MPK6 signaling module participates in nutrient signaling to influence nitrate assimilation, enhances phosphate acquisition and becomes activated upon iron deficiency [12]. Recently, it was shown that MPK3/MPK6 exert essential functions in the induction of camalexin, the major phytoalexin in Arabidopsis, and promote the indole glucosinolate biosynthesis pathway [9]. In addition, MPK3/MPK6 activation rapidly alters the expression of photosynthesis-related genes and inhibits photosynthesis, which promotes the accumulation of superoxide (O2) and hydrogen peroxide (H2O2), two major reactive oxygen species (ROS), in chloroplasts under light [13]. The INDETERMINATE DOMAIN (IDD) /BIRD family of transcription factors (TF) is highly conserved in both monocots and dicots and functions in multiple developmental processes [14,15]. IDDs are a plant-specific group of TF comprising of 16 members in Arabidopsis which are characterized by a conserved N-terminal ID domain composed of four zinc fingers (ZFs) and a long undetermined sequence for protein interaction [16]. The four ZFs can be subdivided into the C2H2 type ZF1 and ZF2, which are dedicated to DNA interaction, and the C2HC type ZF3 and ZF4. ZF3 and in particular ZF4 are essential for the interactions of IDD3/MAGPIE and IDD10/JACKDAW to the SHORT-ROOT (SHR) —SCARECROW (SCR) complex. By contrast, ZF1-ZF2-ZF3 of IDD3 and IDD10 are involved in DNA binding [17]. In Arabidopsis, IDD4/IMPERIAL EAGLE functions in ad-/abaxial leaf development and leaf blade formation and its expression is subject to KANADI1 and the HD-ZIPIII family protein REVOLUTA. [18]. Furthermore, among other IDD family members, IDD4 contributes to the root ground tissue organisation and coordinates the differentiation of the endodermis initial stem cell niche in order to give rise to cortex and endodermis cells [16,19,20] and serves as transcriptional scaffold to enable transactivation activity of the gibberellin-inhibitor DELLA/RGA proteins of the GRAS-family in association with the transcriptional regulator SCARCROW-like 3 (SCL3) [16,21,22]. Several independent studies identified IDD4 to be phosphorylated on serine-73, a highly conserved putative MAPK motif that is conserved in all family members of the IDD/BIRD family [23–29]. Moreover, using an inducible MAPKK-activation system, Ser73 [29] was identified as a target of the MAPKs MPK3 and MPK6. These studies prompted us to test IDD4 as putative regulator and MAPK substrate in plant innate immunity. Transcriptome and global ChIP-SEQ analysis of idd4 and IDD4ox plants revealed that IDD4 plays a role in coordinating innate immunity with growth and development. ChIP-qPCR analysis showed that flg22-treatment correlates with the recruitment of IDD4 to ID1 motif-containing promoter regions. Moreover, IDD4 interacts and becomes phosphorylated by the immune MAPK MPK6, and that IDD4-phosphomimichking versions show enhanced DNA-binding and transcriptional activity of ID1 motif-containing promoters. Phosphosite-mutated IDD4 plants show opposite susceptibility to pathogen attack and transcriptome reprogramming, confirming the function of IDD4 in regulating genes related to immunity and plant-growth. So far, it was reported that IDD4 is expressed in the root ground tissue of the basal meristem [20]. To investigate the contribution of IDD4 in defense response, its expression was analysed by generating stably transformed Arabidopsis lines expressing GUS or NLS: 3xGFP under a 2. 5 kb IDD4 promoter sequence. Intense staining of the GUS reporter was observed in cotelydons, root tips and in all stages of rosette leaf development (Fig 1A and 1B). In leaves, we detected GUS/GFP signals in the trichomes, stomata, epidermal cells (S1A–S1D Fig) and mesophyll cells (S1E Fig). Moderate GUS staining was observed in sepals and petals (Fig 1C) as well as in ovules embedded in carpels of Arabidopsis flowers (S1F Fig). Various public microarray datasets (Genevestigator) (S1G Fig) showing endogenous IDD4 transcript abundance corresponds to our histological results and reveal the expression of IDD4 in a wide range of tissues throughout the life cycle in Arabidopsis. The functional contribution of IDD4 in response to bacterial pathogen attack was investigated by challenging idd4 mutant, a complementation line (idd4/pIDD4: : IDD4: YFP) and an overexpressor line IDD4ox1 (pUBI10: : GFP: IDD4) with the virulent hemi-biotrophic plant pathogen Pst DC3000. We used an IDD4 insertion line (Salk_148352) containing the T-DNA insertion in the first exon that could be confirmed by sequencing as a true knockout line (Fig 1D and 1E and S2A Fig). The idd4 complementation line expresses IDD4: YFP driven by the 2. 5 kb upstream sequence of the IDD4 ORF. Interestingly, by determining the fresh weight of 18 day-old plants, we consistently found an increase of approximately 20% biomass in the aerial part of the idd4 mutant. By contrast, the shoot fresh weight of the four analyzed IDD4ox lines showed a reduction of about 11. 5% in IDD4ox1 (Fig 1F and 1G) up to approximately 65% in IDD4ox4 when compared to WT (S2B–S2D Fig). Interestingly, the growth reduction seems to be in accordance with the ectopic expression of IDD4 in IDD4ox1-4 lines (S2B Fig). The differences in shoot growth also corresponded to altered root formation in the idd4 and IDD4ox lines. The root biomass in idd4 is increased by about 36% whereas that of the IDD4ox lines show a reduction of approximately 14% (Fig 1F and 1H) in IDD4ox1 and 75% in IDD4ox4 (S2C and S2D Fig). The enhanced growth of idd4 could be reverted to WT by the expression of the complementation construct (S2E Fig). The restored phenotype in idd4/pIDD4: : IDD4: YFP shows that the improved growth can be traced back to the mutation in IDD4. The differences in shoot and root growth in idd4 and IDD4ox lines suggest a function of IDD4 as a regulator of growth-associated processes. Two hours after spray infection by PstDC3000, the infection levels in the different transgenic lines corresponded to those in WT plants indicating that stomatal immunity was not affected (Fig 1I). However, 72 hours after spay-infection, the proliferation levels of Pst DC3000 in the idd4 mutant were significantly reduced when compared to WT. Furthermore, the bacterial titer in the idd4 complementation line was indistinguishable from WT, suggesting that the reduced susceptibility in idd4 mutants is due to the lack of IDD4 protein function. By contrast, the IDD4ox line exhibited increased susceptibility to Pst DC3000 (Fig 1I). Therefore, we reasoned that IDD4 acts as a negative regulator of basal resistance to hemi-biotrophic pathogen infection. To evaluate the PTI response, the idd4 mutant was challenged by Pst DC3000 hrcC-. The Pst DC3000 hrcC- strain is compromised in virulence due to its inability to inject any of its type III-secretion system-dependent effectors, one function of which is to suppress plant immunity. In this way, infection with Pst DC3000 hrcC- principally induces only PTI-mediated defense responses. In comparison to WT plants and idd4 complementation lines, the proliferation levels of the bacteria 72 hrs after spray-infection were reduced in idd4 mutants and elevated in IDD4ox (Fig 1J). The higher resistance of the idd4 mutant indicates an enhanced PTI response thereby suggesting that IDD4 also functions to regulate PTI-mediated defense responses. In order to analyse the transcriptome composition of idd4 and IDD4ox lines, we performed RNA-Hiseq analysis on 3 biological replicates of 14 day-old idd4, IDD4ox and WT seedlings without and after flg22 application (1μM flg22,1hr). A close to linear correlation coefficient of WT and idd4 (0. 85), WT and idd4 (flg22) (0. 87), as well as WT and IDD4ox (0. 98) was obtained when considering the expression profiles for all transcripts. The strict correlation suggests that IDD4 does not affect general gene expression, but rather influences subsets of genes in particular biological processes. Hierarchical clustering of significant genes (p<0. 05) in idd4 before and after flg22 treatment, by using normalized FPKM values, revealed distinct differences in gene expression patterns suggesting altered gene induction in idd4 after flg22 perception (Fig 2A, S1 Table). At a stringency of p<0. 05 in untreated idd4 mutants, 2244 differentially expressed genes (DEGs) (S2 Table) could be identified that show a log2-fold change from 0. 27 (1. 21 FC) to 4. 05 (16. 57 FC) of positively regulated genes and from -4. 95 (30. 88 FC) to -0. 27 (1. 21 FC) of negatively regulated genes. Among these 2244 genes, 621 genes are up- and 1623 genes are down-regulated. To categorize DEGs in functional modules, gene ontology (GO) terms were determined by using the AgriGO platform [30] (TAIR9) (Fig 2B, S2 Table). The up-regulated genes can be grouped in very different GO terms describing gene functions in different hormonal pathways and a multitude of cellular biological processes. GO terms are highlighted for defense response, response to salicylic acid stimulus, oxidative stress and response to other organism thereby indicating a function of IDD4 in the repression of defense-related genes and factors contributing to growth and development. Intriguingly, transcript levels of CALMODULIN-BINDING PROTEIN 60g (CBP60g) [31] that regulates expression of the rate-limiting enzyme ICS1 in SA biosynthesis [32] and the SA marker gene PATHOGENICITY-RELATED FACTORS PR2 were expressed at significantly higher levels in untreated idd4 plants compared to WT (Fig 2C and S2F Fig). Similarly, enhanced expression was found in idd4 mutant plants for the pattern-triggered immunity-responsive marker gene FLG22-INDUCED RECEPTOR-LIKE KINASE 1 (FRK1) [33], and the early-defense marker transcription factor WRKY22 (Fig 2C and S2F Fig). Interestingly, the enhanced growth phenotype of the idd4 mutant corresponds with GO terms describing gene functions for response to auxin stimulus, glucosinolate metabolic process and anatomical structure development. The enrichment of auxin response genes, including AUXIN-REGULATED GENE INVOLVED IN ORGAN SIZE (ARGOS) and several members of the SMALL AUXIN UPREGULATED RNA (SAUR) -like auxin-responsive protein family (Fig 2C, S2 Table), promoting plant growth and architecture [34], in idd4 mutant plants might partly explain its elevated biomass. In addition, the idd4 transcriptome was enriched in genes of the glucosinolate metabolism that participates both in defense and growth [9,35]. After flg22 treatment, 2048 DEG (p<0. 05) were obtained in idd4 mutant, with 1063 genes showing enhanced while 985 genes reduced transcript abundance (S2 Table). In accordance with the untreated idd4 transcriptome, the GO analysis of up-regulated genes after flg22 application emphasized GO terms for hypersensitive response, oxidative stress and defense response (Fig 2D). In particular, the expression of the defense markers WRKY33, WRKY49 and FRK1 is elevated as well as components of the kinase signaling cascades represented by MKK6 and MPK11 (Fig 2E). Additionally, macromolecular biosynthesis process and protein metabolic process are among the most significant GO terms (Fig 2D, S2 Table). By contrast to the idd4 mutant, the overexpression of IDD4 (IDD4ox) reduces defense-related gene expression depicted by the GO analysis of significantly down-regulated genes (p<0. 05) (S2 Table). These DEGs are functionally grouped in GO terms for respiratory burst during defense response, salicylic acid stimulus and innate immune response (Fig 2H). Prominent representatives are WRKY38 [36], PR5 as well as ERF4 and ERF5 [37] (Fig 2I). The deregulation of genes and functional groups dedicated to immunity in the idd4 mutant and IDD4ox lines confirms the role of IDD4 as a regulator of defense-related gene expression. In accordance, the GO analysis before and after flg22 treatment show DEGs involved in the response to reactive oxygen species (ROS) which belong to the first line of defense upon pathogen invasion [38]. Intriguingly, genes coding for the H2O2-scavenging enzymes CATALASE 1 (CAT1) and CAT2, which catalyze the reduction of photo-respiratory generated hydrogen peroxide and protect cells from its toxicity [39], are down-regulated in idd4 (S2 Table). Interestingly, SA-mediated suppression of CAT2 results in increased H2O2 levels and resistance against biotrophic pathogens [40]. Furthermore, PEROXIDASE CA (PRXCA) [41], ASCORBATE PEROXIDASE 1 (APX1) [42] and L-ASPARTATE OXIDASE (AO) [43] are up-regulated after flg22-treatment in idd4 and have been reported to promote H2O2 metabolism and turn-over. By comparing the levels of the ROS H2O2 by 3,3' -diaminobenzidine (DAB) staining in untreated idd4 with WT plants, we observed that the H2O2 levels in idd4 were strongly elevated already before infection (Fig 2F). In the mock-treated control without applied DAB, the staining value is about 180 of an arbitrary unit for WT and idd4 and can be considered as default level. After DAB staining, WT-plants showed a staining value of 165 and idd4 of about 156. This staining difference indicates a higher H2O2 accumulation in the idd4 mutant. Additionally, in a comparative flg22-triggered reactive oxygen species burst assay, we detected strikingly enhanced ROS levels in idd4 mutants and reduced ROS efflux in IDD4ox lines when compared to WT (Fig 2G). Notably, maximum ROS levels were observed in WT at 17 minutes whereas the idd4 mutants achieved maximal ROS production already at 7 minutes after flg22 perception. These results show that the GO term annotations response to oxidative stress in idd4 mutant and respiratory burst during defense response in IDD4ox plants correspond with their biochemical traits. In summary, the constitutive activation of ROS and defense-related traits and the enrichment of factors that promote structural development indicate a role of IDD4 in growth and immunity-related processes. In order to identify the genomic regions targeted by IDD4 in vivo, Chromatin-Immunoprecipitation (ChIP) was performed followed by deep sequencing on the Illumina High-Sequencing platform (SEQ). In order to evaluate the reproducibility of the obtained genome-wide IDD4 binding profiles, ChIP-SEQ was performed on 3 independent lines carrying the IDD4: GFP construct under the control of the UBI10 (At4g05320) promoter. The retrieved binding patterns were normalized by comparison with data obtained from pUBI10: : GFP plants. GFP antibodies were used to immunoprecipitate protein-DNA complexes that were verified by Western-Blot analysis before sequencing (S3A Fig). Sequence coverage at each position on the genome was plotted to identify peaks in the Arabidopsis genome. A significant number of peaks per biological replicate with an FDR <0. 005 could be annotated (Fig 3A) in the 500 bp region upstream of the translational start sequence in the genic region. A highly consistent read pattern in all three IDD4: GFP lines showed predominate IDD4 binding in the vicinity upstream of the transcriptional start site (Fig 3C), which includes the 5’-UTRs and promoter regions. Sequence analysis of the 500 bp regions upstream of the TSS of IDD4 targets revealed the in vivo enrichment of the ID1 motif on a genome-wide scale in about 31% of target genes (Fig 3B) and was ranked as the primary binding site. Previously, the core sequence of the ID1 motif (AGACAA) was initially suggested as binding site of the maize ID1 protein [44], and further in vitro characterized for its association with IDD3 and IDD8 in Arabidopsis [16,45]. To compare the independent transgenic lines, the binding of IDD4: GFP to the same target sequences was monitored by co-occurrence matrix analysis (S3C Fig). The comparison illustrates that replicates 1 and 3 provide the most reliable overlapping binding sites. After removing genes that showed similar sequence peaks in the negative control, 837 genes could be defined as significant binding targets of IDD4 (S3 Table). A GO term analysis of the 837 major IDD4 target genes highlighted gene functions in innate immune response, response to salicylic acid stimulus, hypersensitive response and respiratory burst during defense response (Fig 3D). These GO terms demonstrate that a significant part of the direct IDD4 targets are assigned to functions in defense response and pathogenicity (S3 Table). On the other hand, IDD4 exhibits binding preferences on genes that mediate cellular processes governing growth and development, including auxin and glucoside metabolic process as well as response to gibberellin and light stimulus. To gain a more precise view on global IDD4 function, target genes were grouped into protein interaction networks by performing gene cluster analyses using STRING [46]. Versatile functional networks could be generated with IDD4 in various distinct biological processes (Fig 3E and 3H–3J). Several genes are integrated into a ramified functional network that contributes to defense-associated processes (Fig 3E), including WRKY33, whose phosphorylation by MPK6 promotes the transcriptional induction of camalexin biosynthesis genes [47]; ERF104, a key regulator of basal immunity whose stability and interaction with MPK6 is compromised in response to flg22 [48]; AP2C1, which interacts and dephosphorylates MPK6 and consequently suppresses immunity [49]; CAF1a, a mRNA deadenylase targeting transcripts for post-transcriptional modification with temporal specificity during plant defense response [50], and BIR1, encoding a BAK1-interacting receptor-like kinase and activates plant defense responses [51]. IDD4 binding profiles of AP2C1 and ERF104 generated by ChIP-SEQ and ChiP-qPCR confirmed the association of IDD4 to the 5’upstream region of the respective gene loci (Fig 3F and 3G). The association of IDD4 to these target genes indicates its direct influence on factors shaping immunity response (Fig 3E). Furthermore, a prominent hub describes IDD4 function in the ad/abaxial pattern specification and leaf blade formation by regulating REVOLUTA (REV), KANADI (KAN), PHABULOSA and ABNORMAL FLORAL ORGANS (Fig 3H and 3K) [18]. A further functional node combining factors involved in auxin-signaling was formed and comprises the AUXIN RESPONSE FACTOR ARF9 and ARF10, and the INDOLE-3-ACETIC ACID INDUCIBLE IAA18 and IAA26, as well as one member of the SAUR family, suggesting a regulatory role of IDD4 in auxin-mediated growth processes (Fig 3I and 3K). Moreover, recently published data indicate IDD4 as a transcriptional activator of nuclear-encoded photosynthetic gene expression and photomorphogenesis [52]. In this context, chloroplast maturation and import machinery, as well as chlorophyll biogenesis, seems to be targeted by IDD4 (Fig 3J), as shown by the binding to the promoter regions of PHYTOCHROME B, PHYTOCHROME INTERACTION FACTOR (PIF) 3, PIL1, PIL2/PIF6, PIL5/PIF1 [53] (Fig 3K) and Chloroplast heat shock protein 70–1 (cpHsc70-1) [54] together with DE-ETIOLATED 1 (DET1) [55]. Moreover, RNA POLYMERASE SIGMA SUBUNIT 2 (SIG2) [56] encodes a subunit of chloroplast RNA polymerase, and REDUCED CHLOROPLAST COVERAGE 2 (REC2) [57] which contributes to the size establishment of the chloroplast compartment are two of several further IDD4 target genes that are involved in the chloroplast biogenesis and function (S3 Table). To pinpoint whether IDD4 acts as a positive or negative regulator of major ChIP-SEQ target-genes, we compared DEGs identified in the transcriptome analysis of the idd4 mutant and IDD4 overexpression lines. The comparison of CHIP-SEQ targets with DEGs in idd4 and IDD4ox revealed an overlap of 11. 7% and 6. 7%, respectively. Altogether, the search for direct targets yielded 135 genes (S4 Table) that can be grouped by GO terms for glucoside metabolic process (p = 1. 00e-08), defense (p = 1. 10e-06) and immune response (p = 8. 10e-05). The predominantly opposite regulation of these target genes in the two genetic backgrounds suggests that IDD4 functions as a direct regulator of the respective genes. In this regard, hierarchical clustering of differentially regulated major targets yielded 49 genes matching the criteria of being predominantly up-regulated in idd4 and down-regulated in IDD4ox which indicates IDD4 as transcriptional repressor of this cluster (Fig 4A, S4 Table). The corresponding GO terms emphasise hormone-mediated signaling, response to other organism, defense response and developmental process. In addition, transcriptionally repressed targets of IDD4 are involved in various defense processes including AP2C1, CPK28, CAF1a and SERK1, thereby indicating IDD4 as a direct regulator of genes in immunity. STRING-based protein interaction network of these 49 genes created clusters combining target genes involved in immunity like AP2C1, CAF1a, CPK28 and ERF2 with those involved in vesicle-transport, like EXO70H7 [58] (S5A Fig). On the other hand, 86 genes identified in the DEGs and CHIP-SEQ overlap predominantly correspond to down-regulated genes in idd4 but enhanced expressed in the IDD4ox lines (Fig 4B, S4 Table). The selected genes can be grouped in GO terms for response to blue light, response to salt and carbohydrate biosynthesis process. Furthermore, the supported GO terms indoleacetic acid biosynthesis, and glucoside metabolic process refer to IDD4 function as a regulator of genes that participate in growth and development. Gene cluster analyses using STRING created several functional modules (S5B Fig) for light response/ photorespiration/ photo-morphogenesis and chloroplast import machinery emphasizing again the contribution of IDD4 in processes mediating growth and development. Noteworthy, the expression of IDD4 is compromised by defects in the chloroplast import machinery and the retrograde transport, and it is postulated to act as a transcriptional activator of nuclear-encoded photosynthetic gene expression [52]. In this context, the expression of TRANSLOCON AT THE OUTER MEMBRANE OF CHLOROPLASTS 64-III (TOC64-III) [59], Ankyrin repeat-containing protein 2 (AKR2) [60], EARLY LIGHT-INDUCABLE PROTEIN (ELIP1) [61], Tonoplast dicarboxylate transporter (TDT) [62] and VARIEGATED 1 (VAR1) [63] which collectively contribute to chloroplast maturation is subject to IDD4 function (S5B Fig). In addition, our data indicate that IDD4 exerts transcriptional regulation of the blue light photoreceptor PHOTOTROPIN2 (PHOT2) [64] and the J-DOMAIN PROTEIN (JAC1) [65,66] with assigned functions in the adaptation to light stress (blue light) and photo-morphogenesis in a signal-transduction pathway for photo-chloroplast movement and accumulation (S5B Fig, S4 Table). Altogether, our data indicate that IDD4 is embedded in widely-ramified regulatory pathways thereby exerting transcriptional control on key factors that shape and balance growth and defense. To test the interaction between the immune MAPKs MPK3,4 and 6 with IDD4, we searched for MAPK docking sites in IDD4 that are essential for the binding of substrates to MAPKs. Analysis of the IDD4 amino acid sequence revealed a highly conserved MAPK docking motif that lies between ZF1 and ZF2 (S4A Fig) [67]. The interaction of IDD4 with the immune MAPKs MPK3, MPK4 and MPK6 was tested by in vitro pull-down assays using MBP-His-tagged IDD4 and GST-tagged MAPKs (Fig 5A). IDD4 predominately interacted with MPK6. The interaction with MPK3 and MPK4 was at background intensity and the negative controls, single GST and MBP, did not associate with the respective proteins. The association of IDD4 with MPK6 was then evaluated by bifluorescence complementation experiments (BiFC) in Nicotiana benthaniama. A strong fluorescence signal was detected in the nucleus of Nicotiana epidermis cells that were co-transfected with YFPN-IDD4 and YFPC-linked MPK6 (Fig 5B). As a positive control, the interaction of SCL3 with IDD4 could be confirmed [16], whereas no association could be detected for UBI10 and as well the empty vector control. The data suggest that among the three immune MAP kinases MPK6 is the principal interaction partner of IDD4. To evaluate the capability of MPK6 to phosphorylate IDD4, in vitro kinase assays followed by LC-MS/MS analysis were performed with a constitutively active version of MPK6 and recombinant IDD4 protein as a substrate. The MS/MS spectra for IDD4 (Fig 5C and 5D) show that MPK6 targets the same amino acid residue (Ser-73) that was identified in several phosphoproteomic studies [43–48]. Moreover, using an inducible MAPKK-activation system, Hoehenwarter et al. (2013) [29] identified the equivalent phosphosite as a putative target of the MAPKs MPK3 and MPK6. Intriguingly, an additional phosphosite in IDD4 (Thr-130) is phosphorylated by MPK6 in the highly conserved N-terminal ID domain (Fig 5E, S4A Fig). Recently, it was shown that MPK6 targets besides the common SP and TP sites PT sites after flg22 treatment [68], suggesting a biological function of Thr-130 after bacterial perception. Notably, the first phosphorylation site resides 11 amino acids upstream of ZF1 whereas the second phosphorylation site is located inside ZF2 (Fig 5E). The post-translational modifications in this region of the ID domain suggest an inherent phosphorylation-dependent regulation mechanism for DNA-binding of IDD4 [17]. The elevated resistance of the idd4 mutant accompanied by the elevated expression of genes involved in salicylic acid (SA) metabolism and signaling suggest a function of IDD4 in these processes. The 500 bp upstream region of the translational start sequence (TSS) of the SALICYLIC ACID GLUCOSYLTRANSFERASE 1 (SAGT1) gene has two ID1 motifs in the promoter region between positions -259 and -224 upstream of the TSS (Fig 6A) and becomes bound by IDD4 with a log2 (FC) of 2. 20. This finding was confirmed in vivo by ChIP_qPCR using transgenic lines expressing IDD4-GFP under the UBI10 promoter (Fig 6B) and its native promoter (Fig 6C) showing consistently that IDD4 binds to the SAGT1 promoter region P1 and P2 (-300 to 0) close to the TSS. SAGT1 converts SA to the biologically inactive storage forms SA-2-l-β-D-glucoside (SAG) and SA-glucose-ester (SGE) [69]. In agreement with this function, transgenic SAGT1ox plants exhibited increased susceptibility to Pst DC3000 and showed reduced levels of free SA [70,71]. The expression of SAGT1 is induced by Pst DC3000 infection and Methyl-SA [69,71]. Genome-wide DNA binding and ChIP-qPCR analysis showed that IDD4 binds to the SAGT1 promoter and consequently identifies SAGT1 as a direct target of IDD4. Indeed, SAGT1 transcript levels are reduced in the idd4 mutant (log2 (FC) -0. 39, p<0. 01) in untreated conditions and after flg22 treatment (log2 (FC) -0. 4, p<0. 05) (S2 Table) when compared to WT plants. These results could be confirmed by qPCR (Fig 6D). On the other hand, SAGT1 transcript levels are enhanced in IDD4ox (Fig 6E) after flg22-treatment, indicating that IDD4 acts as a positive regulator of SAGT1 expression. To study the biological function of IDD4 phosphorylation in planta, we replaced Ser-73 and Thr-130 by aspartic acid (D) to generate a phospho-mimicking IDD4 version S-73-D, T-130-D (DD) or by alanine (A) to produce a phospho-dead IDD4 S-73-A, T-130-A (AA) version. As shown in S4E and S4F Fig, both phospho-modified IDD4 versions still interact with the M5GAI (RGA/DELLA) protein in Y2H and SCL3 in BiFC analysis. These results suggest that phosphorylation of IDD4 in the first 2 ZFs does not hamper the interaction with these partners. To test the influence of phosphorylation on the DNA binding capability of IDD4, we produced stable transgenic plants expressing the phospho-modified IDD4: RFP versions controlled by the UBIQUITIN10 promoter. For further studies, transgenic lines were selected that exhibit comparable transcript levels of IDD4-modified-versions IDD4-AA and IDD4-DD thereby accumulating equal amounts of the encoded protein (S3D Fig). The phenotype of the IDD4 phospho-modified plants is indistinguishable from WT and the shoot fresh weight of 18 day-old IDD4-AA lines is slightly reduced compared to IDD4-DD lines but insignificant to WT. The root fresh weight of IDD4-AA and IDD4-DD lines is comparable to WT (S2G–S2I Fig). These lines were used for ChIP-PCR to determine binding preferences of phosphorylated IDD4. Quantitative PCR of DNA that was immunoprecipitated by RFP antibody confirmed that the phospho-mimicking version IDD4-DD shows higher binding affinity to the SAGT1 promoter compared to the IDD4-AA line or WT control under both flg22 and mock-treated conditions (Fig 6F). The inability of IDD4-AA to increase DNA binding to the SAGT1 promoter after flg22-treatment demonstrates the importance of the flg22-mediated post-translational modification of IDD4 at the phosphorylation sites. In order to analyze whether the binding ability enhancement of the unmodified IDD4 to the ID1 element also occurs in vivo in the context of PAMP signaling, ChIP of IDD4: GFP lines was performed 1 hour after flg22 treatment (Fig 6H). Previously, we tested whether flg22 perception influences IDD4 protein stability by analyzing IDD4ox lines for IDD4: GFP degradation at 30 and 60 min after flg22 treatment in 10 day-old seedlings. No change of stability was seen under these conditions by Western blotting or fluorescence microscopy in roots (S3E and S3F Fig) demonstrating that flg22-perception does not compromise IDD4 protein stability. However, the association of IDD4: GFP to the SAGT1 promoter in the ChIP approach was significantly elevated after flg22 application unlike the IDD4: GFP mock-treated control (Fig 6H). This finding suggests that PAMP-treatment determines the binding ability of IDD4 to the ID1 motifs of the SAGT1 promoter. These studies were extended by DNA-shift experiments using the recombinant N-terminal domain of IDD4 (IDD4-WT, IDD4-AA and IDD4-DD), containing the 2 phosphorylation sites and the 4 zinc-finger-containing ID domain previously shown to be necessary for DNA binding and interaction with other partners, respectively. As a probe, the sequence of the SAGT1 promoter region -259 to -224 (Fig 6A) was used that harbours 2 ID1 motifs and was shown to be targeted by IDD4: GFP in the ChIP-SEQ and ChIP-qPCR approaches (Fig 6A–6C). Predominant binding of IDD4-DD to the biotinylated ID1 motifs could be confirmed; whereas IDD4-AA and unmodified IDD4 exhibited much weaker binding preferences (Fig 6G). By adding an unlabeled specific competitor to the reaction, the interaction between IDD4-DD and the ID1 probe could be significantly diminished. By contrast, the addition of an unlabeled probe containing mutated ID1 motifs only resulted in a minor competition on IDD4-DD binding. Taking altogether, these results demonstrate the enhanced DNA-binding ability of the phospho-mimiking IDD4 version. After obtaining indications that the DNA-binding property of IDD4 might depend on the phosphorylation status in vivo and in vitro, we analyzed whether the phosphorylation status of IDD4 determines the transcriptional activity as well. To prove the biological relevance of the transcriptional activity of IDD4, the expression of SAGT1 was studied in the phospho-modified versions expressing lines under untreated conditions and after flg22 application (after 4 hrs). The relative gene induction level of SAGT1 after flg22 treatment was significantly induced in the IDD4-DD line, reminiscent to the IDD4ox line (Fig 6E), however, unlike to the IDD4-AA line that does not show an increased expression level when compared to WT (Fig 7A). These findings suggest that IDD4-DD acts as an activator of SAGT1 gene expression. By contrast, the expression of SAGT1 in the phospho-modified lines under untreated conditions was indistinguishable from WT, accordingly to the WT-like expression of SAGT1 in the IDD4ox line (Fig 6E). Altogether, these results imply the importance of the IDD4 phosphorylation status to initiate target gene expression in a flg22-dependent manner. After showing that phosphorylation of IDD4 coordinates gene expression, the biological relevance of this mechanism was assessed after hemi-biotrophic pathogen infection. Therefore, we measured Pst DC3000 hrcC- infection levels in two independent transgenic lines harbouring IDD4-AA and IDD4-DD. At 2 hrs after spray-infection of the different phospho-modified line, similar bacterial cfu were obtained, indicating that the accessibility of the bacteria to the leaf apoplast tissue was not affected. However, 72 hrs after plant infection, the proliferation levels of the bacteria were significantly elevated in both IDD4-DD lines compared to IDD4-AA lines and WT. Furthermore, the bacterial titers in the IDD4-AA lines were reduced even below WT levels indicating that the posttranslational-modification of IDD4 impacts bacterial growth and defense response (Fig 7B). In order to analyse the susceptibility of the modified IDD4 expressing lines at the molecular level, the transcriptomes of 14 day-old seedling were studied by RNA-Hiseq of 3 biological replicates. The transcriptome composition of IDD4-DD lines revealed 429 genes as being down-regulated and 354 up-regulated (p<0. 05) when compared to WT or IDD4-AA lines. GO terms of down-regulated genes could be determined for defense response, respiratory burst and innate immunity (Fig 7C, S6 Table). GO terms of the up-regulated genes in IDD4-DD highlighted gene functions in energy metabolism, RNA biosynthesis process and photosynthesis (Fig 7D). On the other hand, GO terms of down-regulated genes in IDD4-AA (p≤0. 05,211 genes) referred to response to starvation, root system development and metal ion transport (Fig 7E). Furthermore, GO terms for up-regulated genes in IDD4-AA (p≤0. 05,169 genes) emphasize gene functions for sugar and carbohydrate response (Fig 7F, S6 Table). In summary, the GO term analysis indicates a distinct molecular setting between IDD4-AA and IDD4-DD lines, e. g. favouring growth and energy metabolism at the expense of defense response in the case of the phospho-mimicking (DD) line. The transcriptome compositions of IDD4-AA and IDD4-DD lines with respect to significant differentially regulated genes (p<0. 05) resulted in 3 main clusters illustrating a distinct and partly opposite expression of DEGs in the particular genotypes (Fig 7G). Cluster I and II show a profound differential regulation of downstream targets in the IDD4-DD lines whereas their deregulation in IDD4-AA lines is rather moderate and oppositely to the IDD4-DD line, such as in cluster I. To assess the biological functions of DEGs in the respective clusters, we carried out a GO term analysis. Cluster I highlights among others GO terms for response to water deprivation (5. 4e-13) and abiotic stress stimulus (2. 1e-08), whereas genes in cluster II were dedicated to the GO terms plant-type cell wall organisation (6. 6e-10) and defense response (1. 7e-07). By contrast, cluster III emphasises genes that are down-regulated in the IDD4-AA lines and moderately up-regulated in the IDD4-DD lines supporting additionally the notion of a distinct regulation of subsets of target genes indicating a distinct transcriptome composition (Fig 7G). In this context, GO terms for cluster III could be assigned to gene function in response to starvation (4. 6e-08) and inorganic cation transport (3. 1e-07). The molecular characterisation by qPCR of the phospho-modified lines 4 hours after flg22 treatment showed differential regulation of the SA response marker PR1 being higher expressed in IDD4-AA lines than in IDD4-DD and WT (S4B Fig). By contrast, the expression of the SA-signaling inhibitors WRKY38 [72] and NIMIN1 [73] are elevated in the IDD4-DD lines and the expression of NIMIN1 is reduced in IDD4-AA compared to WT and IDD4-DD line (S4C and S4D Fig). These expression patterns of SA-signaling markers indicate a reduced SA response in the IDD4-DD lines. Therefore, to determine the concentration of free SA in the IDD4-phospho-modified lines after Pst DC3000 hrcC- infection, hormonal measurements were conducted. Quantitative analysis of free SA in 3 biological replicates of WT, IDD4-AA and IDD4-DD transgenic lines before pathogen infection (Fig 7H) revealed comparable amounts of about 0. 42 ng/mg dry weight. However, 24 hrs after being challenged by Pst DC3000 hrcC-, WT and IDD4-AA lines showed strongly increased comparable levels of free SA, whereas the IDD4-DD lines exhibited significantly lower amounts. These findings are compatible with the notion that the higher susceptibility of the IDD4-DD lines to Pst DC3000 hrcC- is to some extent caused by its diminished SA levels as reflected by the elevated expression of the negative SA regulators NIMIN1, WRKY38 and SAGT1. If IDD4 and in particular its phosphorylation status contributes to SA homeostasis, then the SA accumulation after pathogen treatment should as well as be compromised in the idd4 mutant. Therefore, eventually, we determined the free SA levels in the idd4 mutant 24 hrs after Pst hrcC- infection. Interestingly, we found already a moderately elevated SA level in the idd4 mutant (Fig 7I). Furthermore, the pathogen-induced SA rise was also higher than in infected WT plants. Taking into account the higher resistance of idd4 against hemi-biotrophic pathogens (Fig 1I and 1J) and the differential regulation of components contributing to SA metabolism/ catabolism such as SAGT1 (Fig 6D and S2F Fig), then these results corresponds with a function of IDD4 in the modulation of SA homeostasis. All in all, these findings suggest a profound biological significance of IDD4 phosphorylation to coordinating plant defense with growth and developmental processes. Plant response to pathogen attack often requires “trade-off” processes in which resources initially dedicated to growth and pattern formation will be redistributed to pay the high metabolic costs of defence. It follows that plants have to precisely regulate resources to be allocated to fight against a pathogen [74]. In this regard, it was shown that the members of the PHYTOCHROME-INTERACTING FACTORS (PIFs) play a pivotal role in the regulation of growth defense trade-offs to adapt to changing environmental conditions [75]. PIFs redundantly inhibit skotomorphogenesis and individually regulate other light-mediated processes such as shade avoidance responses, chloroplast differentiation and seed germination but their own transcriptional regulation is poorly understood [76]. In addition to the light responses, some PIF members including PIF4 are involved in the hormonal responses (gibberellic acid (GA), brassinosteroid (BR), and auxin) [53]. In this regard, PIF4-mediated thermosensory growth and architecture adaptations are directly linked to suppression of immunity at elevated temperature. Accordingly, the pif4-101 and pifQ (pif1, pif3, pif4, pif5) quadruple mutant exhibited increased resistance to PstDC3000, demonstrating PIF4 as a positive regulator of growth and development and negative regulator of immunity [77]. Interestingly, we found that IDD4 binds to the promoter region of several PIF members (PIL1, PIL2/PIF6, PIL5/PIF1, PIF3) and photoreceptor PHYTOCHROME B (PHYB) that regulates PIF protein levels through promoting light-dependent protein degradation [75] thereby supporting resistance against PstDC3000 [77]. The association of IDD4 to these promoter regions shows an important function in the transcriptional regulation of these system integrators in plant development. In addition, our ChIP-SEQ analysis revealed the association of IDD4 to the G-Box, also called PIF4-box (Fig 3B), which is highly enriched in PIF4 target promoters previously shown by PIF4 ChIP-SEQ analysis [78]. The PIF4 DNA binding ability depends on its capacity to form hetero- and homodimers. The shared usage of the G-Box cis-regulatory elements by PIF4 and IDD4 might be mediated by heterodimer formation of these factors to shape the DNA-binding ability and expression of particular gene subsets. We have shown that IDD4, in accordance with PIF4, acts as repressor of immunity against hemi-biotropic pathogens. However, the enhanced growth of the idd4 mutant demonstrates that IDD4 acts as a suppressor of plant growth and development, unlike PIF4 that positively regulates these processes. Consequently, we suspect that IDD4 and PIF4 cooperatively mediate the expression of defense-related genes that might antagonistically contribute to processes promoting growth and development. Moreover, IDD4 was previously reported being transcriptionally repressed by KAN and REV, and IDD4ox lines are compromised in leaf blade formation and leaf size [18]. The overexpression of IDD4 causes downward leaf curling, resembling reduced HD-ZIPIII gene function which indicates a negative feedback regulatory network of the three factors. The discovered binding of IDD4 to promoter regions of KAN and REV suggest an up to now uncharacterized feedback regulation. In addition, pathogen attack has been shown to suppress components of photosynthesis at the level of gene expression and protein abundance; and defense exerts a negative impact on photosynthesis which results in a reduction of components essential for light harvesting and carbon fixation [74]. We show that IDD4 affects the expression of genes involved in chloroplast maturation, localization in response to changing light conditions and photo-morphogenesis. Furthermore, the auto-immune phenotype of idd4, reflected by the internal increase of H2O2, indicates a role of IDD4 in the photo-respiratory H2O2 accumulation. We showed that IDD4 is phosphorylated by the immune MAPK MPK6 whose activation is triggered in planta upon PAMP-perception to modulate immune reaction [33]. We found that the IDD4 phospho-modified versions behave differently regarding the DNA-binding ability, target-gene activation and response to pathogen-attack. In this context, the IDD4-phospho-mimiking version (IDD4-DD) shows stronger recruitment to the DNA and acts as transcriptional activator of SAGT1 expression. By contrast, the phospho-dead version (IDD4-AA) displays weak DNA-binding and low transcriptional activation after flg22-treatment. By exploiting the native IDD4 version, we found that IDD4-binding and transcriptional activity are increased upon PAMP-perception. After flg22-application, the recruitment of IDD4 to the DNA and the transcriptional activity are elevated in accordance with the results of the phospho-mimicking IDD4-DD version. Moreover, the opposite behavior of IDD4-AA and its unresponsiveness to flg22-treatment support the notion of a post-translational modification-based mechanism to regulate IDD4 DNA-binding properties. Phosphosite-dependent transcriptional deactivation of IDD8 mediated by AKIN 10 in the process of carbon metabolism was recently reported [79]. The phosphorylation of IDD8 at Ser-178 and Ser-182, which reside both in the fourth ZF domain did not affect the subcellular localization and DNA-binding property of IDD8 but diminished the transcriptional activation activity. Interestingly, the function of a putative transactivation domain might be compromised by the phosphorylation of the closely adjacent ZF4 [45]. Recently, the importance of ZF4 for protein-protein interaction of IDD10 and IDD3 with the SCR-SHR complex was reported [17]. Furthermore, the transcriptional activity of IDD10 is modulated by reciprocal interactions with IDD3, SCR and SHR [80]. Therefore, it is conceivable that the phosphorylation of ZF4 in IDD8 interferes with the association of coactivators. In summary, the post-translational modification of particular ZFs in IDD4 and IDD8, as the discussed phosphorylation events, change their characteristic traits and can be considered as a general IDD/BIRD-regulation mechanism to modulate DNA-binding ability, protein-protein interaction and transcriptional activity. Recently, the gibberellin (GA) -inhibitory DELLAs were introduced to control plant immune responses by modulating the balance of JA and SA [81]. Notably, DELLAS cannot directly bind to the DNA because of the lack of a DNA interaction—domain. However, they act as co-activators or repressors, respectively, by binding to transcription factors in a stress-dependent manner in order to coordinate target gene expression. We showed the interaction of IDD4 with the DELLA/GAI. The IDD family members 2,3, 4,5, 9 and 10 serve as transcriptional scaffolds to enable transactivation activity of the DELLA/RGA of the GRAS-family [16] [22]. In vitro studies have confirmed the transcriptional activation of DNA-bound IDDs upon the association of RGA and the subsequent expression of the SCL3 locus [16]. Interestingly, genome-wide DNA-binding studies on DELLA/RGA revealed the enrichment of the ID1 cis-motif in about 28% of target sequences underpinned by a total p value of 2. 3e-10. The occurrence of the ID1 motif indicates a common set of RGA and IDD4 controlled genes [82]. In this context, we identified a substantial overlap of 20. 4% of RGA and IDD4 target genes and found associated GO terms, among others, for response to salicylic acid stimulus, light stimulus and regulation of defense response (S3 Table). Interestingly, the introduced quadruple-DELLA mutant is more resistant to Pst DC3000 compared to WT and accumulates higher levels of free SA after pathogen attack [81], demonstrating that DELLAs promote disease susceptibility to hemi-biotrophic pathogens and repress the SA-defense pathway. Therefore, the reduced susceptibility to bacterial infection in the association with the elevated SA levels of the idd4 mutant, and the reduced SA accumulation in the IDD4-DD lines after Pst HrcC- infection suggest a synergistic interaction of IDD4 and DELLA/RGA proteins in the regulation of selected defense processes. Moreover, GAI-ASSOCIATED FACTOR1/IDD2 is involved in the regulation of GA homeostasis and signaling in Arabidopsis for binding to genes which are part of GA feedback regulation. GA converts the IDD2 complex consisting of DELLA, IDD2 and TOPLESS RELATED from a transcriptional activator to a repressor upon the degradation of DELLA [22]. Similarly, the co-repressor SCL3 interacts competitively with DELLAs for the binding to IDD proteins to antagonistically regulate downstream gene expression to control GA signaling pathways [21]. To assess the distribution and abundance of the core ID1 motif (AGACAA), we performed a genome-wide in silico search of the 500 bp upstream regions in the promoter regions of IDD4-bound genes for the occurrence of at least two ID1-core motifs. Previous publications suggested the ID1-cis motif as the main binding sequence of different IDD family members, including IDD4 [22] [44] and it was shown that IDD4 binds to the ID1-core sequence inside the SCL3 promoter in vitro [16]. Noteworthy, our in vivo ChIP-SEQ data confirm the binding of IDD4 to the upstream regulatory sequence of SCL3 (S3B Fig, S3 Table). Our analysis revealed 2840 genes harbouring this element up to 8 times whereas 82. 9% of the genes contained two ID1-core motifs (S5 Table). This high number of genes with two ID1-core motifs in their promoter regions is surprising and raises the question of the specificity of IDD4 DNA-binding and how target gene activation can be achieved. We provide evidence that phosphorylation of IDD4 is a post-translational regulatory mechanism to trigger IDD4 DNA-binding. In order to coordinate the binding of IDD4 to the ID1 motif, binding partners might be involved in a developmental- and stress-dependent manner modulating the accessibility of the binding sites to IDD4. For example, IDD4 activity could be orchestrated by the formation of IDD4 homo- and heterodimers with other transcriptional regulators in a developmental or stress-dependent manner. Further mechanisms might include different splice variants as shown for IDD14, which are generated by cold stress to form a competitive inhibitor regulating starch metabolism. IDD14β lacks one functional DNA binding domain but is still able to create heterodimers with the functional IDD14 form (IDD14α). However, IDD14α–IDD14β heterodimers have diminished DNA-binding activity to their target promoter [83]. Slight modifications in the ratio of functional transcription factors and alternatively spliced variants could sustainably affect the expression of target genes. In this context, it is worth mentioning that IDD4 also forms a second splice variant (IDD4. 2) (TAIR-database), which contains the 4 zinc finger-containing ID domain, but excludes the phosphosite Ser-73. This means that IDD4 can presumably form heterodimers consisting of IDD4. 1-IDD4. 2 as well as homodimers with each of the two splice variants. It is well conceivable that the IDD4. 2 homodimer without the phosphorylation site could act as a competitive inhibitor of the phosphorylated IDD4. 1 protein and further research is necessary to pursue these possibilities. Experiments were performed by using Arabidopsis thaliana of the Columbia accession grown on soil in plant growth chambers (Percival Scientific) under short-day conditions (8h light/ 16 h dark) at 22°C. Nicotiana benthamiana were grown under long day conditions (16 h light + 8 h darkness) at 28 °C. idd4 (Salk_148352C) seeds were obtained from NASC. Total RNAs were extracted from 14 day-old seedlings, grown on sugar-free Murashige and Skoog (MS) plates under long-day conditions. We used the NucleoSpin RNA Plant (MACHEREY-NAGEL) kit, according to the manufacturer’s instructions. First strand cDNA was synthesised from 5μg of total RNAs using SuperScript First-Strand Synthesis System for RT-PCR according to the manufacturer’s instructions. The cDNA stock was diluted to a final concentration of 25ng/ul. Subsequently, 500nM of each primer was applied and mixed with LightCycler 480 Sybr Green I Master mix for quantitative PCR analysis, according to the manufacturer’s instructions. Products were amplified and fluorescent signals acquired with a LightCycler 480 detection system. The specificity of amplification products was determined by melting curves. GADPH was used as internal control for signals normalisation. Exor4 relative quantification software automatically calculates relative expression level of the selected genes with algorithms based on ΔΔCt method. Data were from duplicates of at least three biological replicates. All the sequences of primers used are given in S7 Table. Nuclear proteins were extracted from 14 day-old on half MS-grown seedlings. After quantification with the Bradford method, equal amounts of proteins were resolved by SDS-PAGE and then transferred to a polyvinylidene difluoride membrane using a Mini-Protean 3 Cell. Immunoblot analysis was performed using 1 μg/mL primary polyclonal antibodies raised against GFP and then with secondary antibodies conjugated to alkaline phosphatase. Antibody complexes were detected by chemiluminescence using the Immun-Start AP Substrate kit. ChIP assays were performed on 14 day-old in half MS-grown seedlings using Anti-GFP antibody—ChIP Grade and RNA polymerase II antibodies. Briefly, after plant material fixation in 1% (v/v) formaldehyde, the tissues were homogenised, and the nuclei were isolated and lysed. Cross-linked chromatin was sonicated using a water bath Bioruptor (Diagenode; 15-s on/15-s off pulses, 15 times). The complexes were immunoprecipitated with antibodies overnight at 4°C with gentle shaking and incubated for 1 h at 4°C with 50 μL of Dynabeads Protein A. Immunoprecipitated DNA was then recovered and analysed by qRT-PCR. An aliquot of untreated sonicated chromatin was processed in parallel and used as the total input DNA control. Sequencing of Col-0, IDD4: GFP sample material was performed in Illumina Hi-Seq2000 platform. About 47 million and 53 million paired-end reads (insert size of 300 bp) with 125 bp read length were obtained for Col-0 and IDD4 respectively. Quality statistics of reads were analyzed using FASTQC [http: //www. bioinformatics. babraham. ac. uk/projects/fastqc/]. Trimming and filtering of reads were performed using trimmomatic [84]. Parameters for read quality filtering were set as follows: Minimum length of 36 bp; Mean Phred quality score greater than 30; Leading and trailing bases removal with base quality below 3; Sliding window of 4: 15. Filtered reads were aligned to TAIR10 using Bowtie (Unique mapping of reads was adopted) and enriched regions were identified using MACS2 [85]. Parameters for peaks detection were set as follows: Number of duplicate reads at a location: 1; Bandwidth: 300; mfold of 5: 30; q-value cutoff: 0. 05. Peaks were identified against input DNA sequence. Results from all three peak callers (MACS2, SISSR, HOMER) have shown very high correlation. Bedtools was used for manipulation of these genomic peak intervals [86]. Identification of putative IDD4 binding motifs (p-value < 0. 05) at called peak positions was done using HOMER [87]. Coverage and enrichment of functional elements from TSS to TES and their flanking region were visualized using NGSplot [88]. The complete ChIP-SEQ data sets are available at the GEO repository (GEO accession GSE120068). Sequencing was performed on each library to generate 101-bp paired-end reads on Illumina HiSeq2500 Genome Analyzer platform. Read quality was checked using FastQC and low quality reads were trimmed using the Trimmomatic version 0. 32 (http: //www. usadellab. org/cms/? page=trimmomatic) with the following parameters: Minimum length of 36 bp; Mean Phred quality score higher than 30; Leading and trailing bases removal with base quality below 3; Sliding window of 4: 15. After pre-processing the Illumina reads, the transcript structures were reconstructed using a series of programs, namely, TopHat (ver. 2. 1. 1; http: //tophat. cbcb. umd. edu/) for aligning with the genome, and Cufflinks (ver. 2. 2. 1; http: //cufflinks. cbcb. umd. edu/) for gene structure predictions. For TopHat, the Reference-Arabidopsis thaliana (TAIR10) genome (https: //www. arabidopsis. org) was used as the reference sequences with a maximum number of mismatches as 2. To identify the differentially expressed genes, the following parameters were used: p-value of 0. 05 with a statistical correction using Benjamini Hochberg FDR of 0. 05 in cuffdiff. A cut-off of 2 fold up- or down-regulation has been chosen to define the differential expression. After processing the data, visualisation of differential expression was done using cummeRbund v2. 14. 0 (http: //bioconductor. org/packages/release/bioc/html/cummeRbund. html). Differentially regulated genes that are common among the samples were identified using Venny. The complete RNAseq data is available at the GEO repository (GEO accession GSE120068). For the GO term analysis, AGRIGO analysis tool was used (http: //bioinfo. cau. edu. cn/agriGO/, [30] by using significantly differentially expressed genes between the tested conditions. Protein-network analysis was performed by using STRING [46]. Putative MAPK docking site of IDD4 was searched using the ELM program (http: //elm. eu. org/). Whole seedling Arabidopsis cDNA library was used to amplify the coding sequence (cds) of the IDD4 and SCL3. Subsequently, the entry clone was generated by introducing the cds either in the pENTR- or pCR8/GW/TOPO-vector. Subsequently, entry clones were used to generate protein expression constructs (pDEST-MBP [89]) and protein localisation vectors, fused to GFP/RFP driven by the Ubiquitin promoter [90]. Site-direct mutagenesis of IDD4 was performed in a 2 step approach. First oligo-nucleotide based introducing of the base pair exchange results in two truncated cDNA fragments with an overlapping region. Secondly, both fragments were fused and amplified by using oligos that bind in the end and beginning of the total cDNA. GAI (amino acids 148 to 533), RGA, IDD4, IDD4-AA, IDD4-DD cDNA inserted into pDONR207 by Gateway cloning were recombined with pGADT7 (AD) or pGBKT7 (BD) to generate BD-GAI, AD-IDD4, AD-IDD4-AA and AD-IDD4-DD. All the sequences of primers used are given in S7 Table. Direct interaction assays in yeast were carried out following the small-scale LiAc yeast transformation procedure. The N-terminal part of the DELLA proteins is subject to autoactivation in yeast two-hybrid assays; therefore, only the C-terminal domain of GAI (amino acids 148 to 533) was recombined by Gateway with pGBKT7 to generate BD-GAI. Yeast strain AH109 was co-transformed with BD-GAI and AD-IDD4/IDD4-AA/IDD4-DD or empty vector (pGADT7), and interaction tests were surveyed on selective media lacking Leu, Trp, Ade and His. Plant materials were lyophilised and ground in a bead beater. Aliquots (about 5 mg dry weight) of powdered tissues were extracted with 400 μL of 10% methanol containing 1% acetic acid and internal standards (11. 1 ng of 2H4-SA). The samples were extracted in the bead beater for 1 min, placed in ice for 30 min, and then centrifuged at 13,000 g for 10 min at 4°C. The supernatant was carefully removed and the pellet re-extracted with 400 μL of 10% methanol containing 1% acetic acid. Following further 30 min incubation in ice, the extracts were centrifuged and the supernatants combined. The samples were filtered through 0. 22 μm PTFE filters before LC-MS/MS analysis. Analysis SA was performed by comparing retention times and mass transitions with the standards using an Agilent 1200 HPLC coupled to an Q-TRAP 5500 MS with an electrospray source. Chromatographic separation was carried out at 35°C on a Phenomenex Gemini C18 (150×2. 0 mm, 5 μm) column with the solvent system formic acid/acetonitrile/water (0. 1/94. 9/5, v/v/v; mobile phase A) and formic acid/ acetonitrile/water (0. 1/5/94. 9, v/v/v; mobile phase B). The gradient used was 0–20 min, 0%-100% A; 20–25 min, 100% A; 25–26 min, 100%-0% A; 26–36 min, 0% A. To reduce contamination of the MS, the first 5 min of the run was directed to waste using the inbuilt Valco valve. Analysis of SA was based on appropriate Multiple Reaction Monitoring (MRM) of ion pairs for labelled and endogenous SA using the following mass transitions: 2H4SA 141>97, SA 137>93. The MS was operated in negative ionization mode. The conditions were as follows: Temperature 500 °C, Ion source gas 1 50 psi, Ion source gas 2 60 psi, Ion Spray Voltage -4500 V, curtain gas 40 psi, Collision Gas Medium; DP (-25 V), EP (-9) and CXP (-2) were the same for all compounds. CE (-38), and DT (50) were used for 2H4SA and SA. Data were acquired and analysed using Analyst 1. 4 software. Plants were spray-inoculated with Pseudomonas syringae DC3000 and Pseudomonas syringae DC3000 hrcC- at OD600 = 0. 2 and sampled 2 h and 72 h after inoculation to determine the level of colonisation (colony-forming units (cfu) ) as described previously [91]. In three biological replicates, a total of 30 plants were sampled for each plant genotype by each taking 3 leaf discs per plant. Bradford assays were used to quantify protein levels in extracts and ensure equal loading of total proteins for gels used for immunoblot analysis and EMSA. Recombinant truncated protein of IDD4, IDD4-AA and IDD4-DD (each from amino acid 20–219) fused to MBP-tag was affinity-purified from E. coli Rosetta cells and enriched by Ion Exchange Purification. 3´ End-Biotinylated oligos were ordered. Biotinylated DNA (20fmol) was mixed with 1 μg of the indicated proteins after the instructions of the Lightshift Chemiluminescent DNA EMSA. These experiments were repeated at least three times with similar results. Purified recombinant proteins and constitutively active MAPKs were mixed together in kinase reaction buffer (20 mM Tris-HCl pH 7. 5,10 mM MgCl2,5 mM EGTA, 1 mM DTT and 50 μM ATP) and incubated at ambient temperature for 30 min. SDS-sample buffer was added to stop the reaction followed by boiling at 95°C for 10 min. Protein samples were resolved by SDS-PAGE. The gel was stained with SimplyBlue SafeStain and the band corresponding to the protein of interest was excised out, cut into small pieces of 0. 5 mm3 and destained with four successive washes of 15 min each with ACN and 100 mM NH4HCO3. Proteins were reduced with 10 mM Tris (2-carboxyethyl) phosphine (TCEP) in 100 mM NH4HCO3 at 37°C for 1 h followed by alkylation with 20 mM S-Methyl methanethiosulfonate (MMTS) at ambient temperature for 30 min. Proteins were then digested with trypsin (Porcine trypsin) at 37°C overnight. The digestion was stopped by the addition of 1% formic acid, and the peptides were recovered by incubating the gel pieces in ACN. The recovered peptide solution was desalted using C18 ZipTip and analysed by LC-MS/MS. Briefly, peptide samples were separated on a C18 connected to an LTQ-Orbitrap Velos or a Q-Exactive HF instrument. The LC gradient ramped from 5% solvent B (water/ACN/formic acid, 20/80/0. 1, v/v/v) to 45% solvent B over 45 min, then to 90% solvent B for 10 min. The MS instrument acquired fragmentation spectra on the top 10 peptides using CID fragmentation in the LTQ-Orbitrap or HCD in the Q-Exactive instrument. RAW data files obtained were converted to MGF files using Proteome Discoverer interface (version 1. 4). Database searches were performed with the Mascot server v2. 4 specifying the following parameters: database TAIR10 (release 2010/12/14,35386 sequences); enzymatic specificity: trypsin permitting two allowed missed cleavages; fixed modification of cysteine residues (Methylthio (C) ); possible variable modifications of phosphorylation on S, T and Y residues; 5 ppm tolerance on precursor masses and 0. 5 Da tolerance on fragment ions. The results were filtered based on Mascot scores and MD-scores. To obtain the expression vectors, coding sequences of candidate genes and MAPKs (kindly provided by J. Colcombet) were cloned in fusion with the N- and C-terminal parts of YFP, either as N- or C-terminal fusions, under the control of the cauliflower mosaic virus 35S (CaMV-35S) promoter in the pBIFC1,2, 3 and 4 vectors [92]. Appropriate positive and negative controls were carried out for all combinations. Recombined vectors were individually transformed in Agrobacterium tumefaciens C58C1 strain by electroporation. Agrobacterium cultures from glycerol stocks were inoculated in 10 ml of LB medium with appropriate antibiotics and incubated for 24 h at 28°C with agitation. Each culture was pelleted and resuspended in infiltration buffer (10 mM MgCl2,10 mM MES pH 5. 7,150 μM acetosyringone) to an OD600 of 1. 5 and kept in the dark for 3 h. The P19 viral suppressor of gene silencing was co-expressed with each combination to prevent silencing of transiently expressed proteins [93]. 500 μl of each bacterial culture was mixed before infiltration. For fluorescence complementation, all eight possible combinations between a candidate gene and a MAPK were agro-infiltrated into 3-week-old Nicotiana benthamiana leaves. After 3 days, an upright confocal microscope with a 20X objective (Plan-Apochromat, NA 1. 0) was used to visualise fluorescence. All images were acquired using Argon laser with 514-nm excitation. 3 weeks old plants were incubated in GUS-Staining solution (NaPO4 (pH 7. 2,50mM), EDTA (pH8. 0 10mM), TritonX100 (0. 1%), Ferrocyanid (2mM), Ferricynaid (2mM), X-Gluc (1mg/ml) ) for 12hr at 37°C. Subsequently, chlorophyll was bleached by applying 70% ETOH. Stainings were conducted by using 14-day old seedlings grown under sterile conditions on half MS medium, in accordance to [94]. ROS burst assay was performed as described by [95]. IDD4 (AT2G02080), SAGT1 (AT2G43820), MPK6 (AT2G43790).

This work illustrates the involvement of the IDD family member 4 in the regulation of defense responses against hemibiotrophic pathogens and in processes governing plant growth. IDD4 is embedded in widely-ramified regulatory pathways and exerts transcriptional control of key factors that shape and balance growth with defense. Mutation in IDD4 and overexpression affect the plant shoot and root growth. At the same time, IDD4 acts as a repressor of innate immunity against the hemi-biotrophic pathogen Pst DC3000. IDD4 interacts with and is in vitro phosphorylated by the immune MAP kinase MPK6. Genome-wide IDD4 DNA-binding studies revealed subsets of direct targets contributing to various developmental processes and immunity. Chromatin-immunoprecipitation and DNA-shift experiments identified the ID1 motif as the prime target site of IDD4 and revealed that phospho-modifed IDD4 shows altered DNA binding ability and thereby gene expression and pathogen resistance. Overall, these results indicate that IDD4 links signal transduction to gene expression of genes involved in basal immunity and growth and development.

Abstract Introduction Results Discussion Materials and methods

phosphorylation medicine and health sciences gene regulation immunology brassica dna transcription model organisms experimental organism systems genome analysis sequence motif analysis plants arabidopsis thaliana research and analysis methods sequence analysis genomics bioinformatics proteins animal studies gene expression immune response biochemistry eukaryota plant and algal models post-translational modification database and informatics methods genetics transcriptome analysis biology and life sciences computational biology organisms

2019

INDETERMINATE-DOMAIN 4 (IDD4) coordinates immune responses with plant-growth in Arabidopsis thaliana

The respiratory syncytial virus (RSV) fusion (F) glycoprotein is a major target of neutralizing antibodies arising from natural infection, and antibodies that specifically bind to the prefusion conformation of RSV F generally demonstrate the greatest neutralization potency. Prefusion-stabilized RSV F variants have been engineered as vaccine antigens, but crystal structures of these variants have revealed conformational differences in a key antigenic site located at the apex of the trimer, referred to as antigenic site Ø. Currently, it is unclear if flexibility in this region is an inherent property of prefusion RSV F or if it is related to inadequate stabilization of site Ø in the engineered variants. Therefore, we set out to investigate the conformational flexibility of antigenic site Ø, as well as the ability of the human immune system to recognize alternative conformations of this site, by determining crystal structures of prefusion RSV F bound to neutralizing human-derived antibodies AM22 and RSD5. Both antibodies bound with high affinity and were specific for the prefusion conformation of RSV F. Crystal structures of the complexes revealed that the antibodies recognized distinct conformations of antigenic site Ø, each diverging at a conserved proline residue located in the middle of an α-helix. These data suggest that antigenic site Ø exists as an ensemble of conformations, with individual antibodies recognizing discrete states. Collectively, these results have implications for the refolding of pneumovirus and paramyxovirus fusion proteins and should inform development of prefusion-stabilized RSV F vaccine candidates. Respiratory syncytial virus (RSV) is a ubiquitous pneumovirus which infects nearly all children in the U. S. by the age of two, with repeated infections occurring throughout life [1]. RSV is a common cause of acute lower respiratory tract infections in young children and the elderly, and in 2015 resulted in an estimated 94,000–149,000 deaths globally in children under the age of five [2]. Although few deaths of children in the United States are attributed to RSV [2,3], severe infections requiring hospitalization are frequent and lead to estimated direct health care costs of $750 million dollars annually [4]. Currently, there is no vaccine for RSV and the only FDA-approved therapy is passive prophylaxis with the monoclonal antibody palivizumab (Synagis) [5]. However, the high cost and modest efficacy of palivizumab restricts its usage to high-risk infants [6], making the development of improved interventions a global health priority. RSV is an enveloped virus of the Pneumoviridae family and it has a single-stranded, negative-sense RNA genome. There are two subtypes of RSV, A and B, to which many individual strains belong. RSV has two major glycoproteins on the viral surface important for entry: the fusion (F) and attachment (G) glycoproteins [7]. Whereas RSV G is the primary source of sequence variation and defines the subtype of a specific virus, the fusion glycoprotein is well conserved with sequence identities >90% [8]. RSV F is a class I fusion glycoprotein initially produced as an inactive precursor, F0, that is subsequently cleaved by furin-like proteases to generate a protomer of disulfide-linked subunits, F1 and F2 [9–12]. Three of these protomers associate to form the functional trimeric glycoprotein required for membrane fusion and infection [13–15]. Numerous vaccine trials for RSV are currently underway [16], many of which contain the RSV F glycoprotein as an antigen because it has been shown that F is a major target of neutralizing antibodies [17,18] and is the only protein on the viral surface that is strictly required for entry [19,20]. RSV F initially folds into a metastable prefusion conformation, with fusion peptides buried within the central cavity of the trimer [8]. During viral entry, RSV F triggers to undergo a dramatic conformational rearrangement from the prefusion to postfusion state. The triggering process results in release of the fusion peptides from the central cavity of the trimer and their insertion into the target-cell membrane, resulting in the formation of an unstable prehairpin intermediate. Collapse of this intermediate into the stable postfusion conformation brings the virus and host-cell membranes together, facilitating formation of a fusion pore and release of the viral genome into the target cell [7,15]. However, the mechanism and underlying cause of RSV F triggering is not well understood. Recombinant virus expressing only the RSV F protein on its surface is sufficient for infection of immortalized cell lines in vitro, suggesting that RSV F can facilitate attachment and mediate fusion in the absence of the attachment glycoprotein [7,20–22]. Potential RSV F receptors include nucleolin, EGFR, and heparan sulfate proteoglycans, among others [7,23–27], but the specific role each may play in the setting of natural infection remains to be defined. In addition, in vitro experiments have demonstrated that RSV F has a propensity to trigger upon exposure to elevated temperatures [28] and hypo-osmotic conditions [29], and RSV F has even been shown to spontaneously trigger and refold over time due to the metastable nature of the prefusion conformation [30]. This raises the possibility that RSV F does not have a specific receptor that initiates triggering and fusion, but rather that spontaneous triggering in the presence of attachment factors, such as heparan sulfate proteoglycans [31], is sufficient for entry. The majority of RSV-neutralizing activity in human sera is due to antibodies specific for the prefusion conformation of F [17,18], and recent characterizations of the human antibody response to RSV F has revealed that prefusion-specific antigenic sites, including site Ø (“zero”), are the major target of neutralizing antibodies [18,32,33]. Antigenic site Ø is located at the membrane-distal apex of the trimer and includes the α4-helix and the loop connecting α4 to α5 (α4–α5 loop) of F1, and the F2 loop between β2 and α1. Upon triggering, site Ø undergoes an extensive structural rearrangement in which α4 and the α4–α5 loop refold to form the continuous α5-helix observed in the postfusion F conformation [8]. Comparison of the neutralization potency of two site Ø antibodies, D25 [34] and 5C4 [35], with palivizumab, a site II-directed conformation-independent antibody [36], demonstrated that the prefusion-specific antibodies are 10–100 times more potent [8]. Other potent prefusion-specific human antibodies that bind to the apex of the trimer, such as AM22 and RSD5, have also been isolated in recent years [8,37,38], and one of them (MEDI8897) is now in advanced stages of clinical development [39]. D25 was the first structurally characterized antibody that specifically targets prefusion RSV F and was used to solve the structure of the prefusion RSV F conformation, facilitating the engineering of prefusion-stabilized variants that prevent conformational rearrangement to postfusion RSV F [8,40,41]. Recently, the structure of 5C4 bound to RSV F was determined, revealing a nearly identical conformation of prefusion RSV F as that observed in the D25-bound structure [42]. However, crystal structures of the different prefusion-stabilized variants of RSV F have revealed an alternative conformation of antigenic site Ø or weak electron density in this region, suggesting that this site is flexible. Currently, it is unclear if flexibility in this region is an inherent property of prefusion RSV F that may be important for triggering membrane fusion [41], or if it is related to inadequate stabilization of site Ø in the engineered variants [40]. Therefore, we sought to investigate the conformational plasticity of site Ø by determining and comparing the crystal structures of prefusion RSV F in complex with AM22 and RSD5. Our results demonstrate that prefusion RSV F adopts at least three alternative conformations of site Ø and that potently neutralizing human antibodies can recognize the alternative conformations using distinct binding modes. This suggests that site Ø samples an ensemble of conformations in vivo, at least some of which can be recognized by neutralizing human antibodies. These results should influence future vaccine designs and may have implications for the mechanism of RSV F triggering. Previous studies have demonstrated that AM22 and RSD5 potently neutralize RSV and preferentially bind to the prefusion RSV F conformation, similar to the previously characterized site Ø antibodies D25 and 5C4 [8,35,38,42]. However, differences in antibody kinetics and subtype specificities have not been fully explored. Therefore, we used surface plasmon resonance (SPR) to determine the binding affinity and kinetics of the interaction between three site Ø antibodies (AM22, D25, and RSD5) and prefusion RSV F derived from each subtype (strains A2 and B9320) (Fig 1). For these studies, we worked with a germline-reverted version of RSD5 (RSD5-GL), which had 20 somatic mutations in the framework regions reverted to germline residues to minimize immunogenicity (S1 Fig). Of note, RSD5-GL showed similar neutralization potency and binding kinetics for prefusion RSV F as compared to the parental RSD5 antibody (RSD5-WT) (S2 Fig). Despite similar neutralization potencies, AM22, D25, and RSD5 displayed distinct affinities and binding kinetics when compared to each other as well as when compared individually across the two RSV subtypes. The AM22 antigen-binding fragment (Fab) has an ~50-fold higher affinity for subtype A with an equilibrium dissociation constant (KD) of 0. 12 nM, whereas its affinity for subtype B is 6. 1 nM. Similarly, D25 Fab binds tightly to prefusion RSV F with a slight preference for subtype A, consistent with previously published data [43], having a KD of <0. 064 nM and 0. 33 nM for subtype A and B, respectively. In contrast, RSD5-GL Fab has substantial subtype specificity with a >2,000-fold stronger affinity for subtype B compared to subtype A, with a KD of <0. 016 nM and 34 nM, respectively. Preferential binding of RSD5-GL for subtype B is primarily due to the off-rate, which is fast for the subtype A interaction and slow for the subtype B interaction. To investigate the conformation of site Ø and define the epitope on prefusion RSV F recognized by AM22, we determined the crystal structure of the AM22 Fab alone and in complex with the prefusion-stabilized RSV F variant DS-Cav1 [40]. Crystals of the AM22 Fab alone diffracted X-rays to 1. 7 Å and crystals of the prefusion RSV F–AM22 complex diffracted X-rays to 3. 5 Å (Table 1). The AM22 variable domain (Fv) superimposes very well between the bound and unbound crystal structures, with high structural similarity across the framework and complementarity-determining regions (CDRs) resulting in an r. m. s. d. of ~0. 31 Å for 214 Cα atoms. The crystal structure of the F–AM22 complex shows that three AM22 Fabs bind to prefusion RSV F at the membrane-distal apex of the trimer and have a vertical angle of approach (Fig 2A), in agreement with previously published negative stain EM images [8]. AM22 buries 729 Å2 of surface area on each protomer of prefusion RSV F, mediated primarily through interactions between the heavy chain and F1 subunit. The AM22 heavy chain buries 554 Å2 (76%) on the surface of prefusion RSV F and is involved in 15 hydrogen bonds with RSV F, 14 of which are formed between the CDR H3 and seven residues within Gln202–Ser215 of α4 and the α4–α5 loop of prefusion RSV F. The light chain is responsible for the remaining 175 Å2 (24%) of buried surface area on prefusion RSV F and forms three additional hydrogen bonds with the α4–α5 loop via the CDR L2. The high affinity and specificity of AM22 for prefusion RSV F (Fig 1 and S3 Fig) is due to the formation of a three-strand anti-parallel β-sheet between the CDR H3 of AM22 and α4–α5 loop of F1 (Fig 2B). When bound by AM22, the α4-helix kinks near residue Pro205 and shifts away from α5, stretching the α4–α5 loop and allowing it to adopt a β-strand conformation that pairs with the β-hairpin formed by the CDR H3 of AM22. Upon RSV F triggering and the rearrangement into the postfusion conformation, α4 and the α4–α5 loop refold into the continuous α5-helix, which would disrupt the β-sheet interaction and prevent AM22 binding. Sequence comparison of the two RSV F subtypes demonstrates that the residues comprising the AM22 epitope are well-conserved. However, one of the subtype A RSV F residues that contacts AM22 is Lys209, which is a Gln in subtype B (Fig 2C). The Lys209 side chain of subtype A prefusion RSV F is coordinated by three residues of the AM22 heavy chain. This includes the formation of a salt bridge with Asp100G of the CDR H3 that effectively extends the prominent β-sheet interaction. Substitution of Lys209 with Gln, as found in subtype B, would eliminate the salt bridge and may explain the subtype A preference of AM22. Indeed, incorporating the K209Q substitution in subtype A prefusion RSV F results in a decreased affinity of AM22 with a KD of 6. 7 nM (S4 Fig), which closely matches the KD of 6. 1 nM for subtype B. To further investigate the conformational variability of site Ø and to identify the epitope on prefusion RSV F recognized by RSD5-GL, we determined the crystal structure of prefusion RSV F in complex with the RSD5-GL Fab to 3. 5 Å resolution (Table 1). The crystal structure shows that three RSD5-GL Fabs bind to the membrane-distal apex of the prefusion RSV F trimer (Fig 3A). RSD5-GL binds slightly lower on the trimer than AM22, bridging antigenic site Ø and the recently defined site V [32,44]. The interactions of RSD5-GL with prefusion RSV F are more diverse than that of AM22, with the CDR H2 and CDR H3 as well as all three CDRs of the light chain making contacts with the F protein (Fig 3B). In addition, the contacts on prefusion RSV F span multiple regions including the F2 loop as well as α3 and α4 of F1, burying a total surface area of 855 Å2. The RSD5-GL heavy chain buries 577 Å2 (67%) on the surface of prefusion RSV F. The CDR H3 interacts with α4 and the F2 loop, forming hydrogen bonds with Asp200 and main-chain atoms of Lys65, respectively, whereas the CDR H2 contacts α3 and forms a salt bridge with Lys168. The light chain contributes to the interface through contacts with α4, including hydrogen bonds with Asp200 and main-chain atoms of Pro205 and Gln210. Similar to AM22, the prefusion specificity of RSD5-GL can be explained by the dramatic rearrangement of α3 and α4 to form the single elongated α5-helix upon conversion to the postfusion conformation, which dismantles the RSD5-GL epitope. The binding mode and angle of approach differs for each of the three site Ø antibodies (Fig 4). AM22 and D25 adopt a vertical angle of approach and bind at the apex of the trimer, in agreement with previous negative stain EM images [8,42]. In contrast, RSD5-GL has a more diagonal angle of approach and binds slightly lower on prefusion RSV F, bridging antigenic sites Ø and V, similar to the recently characterized antibody 5C4 [42]. Despite these differences, the epitopes of all three antibodies overlap significantly and large steric clashes would prevent any two of these antibodies from binding simultaneously (Fig 4A). Specifically, all three antibodies make multiple competing contacts with α4 and the α4–α5 loop. In addition, the extent to which the antibodies interact with the F2 subunit varies greatly between the three antibodies. The RSD5-GL interface with F2 accounts for 21% of the buried surface area on prefusion RSV F and includes two hydrogen bonds with Lys65. The D25 interface with F2 contributes 23% of the buried surface area and includes five hydrogen bonds to four residues within Asn63–Lys68. In contrast, the interface between AM22 and F2 accounts for only 9% of the buried surface area on prefusion RSV F, and AM22 forms no hydrogen bonds or salt bridges with F2. Thus, whereas RSD5-GL and D25 make several contacts with the F2 loop, AM22 interacts almost exclusively with the F1 subunit. Alignment of the three antibody-bound prefusion RSV F structures revealed three alternative conformations of site Ø (Fig 5). The structure of DS-Cav1 in complex with AM22 shows a more open conformation of this site, with the α4-helix kinked out and away from α5, resulting in a stretched α4–α5 loop and a larger angle between the α4 and α5 helices. This structure closely matches the unbound prefusion-stabilized RSV F variants DS-Cav1 (PDB ID: 4MMU) and PR-DM (PDB ID: 5C69 [41]), specifically the kink in the α4-helix and greater angle between α4 and α5. In contrast, the prefusion RSV F–D25 structure has a more closed conformation of site Ø, where α4 does not kink out and there is a smaller angle between α4 and α5. The structure of DS-Cav1 bound to RSD5-GL reveals an intermediate site Ø conformation, with the α4-helix only slightly kinked out and away from α5, but not to the same degree as seen in the AM22 complex. Because all three of these antibodies were isolated from humans who had experienced natural RSV infection, these structures indicate that site Ø is naturally flexible and adopts at least three states that can be recognized by the human immune system. It is also possible, and perhaps more likely, that site Ø exists as an ensemble of many conformations, three of which were trapped by these antibodies. Analysis of published crystal structures of prefusion RSV F in complex with neutralizing antibodies targeting various antigenic sites further supports an ensemble of site Ø conformations (S5 Fig). AM22 and RSD5 are two human antibodies that bind to antigenic site Ø and are specific for the prefusion conformation of RSV F. Although both antibodies potently neutralize RSV, similar to D25, their binding kinetics and subtype specificity have distinct differences. AM22 and D25 both preferentially bind to subtype A, whereas RSD5 preferentially binds to subtype B as a result of its fast dissociation rate constant for subtype A F proteins. In addition, previous publications using SPR and flow cytometry-based competition assays have suggested that AM22 and RSD5 do not compete with D25, but rather occupy a separate prefusion-specific epitope [37,38]. However, comparing the crystal structures of prefusion RSV F bound to AM22, RSD5-GL, and D25, demonstrates that all three antibody epitopes overlap substantially and would prevent any two from binding simultaneously due to large steric clashes. This emphasizes the importance of structural characterization of antibody epitopes in addition to competition data, as varying antibody kinetics can mislead epitope classification when using only competition assays. The structural studies presented here reveal alternative conformations of RSV F site Ø in the prefusion state and suggest a natural flexibility of the region that can be recognized in numerous ways by the humoral immune system. This conformational flexibility is further supported by crystal structures of multiple prefusion-stabilized variants, which have identified an array of site Ø conformations, very high atomic B factors, or a distinct lack of site Ø electron density (PDB IDs: 4JHW, 4MMQ–4MMV, 4ZYP, 5C69,5C6B, 5EA3–5EA8,5KWW, 5K6B, 5K6C, 5K6F–5K6I, 5TOJ, and 5TOK [8,40,41,45–49]). Conformational flexibility and transient exposure of different epitopes have been noted for other class I fusion proteins such as HIV-1 Env and MERS-CoV Spike [50–54]. Specifically, identical residues of the V1/V2 loops of HIV-1 Env have been shown to adopt different conformations when bound by two different neutralizing antibodies, demonstrating structural plasticity of an important neutralizing epitope [55,56]. Comparison of two prefusion-stabilized RSV F variants, DS (PDB ID: 4MMQ) and Cav1 (PDB ID: 4MMS), highlights the flexibility of site Ø and suggests that a conformational rearrangement of site Ø is required prior to release of the fusion peptide from the central cavity of the trimer during refolding from the prefusion to postfusion state [40]. When only the fusion peptide was anchored by a disulfide bond in the DS structure (PDB ID: 4MMQ), site Ø was conformationally heterogeneous as indicated by the lack of electron density. However, site Ø cavity-filling mutations introduced in the Cav1 (PDB ID: 4MMS) or DS-Cav1 (PDB ID: 4MMU) variants stabilized the prefusion RSV F structure and showed clear electron density extending from the fusion peptide through site Ø. Taken together, this suggests that movement of the fusion peptide is conditional upon prior rearrangement of site Ø. The only discrepancy between the crystal structures of wild-type prefusion RSV F when bound to D25 and the prefusion-stabilized variant DS-Cav1 is the kinked-open conformation of α4 in the DS-Cav1 model, which was attributed to the cavity-filling V207L mutation being unable to fully stabilize the α4-helix [40]. However, the structure reported here of DS-Cav1 in complex with the human antibody AM22 matches the unbound DS-Cav1 structure and demonstrates that this conformation is a naturally sampled alternative conformation of prefusion RSV F. Recent vaccine strategies targeting viruses with class I fusion proteins have focused on stabilizing the prefusion conformation of the fusion protein for use as an immunogen. A common approach to achieve prefusion stabilization has been through introduction of one or more proline residues within the loop of a helix-loop-helix motif that refolds into a continuous alpha helix in the postfusion conformation. This strategy has been used successfully to stabilize several class I viral fusion glycoproteins including RSV F, MPV F, influenza HA, HIV Env, and coronavirus Spike [41,57–60]. Proline-based stabilization utilizes the restricted phi-psi angles of proline residues and disruption of the alpha-helix hydrogen bond network to inhibit the hinge motion required of the helix-loop-helix motif and the subsequent coil-to-helix structural transition required for refolding into the postfusion conformation. However, for wild-type prefusion RSV F there is a conserved proline residue (Pro205) within the middle of the α4-helix, N-terminal to the α4–α5 loop that may act as a hinge during refolding [41]. Crystal structures of prefusion RSV F demonstrate a variety of alternate conformations adopted by antigenic site Ø, all of which diverge near the conserved Pro205. This indicates that Pro205 may disfavor a rigid alpha-helical structure, which in turn facilitates conformational flexibility of site Ø and the tendency of prefusion RSV F to trigger. We note that Pro205 is absolutely conserved in all human and bovine RSV strains, and a proline at an identical position is also commonly found in F proteins from viruses within the Pneumoviridae and Paramyxoviridae families, with the exception of human metapneumoviruses (Fig 5D). The functional importance of this well-conserved proline residue will need to be evaluated in future studies investigating the triggering mechanism and refolding of pneumovirus and paramyxovirus F proteins. Recent characterization of the antibody repertoire against RSV F highlights the importance of prefusion-specific epitopes when selecting immunogens for RSV F vaccine design, particularly after the recent failure of several postfusion RSV F vaccine trials [16,18,32,33,61,62]. Our findings demonstrate that potently neutralizing human antibodies that target antigenic site Ø can recognize a variety of alternative conformations and have varying degrees of subtype specificity. Because potent antibodies can bind to the various alternative conformations of site Ø, we propose that the ideal prefusion RSV F immunogen would maintain this natural conformational flexibility of site Ø as well as present neutralizing epitopes common to both RSV subtypes. The RSV viruses used in this study were A/A2/61 (ACO83301), A/Randall/58, A/Long/56 (ACO83302), B/WV/14617/85 (ATCC VR-1400), A/9320/77 (AAR14266), A/9846/09 (JX171066), A/9835/09 (JX171067), A/9278/07 (KC618407), A/9395/07 (KC618409), B/9736/09 (JX171068), B/9847/09 (JX171073). Plasmid encoding prefusion-stabilized RSV F (DS-Cav1), subtype A DS-Cav1 with a K209Q substitution, or postfusion RSV F based on subtype A (strain A2) or subtype B (strain B9320) with a C-terminal 6x- or 8x-histidine tag and Strep-tag II was co-transfected with furin into FreeStyle 293-F cells (Invitrogen) at a 4: 1 ratio to ensure full cleavage of prefusion RSV F. Proteins were purified from the media after six days using Ni-NTA Superflow resin (Qiagen) and Strep-Tactin resin (IBA). Tags were removed by digestion with thrombin protease, followed by gel filtration using a Superdex 200 16–600 column (GE Healthcare Biosciences). Prefusion RSV F protein used for SPR was produced in the same manner, except the tags were not removed prior to gel filtration. For crystallization, DS-Cav1 from strain A2 was expressed in the presence of kifunensine (5 μM), digested with 10% (w/w) Endo H overnight, mixed with a 2-fold or 1. 5-fold molar excess of purified Fab for the AM22–RSV F and RSD5-GL–RSV F complexes, respectively, and the resulting complexes were purified by size exclusion chromatography (SEC) using the Superose 6 XK 16–70 column (GE Healthcare Biosciences) in a buffer consisting of 2 mM Tris pH 8. 0,200 mM NaCl, and 0. 02% NaN3. Germline sequences of RSD5 framework regions were determined with reference to the IMGT database [63]. RSD5 and RSD5-GL (fully germlined in VH and VL framework regions, as defined by IMGT) were produced by gene synthesis (GenScript) and confirmed by sequencing. Synthesized VH and VL sequences were cloned into human Igγ1 and Igκ expression vectors (kindly provided by Michel Nussenzweig, Rockefeller University, New York, NY, USA), essentially as described [64]. Plasmids encoding antibody heavy and light chains for AM22, RSD5-WT, RSD5-GL, or for D25 were co-transfected into Expi293 cells or FreeStyle 293-F cells (Invitrogen). AM22 and D25 IgGs and Fabs were purified using Protein A agarose (Fisher) or CaptureSelect IgG-CH1 affinity matrix (Life Technologies), respectively. All IgG antibodies were eluted off the Protein A column using 0. 1 M glycine pH 3. 0 into a buffered solution containing 1/10 (v/v) of 1 M Tris pH 8. 0. All Fabs were eluted off the CaptureSelect IgG-CH1 column using 50 mM acetic acid pH 4. 0 into a buffered solution containing 1/10 (v/v) of 1 M Tris pH 8. 0. To produce RSD5-WT Fab, RSD5-WT was expressed and purified as an IgG with an HRV 3C protease site in the hinge of the heavy chain. RSD5-WT Fab was produced by digesting the IgG with HRV 3C for 2 hours at room temperature, followed by passing the solution over protein A resin to remove the Fc, and subsequently purified by SEC using a Superdex 200 column (GE). Production of RSD5-GL Fab was done by incubating RSD5-GL IgG with papain beads (Pierce). All IgGs and Fabs were buffer exchanged using a desalting column, followed by final purification by SEC using a Superdex 200 column (GE) prior to long term storage at -80 °C. DS-Cav1 or postfusion RSV F from subtype A (strain A2) or subtype B (strain B9320), as well as a subtype A mutant with a K209Q substitution, with a C-terminal 6x-His or 8x-His tag was immobilized on a Ni-NTA sensor chip to a total of 80–150 response units using a Biacore X100 (GE). A buffer-only sample was injected over the DS-Cav1 or postfusion RSV F and reference flow cells for reference subtraction, followed by serial 3-fold dilutions of Fab (AM22, RSD5-GL, RSD5-WT, or D25) from 300 nM to 46. 5 pM in HBS-P+, with a duplication of the 11. 1 nM concentration. For the DS-Cav1 subtype A mutant with a K209Q substitution, only AM22 Fab was evaluated. For the assay evaluating AM22 Fab binding to strain B9320, the highest AM22 Fab concentration used was 1 uM in HBS-P+ buffer, followed by serial 3-fold dilutions to the lowest concentration of 152 pM. For the assay evaluating binding to postfusion RSV F of subtype A or B, a concentration of 300 nM of each Fab in HBS-P+ buffer was used. The data were double-reference subtracted and fit to a 1: 1 binding model using the Biacore X100 or Scrubber2 analysis software. Final binding curves were displayed using GraphPad Prism Version 7. 03 for Windows. Microneutralization assay based on infection of Hep-2 cells by RSV strains. RSD5 human IgG1 monoclonal antibody variants were incubated with 50–100 TCID50 of viruses for 1 hour at room temperature before addition of Hep-2 target cells which were incubated for 4 or 5 days (depending on the strain). Viral infection was measured by indirect immunofluorescence using an automated Pathway 855 analyzer (BD) as previously described [38]. IC50 values were calculated by interpolation of neutralization curves fitted with a 4-parameter nonlinear regression with a variable slope. Crystals for the AM22 Fab alone were produced by sitting-drop vapor diffusion using the Morpheus HT-96 crystallization screen (Molecular Dimensions). AM22 Fab (18. 0 mg/mL in 100 mM NaCl, 1 mM tris pH 8. 0,0. 01% NaN3) was mixed at a 1: 2 ratio with the D2 reservoir condition (0. 02 M 1,6-hexanediol; 0. 02 M 1-butanol; 0. 02 M 1,2-propanediol (racemic); 0. 02 M 2-propanol; 0. 02 M 1,4-butanediol; 0. 02 M 1,3-propanediol; 0. 1 M MES/imidazole pH 6. 5; 10% (w/v) PEG 8,000; 20% ethylene glycol). Crystals were looped directly from the crystallization drop and frozen in liquid nitrogen. Data were collected at the X6A beamline (National Synchroton Light Source, Brookhaven National Laboratories) and scaled to 1. 70 Å. The best diffracting crystals of the complex of Endo H-treated DS-Cav1 with AM22 Fab were produced using the Hampton HT Additive Screen via sitting-drop vapor diffusion. We mixed 100 nL of DS-Cav1–AM22 (5. 7 mg/mL in 200 mM NaCl, 2 mM Tris pH 8. 0,0. 02% NaN3) with 200 nL of reservoir solution containing 0. 1 M sodium acetate pH 5. 5,32. 15% (v/v) PEG 400,4. 02% (w/v) PEG3350, and 0. 01 M cadmium chloride. Crystals were looped directly from the crystallization drop and frozen in liquid nitrogen. Data were collected at the X6A beamline (National Synchroton Light Source, Brookhaven National Laboratories) and scaled to 3. 50 Å. Crystals for the complex of Endo H-treated DS-Cav1 with RSD5-GL Fab were initially identified in position H3 of the ProPlex HT-96 crystallization screen (Molecular Dimensions) via sitting-drop vapor diffusion. The best diffracting crystal was grown in a solution of 1. 8 M lithium sulfate and 0. 1 M Tris at pH 8. 0 via hanging-drop vapor diffusion at a protein-to-reservoir ratio of 1: 3 by mixing 0. 5 μL of DS-Cav1–RSD5-GL (5. 9 mg/mL) with 1. 5 μL of reservoir solution. The crystal was looped directly from the crystallization drop and flash frozen in liquid nitrogen. X-ray diffraction data for this complex were collected at the 19-ID beamline (Advanced Photon Source, Argonne National Laboratories) and scaled to 3. 50 Å. Diffraction data were processed using the CCP4 software suite [65]: data were indexed and integrated in iMOSFLM [66] and scaled and merged with AIMLESS [67]. A molecular replacement solution for the 1. 70 Å AM22 Fab dataset was found by PHASER [68] using a chimeric protein model consisting of the heavy and light chains of PDB ID: 3LMJ and PDB ID: 3QEG, respectively, separated into the constant and variable domains as search models. The structure was built manually in Coot [69] and refined using PHENIX [70]. The structure was built and refined to an Rwork/Rfree of 18. 0%/20. 5% (Table 1). A molecular replacement solution for the 3. 50 Å complex of DS-Cav1 with AM22 Fab was obtained using PHASER with prefusion-stabilized RSV F variant Cav1 (PDB ID: 4MMS) and the 1. 70 Å AM22 Fab structures as search models. The asymmetric unit contained the prefusion trimer bound by three AM22 Fabs. Rigid-body refinement was then performed in PHENIX, followed by refining group B-factors and (x, y, z) coordinates in PHENIX with NCS torsion restraints and reference-model restraints. The reference model was the 2. 40 Å prefusion-stabilized RSV F variant Cav1 (PDB ID: 4MMS). The structure was built and refined to an Rwork/Rfree of 21. 7%/28. 0% (Table 1). A molecular replacement solution for the 3. 50 Å complex of DS-Cav1 with RSD5-GL Fab was obtained using PHASER with prefusion-stabilized RSV F variant PR-DM (PDB ID: 5C69) and a chimeric protein Fab model consisting of the heavy and light chains of PDB ID: 1DFB and PDB ID: 1MCO, respectively, separated into the constant and variable domains and without the Fc domain of the 1MCO heavy chain. The structure was built manually in Coot [69] and refined using PHENIX [70]. Rigid-body refinement was initially performed in PHENIX, followed by refining individual B-factors and (x, y, z) coordinates in PHENIX with reference-model restraints. The reference model was the 2. 3 Å prefusion F variant PR-DM (PDB ID: 5C69). The structure was built and refined to an Rwork/Rfree of 18. 3%/20. 5% (Table 1). Structural features were analyzed using the “Interfaces” feature of PDBePISA [71]. This analysis defined the antibody epitope and paratope, specific residues and contacts involved in the interface, as well as the buried surface area. The modeled structure of each complex was displayed using PyMOL [72] to facilitate structural comparison between the different complexes. The amino acid sequence for the human respiratory syncytial virus subtype A (strain A2) fusion protein was used as the original sequence for comparison with all known pneumovirus and paramyxovirus fusion proteins. NCBI basic local alignment search tool (BLAST) was used to identify homologous regions between the hRSV fusion protein sequence (strain A2) and other pneumovirus and paramyxovirus fusion proteins. Specifically, we identified sequences that were indicated to be partially homologous with the residues 195–214 of the hRSV fusion protein sequence derived from strain A2. These residues correspond to the α4-helix and α4–α5 loop within the prefusion RSV F structure. For all known structures of prefusion pneumovirus or paramyxovirus fusion proteins, the homologous sequence also corresponds to the equivalent α4-helix and α4–α5 loop, even if the residue numbering differs. To prevent overrepresentation from viral species which have multiple subtypes sequenced, we only included a single amino acid sequence from each species when the multiple subtypes were >90% identical throughout the residue range corresponding to 195–214 in hRSV F. However, if two subtypes within a single viral species differed by >10% in the residue range corresponding to 195–214 in hRSV F, then they were both included when performing sequence analysis and generating the sequence WebLogo. For example, there are four sequenced strains of human metapneumovirus F, but they are all mostly identical and hence only two representative sequences were included in the final WebLogo (strain A1 and B1). However, there are multiple distinct sequences for the different types of parainfluenza virus (PIV), and therefore all the distinct sequences are included separately when generating the WebLogo. See S1 Table for a full list of sequences used in the WebLogo. The WebLogo was generated using publicly available software at weblogo. berkley. edu. The coordinates and structure factors for the F–RSD5-GL complex, the unbound AM22 Fab, and the F–AM22 complex, have been deposited in the Protein Data Bank (PDB) under accession codes 6DC3,6DC4, and 6DC5, respectively.

Respiratory syncytial virus (RSV) is a pervasive pathogen that causes severe lower respiratory tract infections, leading to ~100,000 deaths globally each year and hundreds of millions of dollars in healthcare costs. However, despite its prevalence, there is no vaccine for RSV and the only available therapy is limited to high-risk infants, leaving the vast majority of people with no effective means of prevention or treatment. The RSV fusion (F) protein is a major target of neutralizing antibodies, including extremely potent antibodies that recognize prefusion-specific epitopes. In this study, we determined the crystal structures of two neutralizing human antibodies bound to prefusion RSV F. Our results reveal that each antibody recognizes a different conformation of a neutralization-sensitive epitope, suggesting that this region is inherently flexible and may be important for RSV F function. These results should guide future vaccine-design efforts and help elucidate the mechanisms of RSV F triggering and fusion.

Abstract Introduction Results Discussion Methods

chemical bonding medicine and health sciences immune physiology crystal structure chemical compounds immunology condensed matter physics microbiology viral structure organic compounds crystals materials science amino acids crystallography antibodies glycoproteins hydrogen bonding physical chemistry immune system proteins solid state physics proteins antigens chemistry cyclic amino acids proline physics biochemistry organic chemistry virology physiology virus glycoproteins biology and life sciences physical sciences materials glycobiology

2019

Alternative conformations of a major antigenic site on RSV F

Eosinophilia is a typical finding of the acute/juvenile form of paracoccidioidomycosis (PCM), a systemic mycosis endemic in Latin America. This clinical form is characterized by depressed cellular immune response and production of Th2 cytokines. Moreover, it has been shown that the increased number of eosinophils in peripheral blood of patients returns to normal values after antifungal treatment. However, the role of eosinophils in PCM has never been evaluated. This study aimed to assess the phenotypic and functional characteristics of eosinophils in PCM. In 15 patients with the acute form of the disease, we detected expression of MBP, CCL5 (RANTES) and CCL11 (eotaxin) in biopsies of lymph nodes and liver. In addition, there were higher levels of chemokines and granule proteins in the peripheral blood of patients compared to controls. Isolation of eosinophils from blood revealed a higher frequency of CD69+ and TLR2+ eosinophils in patients compared to controls, and a lower population of CD80+ cells. We also evaluated the fungicidal capacity of eosinophils in vitro. Our results revealed that eosinophils from PCM patients and controls exhibit similar ability to kill P. brasiliensis yeast cells, although eosinophils of patients were less responsive to IL-5 stimulation than controls. In conclusion, we suggest that eosinophils might play a role in the host response to fungi and in the pathophysiology of PCM by inducing an intense and systemic inflammatory response in the initial phase of the infection. Paracoccidioidomycosis (PCM) is a systemic mycosis caused by dimorphic fungi of the Paracoccidioides genus. It is the most prevalent systemic mycosis of Latin America and, in Brazil, it is the leading cause of death among immunocompetent patients [1–4]. PCM is caused by inhalation of environment Paracoccidioides conidia. The fungus may remain latent in tissues for years, without any clinical manifestation. Depending on the inoculum or host immune response, the disease may develop into two clinical forms: the acute/subacute form, which affects young adults and children, or the chronic form, which affects older adults [5]. The acute/subacute or juvenile form comprises 10% of all cases. It is the most severe form of PCM, characterized by diffuse lymph node involvement, hepatosplenomegaly and bone marrow dysfunction. It may also affect skin and bones. Young patients of both genders are equally affected [3,6, 7]. Patients with acute form of PCM have a depressed cellular immune response as evidenced by delayed-type hypersensitivity (DTH) negative tests, deficient lymphocyte proliferation to yeast antigens and the production of Th2 cytokines such as IL-4, IL-5, IL-10 and TGF-β [8]. In addition, these patients produce high levels of IgE and IgG4 antibodies against P. brasiliensis [9]. Also in this form, eosinophilia had been correlated with negative delayed hypersensitivity skin tests, lower CD4 cells number and high levels of anti-P. brasiliensis antibodies, in addition to disease activity and severity [10,11]. This increased number of eosinophils typically returns to normal after antifungal treatment [10,12–14]. However, little is known about the role of these cells in the pathogenesis of PCM. The role of eosinophils in health and disease has received more attention in the past decades [15–17]. Eosinophils, commonly correlated with immune responses during allergic and parasitic diseases [18,19] participate in both innate and adaptive immunity, since it activates and interacts with several immune cells, including dendritic cells and T lymphocytes [20]. Eosinophils are recruited from the circulation to the inflammatory foci in response to various stimuli. Eosinophil degranulation and release of cytotoxic molecules, i. e. MBP, ECP, EPO and EDN, can quickly affect the microenvironment and influence cell recruitment, tissue repair, homeostasis and remodeling, and also promote a direct response against the pathogen [17,21]. In addition, eosinophils can present antigen to T lymphocytes and, therefore, act as antigen presenting cells (APC) and initiate an immune response to specific antigen [22]. Eosinophils can also act as an effector cell, inducing tissue destruction and dysfunction, as well as promoting exacerbation of the inflammatory response through the release of toxic proteins from their granules, cytokines and lipid mediators [23,24]. To date, there are no studies evaluating the role of eosinophils in PCM, even though eosinophilia is part of the diagnostic criteria of the acute form [10,12–14]. The aim of this study was to evaluate the functional capacity of peripheral blood eosinophils in patients with the acute form of PCM. All study procedures were performed after informed consent, in accordance with standards established by the Committee of Ethics in Research School of Medical Sciences, UNICAMP (No 449/2008). Written informed consent was obtained from all participating subjects or their parents/guardians (on behalf of child participant). We included 15 patients with the acute/juvenile form of PCM evaluated at a Pediatrics clinics at the University of Campinas, São Paulo, Brazil. The diagnosis was established by direct examination and/or serology (double immunodiffusion test). The control group was composed of 11 healthy individuals with a maximum age of 35 years old. For each patient and control three fecal samples were collected in alternated days and examined microscopically for the presence of intestinal parasites. All experiments were performed with peripheral blood eosinophils obtained at the moment of diagnosis, before the beginning of treatment. Paracoccidioides brasiliensis strain 18 (Pb18) and strain 265 (Pb265) yeast cells were obtained from Microbiology Laboratory (State University of Campinas, Sao Paulo, Brazil) and maintained by weekly subcultivation in Fava-Netto medium at 37°C. After 5 days of culture, yeast cells were washed and suspended in phosphate-buffered saline (PBS, pH 7. 2). Fungal suspensions were homogenized with glass beads in a vortex homogenizer in order to obtain individual cells. Yeast viability was determined by Trypan blue exclusion test and we used fungal suspensions containing more than 85% viable cells. Biopsies specimens were taken from lymph nodes and liver of acute PCM patients before the beginning of treatment. Samples were fixed in 4% formaldehyde and embedded in paraffin. Immunohistochemical analysis was performed using antibodies for MBP, CCL5 (Santa Cruz, CA, USA) and CCL11 (R&D Systems, MN, USA). A polymer-based method was used (MACH 4 Free biotin-Detection, Biocare Medical, USA), according to the manufacturer´s instructions. Brown staining indicated positive reaction. Eosinophils were isolated using negative selection method, as described elsewhere [25,26]. Peripheral blood collected in heparin tube from healthy donors and patients with acute PCM before the beginning of treatment. It was then diluted 1: 1 with phosphate buffered saline/bovine serum albumin 1% (PBS/BSA 1%) and overlaid onto Percoll gradient (density = 1. 088 g/ml), centrifuged at 1000 × g for 20 min, 20°C and the pellet containing red cells and granulocytes was collected. Red cells were lysed using lysing buffer (155 mM NH4Cl, 10 mM KHCO3 0. 1 mM EDTA). For eosinophil purification, granulocytes were incubated with an antibody cocktail of eosinophil isolation kit (MACS, Miltenyi Biotec Inc. , Auburn, CA, USA) for 15 minutes. After this period, cells were incubated with anti-CD16 immunomagnetic microbeads for 15 min at 4°C and added to a magnetic column field. The negative cells, eosinophils, were collected. Eosinophils were suspended in Eagle' s minimum essential medium (MEM, Sigma Chemical CO. , MO, USA), pH 7. 2 (> 92% eosinophils). The cell viability (97%) was assessed in the trypan blue dye exclusion test. The purity of eosinophils (CD16-CD49d+) was verified by FACs and reached > 98%. Serum levels of CCL11, CCL5, CXCL9, and CXCL10 of patients and healthy individuals (controls) were quantified by ELISA technique using kits from R&D Systems (Minneapolis, MN, USA). The granule proteins of eosinophils were assessed by commercially available ELISA according to the manufacturer’s instructions. For MBP, EPO and ECP we used USCN Life Science Inc (Wuhan, Hubei, China) and for EDN we used ALPCO immunoassays (Salem, NH, USA). Eosinophil migration was measured using a 96-multiwell ChemoTx 101–5 chamber (NeuroProbe Inc. , Cabin John, MD, USA). Eosinophils from patients with acute PCM and healthy volunteers were ressuspended in MEM at a concentration of 3 x 106 cells/mL. First, microplate wells were filled with 29μL of chemotactic agent (CCL5 - 100ng/mL, CCL11 - 100ng/mL or interleukin-5–100 and 50ng/mL) diluted in MEM or only MEM (spontaneous migration). The upper part was separated from the lower chamber by a 5μM polycarbonate membrane. Subsequently, 25μL of eosinophil suspension was added to the polycarbonate membrane. Eosinophils were isolated ex vivo or were previously incubated with P. brasiliensis yeast cells for 4 hours. The chamber was then incubated for 2 h at 37°C in a humid atmosphere with 5% CO2. After incubation, the non-migrating cells on the top side of the filter were removed by a tissue and the chamber was centrifuged at 200 x g for 5 min at 20°C. The filter was then removed and the number of cells that had migrated into the bottom compartment was determined by measuring residual eosinophil peroxidase (EPO) [27,28]. Fifty μL of EPO substrate (1 mM H2O2,1 mM o-phenylenediamine and 0. 1% Triton X-100 in Tris buffer, pH 8. 0) were added to each well. After 30 min at room temperature, 25 μL of 4M H2SO4 were added to each well to stop the reaction and absorbance measured at 490nm in a microplate reader (Bio-Rad Laboratories Inc. , Philadelfia, PA, USA). The number of migrated cells was calculated by comparing absorbance of unknown samples to that of the standard curve (eosinophil number versus EPO activity) for cells from controls and PCM patients. The standard curves for cells from control and PCM patients did not differ significantly, indicating that EPO activity can be used as a marker of eosinophil number. We calculated the relative migration by dividing the number of eosinophils that migrated towards the chemoattractants by the number of eosinophils that migrated towards MEM (spontaneous migration). Each experiment was carried out in triplicate. To assess whether eosinophils have the capacity to adhere to endothelial cells we used HLECs. Ninety-six-well plates were prepared by coating individual wells with HLECs in DMEM (100 μl; 2x105cells/mL) for 2 days at 37°C in a humid atmosphere with 5% CO2. After this period, fresh DMEM-FBS was added and part of the cells were stimulated with TNF-α (10ng/mL) [29,30] for 4h. The plates were washed twice again with DMEM-FBS. Eosinophils of patients and controls were incubated for 30 minutes with CCL5 (100ng/ml), CCL11 (100ng/ml) or IL-5 (100ng/mL), as previously documented [23] and then added to HLECs, previously stimulated or not with TNF-α. After 1h incubation with HLECs, non-adhered cells were removed and the remaining cells were washed twice with DMEM-FBS. At the end of the washings, the wells were filled with 50 μL of MEM. Eosinophil adhesion was calculated by measuring residual eosinophil peroxidase (EPO) activity of adherent cells, as previously described [27,28]. Characterization of T cells activation markers and PRRs on the surface of eosinophils was performed using flow cytometry. Eosinophils were purified from the peripheral blood of patients with the acute/juvenile form of PCM and controls as described (Eosinophil isolation). Alternatively, some cells were incubated for 4 hours in the presence of P. brasiliensis yeast after purification. For the flow cytometry assay, eosinophils were incubated with 200μL human AB serum for 10 minutes at 4°C. Cells were washed with 1 mL of dilution buffer [PBS-BSA (0. 1%), NaN3 (0. 2 mM) ], the supernatant was discarded and the precipitate suspended in dilution buffer. After this procedure, 20μL of the cell suspension were distributed in 96-well plates with U bottom containing the following antibodies: anti-CD16 (FITC), anti-TLR2 (FITC), anti-CD86 (FITC), anti-TLR4 (PE), anti-HLA-DR (PE), anti-CD80 (PE), anti-CD69 (PE-Cy5) e anti-CD49d (PE-Cy5) and isotype controls, diluted in dilution buffer. After 20 minutes incubation at 4°C, in the dark, cells were washed with 120μL of dilution buffer. The supernatant was discarded and the cells were suspended in 200μL of 2% formaldehyde. After the transfer to appropriate tubes, the reading was held in flow cytometer (FACScalibur, Becton & Dickson, USA). For each sample, a minimum of 25,000 events were acquired. The results (relative cell percentage for each parameter in the isolated population of eosinophils) were analyzed using FSC Express (v. 3, De Novo Software, USA). The fungicidal activity of eosinophils directly against P. brasiliensis yeast cells was evaluated by incubating eosinophils with P. brasiliensis in a 1: 100 fungus/eosinophils ratio, in a U bottom 96-well plate, for 4h at 37°C. After this period, 100μL of this culture (diluted or not) was plated on brain-heart infusion agar (BHI, Difco Laboratories, MI, USA), which was supplemented with 4% horse serum and 5% P. brasiliensis isolate 192 culture filtrate. In some cultures we added yeast cells and human rIL-5 (25ng/mL) (Peprotech, NJ, USA). Plates were incubated at 37°C and the number of colony forming units (CFU) were determined from the fifth to the thirtieth day of culture and then corrected by the dilution factor and expressed in number of CFUs/mL. GraphPad Prism software (v. 5, San Diego, CA, USA) was used for statistical analysis. After analyzing data for normality, results were presented as mean ± SEM or median, accordingly. Student' s t-test or Mann Whitney U test was used to compare parameters between patients and controls, as appropriate. One way repeated measures ANOVA, followed by Tukey post test, was used to compare different conditions in the same experiment. A P value < 0. 05 was considered statistically significant. Table 1 shows clinical characteristics of the 15 patients included in the study at the moment of diagnosis, before the beginning of treatment. Patients' ages ranged from 5 to 14 years-old (mean: 9. 57 years-old). We decided to include a 32-years-old female patient (patient 15) because she presented clinically with the acute form of PCM, including eosinophilia and lymph node involvement. In agreement with previous data on the acute/juvenile form of PCM, there was no significant difference between the proportion of male and female patients [1,31]. As expected, lymph nodes and liver were commonly affected [5]. The absolute number of eosinophils in peripheral blood was found to be increased in most patients, with exception of two. Cell counts ranged from 0. 29 to 12. 93 x 103/mm3. The anti-P. brasiliensis antibody was detected in the serum (immunodiffusion test—ID) of most patients (Table 1). For those patients presenting with a negative immunodiffusion (P10 and P12) we performed ELISA tests using P. brasiliensis gp43 as antigen and detected antibodies at 1/51. 200 and 1/12. 800 titers, respectively (Table 1). C-reactive protein levels were elevated in most patients, ranging from 3. 8 to 175. 0 mg/L. All patients tested negative for intestinal parasitosis and did not report any allergies. Liver and lymph nodes biopsies of patients with the acute/juvenile form of PCM were evaluated by immunohistochemistry in order to detect MBP, CCL5 and CCL11. Fig 1A and 1D show cells with morphology of eosinophils expressing MBP in the inflammatory infiltrate around fungal cells in the liver and lymph nodes, respectively. CCL5 was detected in areas of hepatic parenchyma, as well as in giant cells (Fig 1B). In lymph nodes, CCL5 was mainly detected in giant cells, around P. brasiliensis yeast cells. CCL11-stained cells were diffusely demonstrated in the hepatic parenchyma (C) and lymph nodes (F). Altogether, the examined biopsies showed rich eosinophils infiltrates in close contact with fungal cells and MBP, an eosinophilic granule protein. In accordance, CCL5 and CCL11, chemokines involved in eosinophils recruitment, were also detected in P. brasiliensis infected tissues. Regulation of expression of adhesion molecules is crucial in controlling inflammation, and the adhesion of leukocytes to endothelial cells is the first step in the recruitment of these cells to the inflammatory sites. Eosinophils express adhesion molecules such LFA-1 and Mac-1, which interact with endothelial cell through ICAM-1 [34]. In addition to demonstrating an increased production of chemokines and granule proteins in the serum of PCM patients, as well as higher in vitro migratory capacity, in response to CCL11, we evaluated the adhesion of eosinophils to lung endothelial cells (HLECs). Under baseline conditions, eosinophils of patients had a higher adhesive capacity than the ones from controls (Fig 4A). We also stimulated eosinophils with CCL5, CCL11 and IL-5 to evaluate the effect of these inflammatory mediators in the adhesion of eosinophils to HLEC. The stimulation with CCL11 and IL-5 did not change the adhesion capacity of eosinophils (from both PCM and controls) (Fig 4B and 4C). The stimulation of eosinophils with CCL5 promoted a slight increase in the adhesion of control eosinophils to HLECs, but not PCM eosinophils (Fig 4D). Given that we found a higher migratory and adhesion capacity in eosinophils obtained from PCM patients, in addition to increased serum chemokines and granule proteins, we evaluated whether the disease would lead to an increment in activated eosinophils in peripheral blood from patients. We then analyzed the frequency of eosinophils positive for the activation marker CD69 in the peripheral blood from PCM patients and controls. We demonstrated a much higher frequency of CD69+ eosinophils in PCM patients than in controls (Fig 5). In the analysis of the frequencies of TLR2+ and TLR4+ eosinophils in the peripheral blood of both groups, patients with acute/juvenile form of PCM had more circulating TLR2+ eosinophils than controls. There was no difference in TLR4 expression between groups (Fig 5). Eosinophils can function as antigen-presenting cells and stimulate T cell response in vivo. Therefore, they must express cell surface components necessary for antigen presentation and T cell activation (MHC-II, CD80 and CD86) [22,35]. We next evaluated whether these molecules were differentially expressed in patients and controls. We demonstrated decreased CD80+ circulating eosinophils in PCM patients when compared with the control group. Conversely, we did not find differences in the percentage of CD86+ or MHC-II+ eosinophils between groups (Fig 5). Altogether, our results suggest that during PCM, peripheral eosinophils are more activated and express more TLR2 than non-infected controls. However, the percentage of CD80+ cells is reduced in patients, which could have an inhibitory effect on T cell activation. These results indicate that during paracoccidioidomycosis there is an activation of eosinophils, revealed by high levels of serum chemokines and granule protein release, increase adhesion and migration capacity and CD69 expression. In order to assess differences between PCM patients and healthy controls in eosinophils' ability to kill fungal cells, P. brasiliensis yeast cells were cultured with eosinophils for 4h and the number of CFU/mL was estimated. Eosinophils from healthy donors and patients showed direct cytotoxic activity against P. brasiliensis yeasts, independent of the virulence of the strain (Fig 6). However, the addition of IL-5 (25ng/mL) induced higher fungicidal activity to Pb18 in controls (p<0. 05 in relation to unstimulated eosinophils) than in patients (Fig 6A). In relation to Pb265 eosinophils of PCM patients are also less responsive to IL-5 than controls (Fig 6B). Peripheral eosinophilia is a common finding in patients with the acute/juvenile form of PCM, especially in children [13,36,37]. However, prior studies evaluating the role of these cells in the immune response against P. brasiliensis were lacking. We analyzed phenotypic and functional characteristics of eosinophils in patients with acute PCM when compared with healthy donors. We detected high levels of antibodies against P. brasiliensis and C-reactive protein, a systemic inflammatory marker. Two patients had a negative immunodiffusion test (P10 and P12), despite positive detection of fungal cells in biopsy specimens. ELISA assay detected high titers of antibodies against gp43 of P. brasiliensis in the serum of both patients. In previous work, we demonstrated that lack of reactivity in immunodiffusion test could be associated with production of low affinity IgG2 class antibodies against carbohydrate epitopes [38]. In accordance with previous reports, all patients presented with mononuclear phagocytic system involvement (lymphadenomegaly and hepasplenomegaly), in addition to bones and skin lesions [39–41]. Activated eosinophils can produce chemokines such as CCL5, CXCL9 and CXCL10. CXCL9 and CXCL10, IFN-γ induced chemokines, are chemotactic for activated T cells and signal through the common receptor CXCR3, which is expressed by memory T cells (CD45RO+) preferably of Th1 and NK cells, but not monocytes or neutrophils [42,43]. In allergic inflammation, the CC chemokines i. e. CCL5 (CCL5), promote cell migration and activation inflammatory cells, including eosinophils [44]. High concentrations of CCL5, CCL2 (MCP-1), CXCL10 and CXCL9, associated with infiltration of mononuclear cells, were detected in the lungs of P. brasiliensis infected mice and sera of patients with PCM [45,46]. In our study we detected high concentrations of CXCL9 and CCL5 in the serum of patients with acute PCM compared to healthy subjects. This may be related to higher eosinophil counts and inflammation typically found in patients with this form PCM. Moreover, biopsies of patients also revealed the presence of MBP, CCL5 and CCL11 in tissues, surrounded by fungal cells. Wagner and colleagues also described eosinophils infiltration and deposit of MBP on P. brasiliensis yeast cells in granulomas of patients with PCM [47]. These results suggest that eosinophils may assist in the synthesis of chemokines that promote the recruitment of effector T cells and other eosinophils both systemically as well as to the infection site. We detected higher levels of eosinophil granule proteins in sera of PCM patients when compared to controls. Yang and colleagues have shown that EDN, an eosinophil granule-derived secretory protein, is a TLR2 ligand, and enhances antigen-specific Th2 immune responses [48]. The higher production of EDN by PCM patients may enhance the Th2 polarized immune response present in acute PCM [49–51]. Furthermore, Eosinophils from PCM patients presented a higher migratory capacity in response to CCL11 than eosinophils from controls. However, pre-incubation with both P. brasiliensis strains did not change this capacity, showing similar results as the unstimulated condition. Eosinophil adhesion to endothelial cells is an important step in the migration process of these cells from blood vessels to the tissues. In this context, CCL11 have a central role by increasing the adhesion and transendothelial migration of eosinophils [17]. We also demonstrated that the adherence of eosinophils to human lung endothelial cells was higher in eosinophils from PCM patients when compared with healthy subjects. The adhered eosinophil numbers were higher in PCM patients, both in basal conditions and when cells were pre-stimulated with CCL11 or IL-5. Our results suggest that eosinophils from acute PCM are more activated, able to produce higher levels of chemokines and granule proteins. Moreover, eosinophils from PCM patients have a higher migratory capacity under CCL11 stimulus and are able to adhere more to lung endothelial cells, which could favor the influx of these cells to inflammatory sites. Although both migratory and adherence capacity of eosinophils were higher in patients than in controls, no change was observed when the cells were stimulated with fungal cells or chemokines/cytokines (Figs 3 and 4, respectively). These results may imply that eosinophils from patients with the acute form of PCM are already fully activated to migrate and to adhere to endothelial cells. Accordingly, we next evaluated the expression of CD69, pattern recognition receptors and T-cell activation molecules in eosinophils from PCM patients and healthy controls. As we previously concluded, we found more CD69+ eosinophils in the peripheral blood of PCM patients. The same group presents more TLR2+ circulating eosinophils. On the other hand, the frequency of CD80+ eosinophils in individuals with acute PCM is lower when compared with controls. Besides this apparently increased activation of PCM eosinophils, these patients are not able to clear the infection, and some patients succumb to the disease [39,52–54]. Loures and colleagues have shown that the presence of TLR2 in mice is associated with susceptibility to P. brasiliensis infection [55]. In this study using TLR2-deficient mice, the authors reported that P. brasiliensis might use TLR2 to interact with host cellls, as a fungal virulence mechanism, resulting in lower fungal burden in TLR2-/- mice. The TLR2 deficiency is associated with activation of Th17 cells and lower expansion of regulatory T cells, which leads to an uncontrollled inflammation and tissue damage. A suppressor role of TLR2 was also reported for Candida albicans, through induction of IL-10 and Treg [56]. It was also previously demonstrated the presence of TLR2-expressing regulatory DC in the lungs of mice susceptible to P. brasiliensis, which was not associated with protection [57]. The versatility of recognition by TLR2 is seen as a result from the ability of this receptor to act together with other toll-like receptor, such as TLR1 and TLR6 [58], thus providing proinflammatory (TNF-α and IL-12 production) and anti-inflammatory responses (IL-10 production), a condition which provides an escape strategy used by some pathogens [56,59]. Indeed, our group has previously reported that PCM patients with active disease present higher numbers of Treg cells, compared to patients who had received treatment or controls [60]. The analysis of Treg cells from PCM patients have shown that these cells presented regulatory phenotype associated with suppressive activity, which may contribute to the onset of the disease. Moreover, the increase in TLR2+ population of eosinophils might collaborate to fungal infectiveness, associated with the increase in T regulatory populations, which might lead to a more severe disease, as observed in acute PCM patients. Indeed, it has been suggested that PAMP recognition through TLR by eosinophils is linked to a Th2 phenotype in humans [61]. We next performed a fungicidal assay in order to evaluate the capacity of eosinophils to direct kill P. brasiliensis yeasts. Our results showed that eosinophils from healthy individuals and PCM patients were able to kill Pb18 and Pb265 yeast cells, although patient cells were less responsive to IL-5 stimulation. These results might indicate that in PCM the fungicidal activity of eosinophils does not account for protection. Given the importance of the innate immune response leading to protection or immunopathology, and the quality of the adaptive immune response, we believe that eosinophils might have a role in the pathophysiology of the acute form of PCM. In this context, the results of this study confirm the initial hypothesis that eosinophils participate in the early stages of host response to fungi promoting an intense and systemic inflammatory response that result in an inefficient immune response against P. brasiliensis. Moreover, the higher TLR2 expression suggests that they might have an auxiliary role in the Th2 immune polarization found in patients with acute PCM.

Paracoccidioidomycosis (PCM) is a fungal disease endemic of some Latin America countries. The acute clinical form of the disease, which affects children and young adults, is the most severe form of PCM. It is characterized by a depressed T cell immunity and increased number of blood eosinophils that decreases after antifungal treatment. The role of eosinophils in PCM has never been investigated. We found high levels of eosinophil granules and chemokines in serum of patients. Moreover, patients eosinophils have a higher migratory and adhesion capacity compared to controls. Our results indicate that eosinophils may participate in the early steps of host response to fungi promoting an intense and systemic inflammatory response, which may result in an inefficient immune response against P. brasiliensis in vivo.

Abstract Introduction Methods Results Discussion

blood cells cell motility medicine and health sciences immune cells body fluids pathology and laboratory medicine immunology tropical diseases lymph nodes signs and symptoms lymphatic system paracoccidioidomycosis neglected tropical diseases fungal diseases immune system proteins infectious diseases white blood cells inflammation animal cells proteins chemotaxis immune response toll-like receptors biochemistry signal transduction eosinophils diagnostic medicine blood cell biology anatomy physiology chemokines biology and life sciences cellular types immune receptors

2017

Functional and phenotypic evaluation of eosinophils from patients with the acute form of paracoccidioidomycosis

Accumulating experimental evidence suggests that the gene regulatory networks of living organisms operate in the critical phase, namely, at the transition between ordered and chaotic dynamics. Such critical dynamics of the network permits the coexistence of robustness and flexibility which are necessary to ensure homeostatic stability (of a given phenotype) while allowing for switching between multiple phenotypes (network states) as occurs in development and in response to environmental change. However, the mechanisms through which genetic networks evolve such critical behavior have remained elusive. Here we present an evolutionary model in which criticality naturally emerges from the need to balance between the two essential components of evolvability: phenotype conservation and phenotype innovation under mutations. We simulated the Darwinian evolution of random Boolean networks that mutate gene regulatory interactions and grow by gene duplication. The mutating networks were subjected to selection for networks that both (i) preserve all the already acquired phenotypes (dynamical attractor states) and (ii) generate new ones. Our results show that this interplay between extending the phenotypic landscape (innovation) while conserving the existing phenotypes (conservation) suffices to cause the evolution of all the networks in a population towards criticality. Furthermore, the networks produced by this evolutionary process exhibit structures with hubs (global regulators) similar to the observed topology of real gene regulatory networks. Thus, dynamical criticality and certain elementary topological properties of gene regulatory networks can emerge as a byproduct of the evolvability of the phenotypic landscape. Several models of network growth and evolution have been devised to generate networks with specific topological properties (such as hub-like structures, [10], [11]) or with a particular type of dynamical behavior [12], [13]. The ‘dynamics’ of a genetic network, that is, the collective change of gene expression of all the genes in the network, (i. e. of the gene expression pattern), is obviously the more appropriate phenotype on which evolution acts than the topology itself. However, networks are often, again, trained explicitly to exhibit a particular behavior, such as robust dynamics under certain kinds of perturbations [14]–[18], or to perform some arbitrarily imposed task [19]. Usually, the training is achieved by selecting the networks that score highest with respect to a suitable fitness function. In contrast to such explicit targeting of particular phenotypes as endpoints we propose that an elementary and more encompassing set of constraints must be taken into account, which is epitomized in these two distinct trade-offs of opposing features: The second trade-off epitomizes the two central properties that underlie evolvability [20]–[22]. Concretely, the evolutionary trade-off, the central subject of this study, implies that when new phenotypic traits are developed, the old, useful traits do not disappear but are conserved or transformed into something similar. A fundamental question in evolutionary biology is whether the evolutionary trade-off is the result of adaptation by natural selection, or arises through non-adaptive mechanisms. There is a great amount of evidence suggesting that evolvability itself is a selectable trait and hence, evolvability evolves [23]–[26]. However, the mechanisms through which evolvability evolves are still under debate. The two dualisms, the evolutionary trade-off and the developmental trade-off, are of course interconnected in the sense that the latter is an adaptive phenotype of the evolving individual, that is, it is shaped by selection pressure. Indeed it was precisely because of the developmental trade-off that critical dynamics has been hypothesized to play an important role in evolution [27]–[29]. Critical dynamical systems operate at, or close to, a phase transition between ordered and chaotic dynamics. They exhibit a series of very remarkable properties that would be difficult to explain in the absence of criticality, such as collective response to external stimuli without saturation [30]–[32], optimal computational capabilities [33], fast information storage, transfer and processing [34], etc. In fact, the existence of critical dynamics in living systems has been increasingly recognized as an important property that confers collective behavior over many different scales [35]. In general terms, critical dynamics in gene regulatory networks implies that perturbations of gene expression would neither amplify and percolate through the system (manifest by the overwhelming divergence of the trajectories of any two initial states, as seen chaotic systems) nor would they immediately “die out” (manifest by the overwhelming convergence of the trajectories of any two initial states, as seen in ordered systems). In computational models gene regulatory networks that operate in the dynamically critical regime (between order and chaos) have been shown to exhibit both homeostasis (robustness of gene expression states) and developmental progression (change of gene expression state), thus achieving some sort of optimization (or balance) in the developmental trade-off [4], [5], [27]–[29]. Thus, criticality is a mechanism that, within an organism, engenders dynamical robustness to the network while at the same time allowing the network to respond to developmental perturbations. Therefore, for the development of the individual organism there are compelling reasons to assume that dynamical criticality in their genetic networks is a desirable property. This may explain why experimentally observed gene expression patterns in several organisms indicate that the regulatory networks indeed operate in the critical regime [4]–[9]. However, to our knowledge in previous work on dynamical criticality in genetic networks, this property has either been taken for granted or externally imposed by adjusting the value of a network control parameter that is known to operate the order-chaos phase transition. In these studies the networks are constructed by design to be in the critical phase, (or in the ordered or the chaotic phase) followed by the analysis of their properties and contribution to evolution [30]–[34]. In a case where criticality in fact emerged was due to imposed explicit “rewiring” rules [36]. However, little is known about how dynamical criticality emerges without such explicit enforcement but in an evolutionary process that is inescapably subjected to the constraints of evolvability. Therefore, here we ask: what is the role of evolution in poising gene regulatory networks at the critical phase? How does a gene regulatory network evolve a structure that confers criticality in the first place? What properties must be selected for in order for a non-critical network to become critical? In this work we evolve populations of simulated gene regulatory networks and show that criticality is profoundly linked to evolvability. More specifically, we show that critical dynamics, and hence the developmental trade-off in genetic networks, naturally emerge as a robust byproduct of the evolutionary processes that select for evolvability and optimize the evolutionary trade-off. Furthermore, the emergence of criticality occurs without fine-tuning of parameters or imposing explicit selection criteria regarding specific network properties. As a model for gene regulatory networks we use the Boolean network model proposed by Kauffman [27]–[29], [37]–[40]. It has been firmly demonstrated that this model of complex networks effectively captures essential aspects of gene regulation at the promoter which involve highly cooperative, non-linear, conditional relationships. These mechanisms are adequately encoded by logical functions that can reproduce well the observed dynamics of real networks with partially known topology [29], [39], [40]. But more important, the mapping between network architecture space and dynamical regimes is well known for Boolean networks, such that a randomly generated ensemble of networks can be controllably constrained by network architecture parameters. In brief, a Boolean network is defined by a set of nodes, , representing the genes, each acquiring the values and, corresponding to the two states of gene expression: either the gene is expressed (gene locus is active) or it is not expressed. The value of each node is determined by a set of other nodes in the network, the regulators of, denoted as. The network dynamics are then given by the simultaneous update of all the network elements according to the equation (1) where is an average response time (usually taken as) and is a Boolean function constructed according to the activating or inhibitory nature of the regulators of. For specific networks of real organisms, the connections and Boolean functions can be chosen to capture the molecular biology of the regulatory mechanism that is often known in the form of a qualitative proposition that contain logical relationships. Such modeling approach has been shown to reproduce the observed gene expression patterns in a variety of organisms. Since we are not interested in a particular network of a specific organism, in the initial population we use random networks in which the upstream regulators of a given gene are chosen randomly. The Boolean functions of each gene are also assigned randomly in a way such that for each of the activity configurations of the regulators, the Boolean function outputs to with probability p and to with probability 1-p. The value of p, referred to as the ‘bias’ of the Boolean function, is one of the key parameters of the global gene network architecture that influence the dynamics. Since the discrete valued network with N genes has a finite number of possible dynamical states which form the state space, and since the dynamics given by Eq. (1) is deterministic, any trajectory will eventually visit a state previously visited and enter into a periodic pattern of expression that repeats over and over again. More than one periodic pattern can exist for the same network. Such periodic patterns are the dynamical attractors of the network which thus consist of either a set of states that form a “state cycle” (analogous to limit cycle in continuous systems) or of a single steady state (analogous to a fixed-point attractor). The set of states that converge to the same attractor constitute its basin of attraction. Three important aspects of Boolean networks are relevant here. First, the dynamical attractors of the network correspond to the distinct functional phenotypic states of the cell, such as cell types, as has now been experimentally demonstrated [39]–[42]. Therefore, the set of all the dynamical attractors of a given network constitutes its phenotypic landscape (in the sense of Waddington' s epigenetic landscape [43]) which we refer here as the attractor landscape. Second, two broad regimes of dynamical phases that depend on global network topological parameters can be identified: the ordered and the chaotic phase [7], [27]–[29], [37], [38]. Networks in the ordered phase are dynamically too rigid because in such networks any perturbation in the initial condition eventually disappears and the networks relax back to the pre-perturbation state. In the extreme case, all transients converge to only one attractor state, thus permitting only one stable phenotype. By contrast, networks in the chaotic phase typically have large numbers of attractors and/or vastly long transients and are extremely sensitive to small perturbations, making all network states very unstable. And third, there is a continuous phase transition between the ordered and chaotic phases that is characterized by a nontrivial critical point. Networks that operate close to this critical point display a series of interesting properties of complex systems, such as the maximization of information processing needed for ontogenetic complexity [34], [44]. One order parameter that determines in which dynamical phase the network operates is the average network sensitivity S defined as [45] (2) where K is the average number of upstream regulators per gene and p is the fraction of positive (activating) regulations in the set of all Boolean functions in the network. If the network is in the ordered phase, and if it is in the chaotic phase [7], [27]–[29], [37], [38]. The critical phase is attained at. Note that the above definition of the ordered, critical and chaotic phases refers to the level of dynamics, namely, to the response of the network to transient perturbations. However, we recently found that classification of networks into these dynamical regimes has an interpretation that extends to the evolutionary time scale. Specifically, the probability for a change of the attractor landscape, thus of the global phenotypic behavior, following simple mutations to the network structure is very low for networks in the ordered phase and is very high for chaotic networks [46]. Hence, networks in the ordered regime are not evolvable because they absorb the effect of mutations in a large space of neutral mutations. On the other hand, those in the chaotic regime are highly innovative in the sense that their attractor landscape undergoes large scale changes following the small mutations—but they lack mutational robustness. Critical networks are peculiar because following a simple mutation, new attractors emerge with high probability while conserving existing attractors. Thus, critical networks are both robust and capable of useful innovation, hence are evolvable. In view of this relationship between criticality and evolvability, the question of how genetic networks became dynamically critical is thus linked to the question of how evolvability arose through evolution. Could the evolution of evolvability account for the evolution of criticality? We simulated the evolution of genetic networks in a starting population of M0 = 1000 different random Boolean networks each with N = 10 nodes. Initially, all nodes have exactly upstream regulators and the Boolean functions have a bias p = 0. 5. Hence, the sensitivity of the networks in the first generation is entirely determined by the initial network connectivity through (Eq. 2). We mutate the networks in the population by implementing a mutation algorithm that captures fundamental properties of biological genome growth. Specifically, each node represents a gene that is composed of a regulatory region and a coding region, as illustrated in Fig. 1, and mutations can occur in any of these two parts. Mutations in the regulatory region consist in the addition or deletion of binding sites to DNA. These mutations change the way in which the node is regulated by its upstream regulators (see Fig. 1B and the Methods section for a detailed description of the mutation algorithm). Briefly, mutations in the regulatory region of a given node can eliminate or establish regulatory inputs from existing or new upstream regulators, respectively, or produce changes in its Boolean function. On the other hand, mutations in the coding region of node change the spectrum of its target nodes, which translates into the gain of new targets, loss of existing ones or changes in the Boolean functions of the targets of. Finally, network growth is implemented by simulating the evolutionary mechanism of gene duplication followed by divergence [47]. This is done by randomly choosing one node in the network and duplicating it, along with its network connections, thus increasing the number of nodes in the network from N to N+1. We then simulate gene divergence by mutating either the regulatory or the coding regions of the duplicated node. Due to computational limitations, networks were allowed to grow up to a maximum size of N = 100. It is important to mention that even if the mutation algorithm effectively implements the random addition or removal of input or output connections in the network or changes in the Boolean functions of the nodes, the probabilities for these effective mutations to occur change from one node to another and also in time. The reason for this is that these effective probabilities depend on the network size and on the number of binding sites that each node has. Therefore, in the Materials and Methods section we present the mutation algorithm in terms of the probabilities for adding and removing binding sites to the regulatory regions of the nodes because these probabilities remain constant throughout time and across the network elements. Mutations in the regulatory or coding regions of the nodes occur randomly with a mutation rate per gene per network per generation. Once a given gene is selected to be mutated, one of the mutations [ (1) – (6), as described in the Methods section] is performed. Let be the number of networks in the population at generation. Then, on average networks undergo mutations in one of their genes and are subjected to selection. To select for mutational robustness we evaluate at each generation whether or not the mutated networks conserve the attractors that they had before the mutation and eliminate from the population those networks which do not conserve all their attractors. By attractor conservation we mean strict maintenance of identity of attractor states. If after the mutations one of the network attractors changes even only by one bit in its binary states, that change is enough to declare that attractor as non-conserved. Only the networks that conserve all the attractors they had before the mutations will pass to the next generation. We will refer to this selection process as the attractor conservation criterion (ACC). The elimination of the networks that do not satisfy this criterion reduces the population size to a new value. If the population is still big enough and we just go to the next generation without replicating any network. However, if we replicate the surviving networks in order to restore the population to its original size M0 = 1000 (or to a size as close as possible to 1000). For this we have to decide whether all the networks will equally reproduce, or if some networks will reproduce more than others. In the latter case, we have to define a fitness function which will determine the number of copies (daughters) generated by each of the surviving networks to maintain the population size. Let us assume that we already have a fitness function that assigns a fitness value to the ith surviving network in the population, with larger values of corresponding to fitter networks. In the next section we give a precise definition of based on the gene expression variability within the attractors but for the time being let us just assume that is already given. Then, if the ith surviving network will produce daughters, where the function gives the closest integer to x and the normalization constant guarantees that the new population size is as close to 1000 as possible. (We cannot always make it exactly equal to 1000; for instance, if for all networks and, then triplicating each surviving network will restore the population to). We will refer to this replication mechanism as the α-fitness criterion. In order to simulate phenotypic innovation, every 2000 generations all networks in the population simultaneously undergo a gene duplication-divergence event. Therefore, the duplication rate is in the order of 10−5 per gene per generation which is in the broad range of estimates based on genome sequence data and similar to numbers used in previous models of network evolution [48]–[50]. After this event the only networks that survive and pass to the next generation are the ones that, in addition to fulfilling the ACC, also generate new attractors. We will refer to this selection rule as the attractor innovation criterion (AIC). Therefore, under this criterion we eliminate from the population all the networks which, after the duplication event, either do not satisfy the ACC or do not generate new attractors (even if some of these latter networks fulfill the ACC). In principle, any mutation can generate new attractors. However, we evaluate the emergence of new attractors only after gene duplication events because it is known that the average number of attractors increases with the size of the network N [51], [52]. Therefore, it is much more likely to find new attractors when the network grows. It is worth noting that before the duplication event the network had N nodes, and after the duplication it has N+1. Hence, to compare the attractors of the network before and after the duplication event in order to check for conservation or innovation, we only take into account the first N nodes of the network (genome) which are the ones common before and after the duplication event, and ignore the value of the (N+1) th node which represents the new gene resulting from the duplication event. Another important point to mention is that, due to computational limitations, in our simulations the whole set of attractors in the attractor landscape was determined only for small networks (N<25). For large networks (N≥25) a thorough search of the state space to find all the attractors is very time consuming and not feasible. Therefore, to assess attractor innovation in large networks we sampled just a subset of the possible states. Clearly, the AIC was applied only to the attractors that were found with this subsampling. In the Methods section we explain the details of the algorithm to find new attractors. Another aspect we took into account when new attractors emerge is that the nodes in these attractors must contribute to a phenotype. In other words, as the network grows and mutates, the new nodes added to the network cannot be all frozen in state 1 or all frozen in state 0. In the attractors some of the new nodes must be 1 and some others must be 0 (or they can oscillate as well). Without this condition, the growing part of the network would carry no useful information. Networks whose attractors have no information are biologically irrelevant, as it is known that real organisms have gene expression profiles with high information content [6]–[8]. Thus, the information content of the attractor states can be used to define the aforementioned fitness function that determines the replication rate of the networks. In order to do so, we define the average gene expression variability of the network attractors as, where and are the average fractions of 0' s and 1' s in all the states of all the attractors of the network (clearly,). Thus, if almost all the nodes in the attractors are in only one state (either 0 or 1), whereas if more or less half of the nodes in the attractors are in the state 1 and the other half in the state 0. In order to implement this phenotypic fitness after the duplication event, when restoring the population to a size close to 1000, we replicate each surviving network by a quantity proportional to its average gene expression variability. This is the α-fitness parameter that we mentioned in the previous section. It is important to note that there are two ways to measure the variability α, as illustrated in Fig. 2. The first way is to measure the variability along the N nodes of each attractor state (Fig. 2A), and then average over all the states in the attractor and over all the attractors in the network. We will refer to this parameter as the horizontal gene expression variability and denote it as. The second way is to measure the variability of each node individually along the attractor cycle (Fig. 2B) and then average over all the nodes in the network and over all the attractors. We will call this quantity the vertical gene expression variability and denote it as. These two parameters need not give the same results, as illustrated in Fig. 2, where whereas for the same attractor. In all the numerical simulations presented here the α-fitness criterion was implemented using the horizontal variability. It is also important to stress the fact that the ACC (corresponding to mutational robustness), the AIC (corresponding to phenotype innovation), and the fitness act on the attractor landscape, which is entirely an intrinsic dynamical property of the network. Therefore, our selection criteria do not train the network to perform an arbitrary task imposed externally. On the contrary, the ACC, the AIC and the fitness acting together throughout the evolutionary process optimize the networks in the population with respect to the evolutionary trade-off by taking into account conservation and expansion of the network' s intrinsic attractor landscape, whatever it is. Fig. 3A shows the evolution of the average network sensitivity of the population, where the average is taken over all the networks in the population at generation. The four different curves correspond to four different starting populations, each consisting either of only ordered networks, only critical networks, or only chaotic networks, according to the initial sensitivity. The curves that converge to (representing chaotic dynamics) show the effect of a control algorithm in which mutations where applied without selection (all networks survive in each generation). Thus, the mutation algorithm alone does not account for the emergence of criticality because it produces chaotic networks. By contrast, when selection is present, the sensitivity of the networks in all populations converge, on average, to the value, indicating evolution to criticality. Therefore, Darwinian selection, realized by the selection filters ACC and AIC, promotes the evolution of networks towards criticality. Fig. 3B shows the distribution of sensitivities in one of the populations that started with chaotic networks () at two distinct generation times in the simulation, early (generation) and at the end (). The distribution reveals that not only does the average evolves towards criticality (mean) but that the initially broad diversity decreases throughout evolution (the standard deviation decreases almost one order of magnitude, from at generation to at generation). The results reported in Fig. 3A are highly reproducible. (In Fig. S1 we present similar plots for 30 more realizations of the evolutionary processes, including seven realizations for which the networks in the initial population had nodes with varying input connectivity. Additionally, in Fig. S2 we present the Derrida maps of the networks that result from the evolutionary process, which show in a more formal way that all the networks become critical. See Text S1 for a definition of the Derrida map.) Another important property to look at is the gene expression variability of the evolved networks. Since in our numerical simulations we used the horizontal variability as the fitness parameter that determines the replication rate of the surviving networks, in the final population all the networks have, as expected (data not shown). However, it turns out that the vertical variability is also distributed mostly around, as Fig. 4A shows. This is a non-trivial result first, because there is no reason a priori to expect, as these two quantities need not bear any relationship (see Fig. 2). But second, and more importantly, because control networks that are explicitly constructed to be critical de novo have a distribution of vertical variability dominated by, as shown in Fig. 4B. Thus, the fact that the evolved networks have both and, cannot be trivially explained as an inherent feature of criticality nor by selection for α-fitness alone. Rather, it is a result of the entire evolutionary processes. To determine how restrictive the selection criteria, ACC and AIC, that must be satisfied for a network to survive selection, are, we measured the survival times of the networks by tracking individual networks (Fig. 5). We tracked all initial networks in the population by labeling them individually with an integer ranging from 1 to 1000 at generation. When one network is replicated into several copies the “daughter” networks acquire the same label from the “mother”. Since the networks that fail the selection criteria are removed from the population, some labels can disappear altogether from the population. This would correspond to the extinction of one lineage. If at generation g only one label is left in the entire population of networks this can be considered the “fixation” of a particular strain in the population and we re-label the networks again from 1 to. Fig. 5A shows the evolution of strains (labels) through 20,000 generations. Presence of individual strains in the population is indicated by the horizontal lines, with the longest surviving strains defining the fixation events indicated by the vertical lines. The vast majority of strains disappeared from the population very quickly while only very few strains survived for long periods. Interestingly, a goodness-of-fit test indicates that the distribution of survival times is highly consistent with a power-law, with exponent (Fig. 5B), as observed for geological life spans of genera from fossil records [53]. Whether or not is in fact best fitted by a power-law is here not of fundamental relevance. Of significance however is the broad tail exhibited by this distribution, for it shows that the vast majority of strains disappear very quickly from the population and only very few strains are able to survive. Therefore, the results reported in Fig. 5 demonstrate that evolution towards criticality via the fitness criteria of attractor conservation and innovation, and of gene expression variability, indeed confronts the population to a series of highly restrictive selective filters (bottlenecks) through which only very few networks are able to go. The data reported in Figs. 3 and 5 also show that, even though there is a great genotypic and phenotypic diversity in the initial population (because initially all the networks are structurally different and have different attractor landscapes), throughout generations the population passes through a series of selective filters which decrease this diversity by eliminating from the population the majority of strains. Indeed, it is clear from Fig. 5 that several fixation events occur throughout the evolutionary processes. Therefore, at the end of the simulation all networks in the population come from one common ancestor. This has the remarkable consequence that all networks in the final population have the same phenotype (the same set of attractors), but slightly different genotypes. These small genotypic differences are reflected in the small, but not vanishing, standard deviation in the final distribution of sensitivities. In the next section we will come back to the structural differences that exist between the networks in the final population. Of great interest is the structure (or topology) of the networks that survive until the end of the evolutionary processes, for such structure should encode the evolutionary trade-off that these networks were optimized for. We started the simulation with homogeneous random networks for which all nodes had the same number of inputs (in-degree) and where the number of outputs (out-degree) was Poisson distributed. However, at the end of the simulation the evolved networks contain global regulators, namely, nodes with a large number of output connections (targets), as illustrated in Fig. 6. In fact, the typical network structure produced by our evolutionary process was qualitatively similar to the structure of the giant component of the E. coli transcription factor interaction network [2], [3], [46] (Fig. 6). This structure is characterized by short-tailed in-degree distributions (Poisson or exponential) and long-tailed out-degree distributions. Such an outcome was unexpected for two reasons. First, the specific structure of the network was never explicitly considered in the selection mechanism nor did we implement any explicit re-wiring rule as in other models of network evolution [36], [54], [55]. Second and more importantly, global regulators introduce strong correlations in the network dynamics and it is not obvious that these correlations offer an advantage in surviving the selection pressure imposed by the ACC and AIC. Although the final networks are too small to accurately determine the out-degree distribution resulting from this evolutionary process (N = 100), the systematic occurrence of nodes with a high number of output connections (hubs) suggests that this type of network structure could also be an emergent property intimately related to the critical dynamics and evolvability of the network, as it has been suggested for other types of networks [56], [57]. It is important to mention that the existence of hubs in the evolved networks is not simply a consequence of the mutagenic algorithm because control networks that “evolved” without selection but subjected to the same type of mutations do not exhibit this characteristic (see Fig. S3). As was mentioned before, although all networks in the final population had exactly the same attractor landscape, the networks themselves are not identical to one another. This is shown in Fig. 7, where three networks randomly chosen from the final population are displayed (A, B, and C). It is clear that, although similar, these networks are not identical. The final diagram D is a superposition of all the networks in the final population. Since all the final networks came from the same common ancestor, the genes in all these networks have the same evolutionary history. Therefore, it is possible to stack up these networks on top of each other and compare them. In order to measure the degree of similarity between these networks, we computed the fraction of occurrence of the link between the nodes and in the population, for all pairs i and j. Thus, if the two nodes and were connected in all the networks in the population, whereas if then these two nodes were linked only in one network of the population and disconnected in the rest of the networks. Very remarkably, Fig. 7D shows that the more persistent links in the networks throughout the population are the ones that belong to the global regulators. The existence of global regulators in the final networks raises the question as to whether the common ancestor network (from which all the other networks evolved) had, just by chance, some nodes with a “special” topological context that predestine them to eventually become the global regulators. For instance, it could be the case that the common ancestor network contained nodes with a number of output connections far above average. These initial hubs might have played an important role in controlling the network dynamics from the very beginning and therefore they may have remained being hubs throughout the evolutionary processes and end up as the global regulators observed in the final networks. To answer this question we performed simulations in which all the networks in the initial population were explicitly constructed with one node with a high number of output connections. Fig. 8A shows a typical example in which the common ancestor network has one hub that regulates 80% of the other nodes in the network (in this particular case the hub is node 9). However, at the end of the evolutionary process (generation g = 200000) this initial hub has turned into just another ordinary node in the network with no special characteristics (Fig. 8C). This can be seen more quantitatively in Fig. 8D, which shows, for each link of the common ancestor network, the fraction of occurrence of that link in the entire population at two generation times: after the first fixation event (black histogram), and in the final population (red histogram). It is apparent from this figure that even after the first fixation event the initial hub has lost some of its connections in many networks of the population. At the end of the simulation processes none of its original connections significantly occurs in the final population. By contrast, two of the original nodes (nodes 2 and 7) without any special property become the global regulators in the final networks. Results similar to the ones reported in Fig. 8 systematically occurred in our numerical simulations, namely, the initial hubs in the common ancestor networks always lost their “hub” property throughout the evolutionary processes and ended up just as random ordinary elements in the final networks. Furthermore, very often the nodes that became the hubs in the final networks were not even present in the initial networks, but added later at some intermediate generation as a result of a duplication/divergence event. The α-fitness criterion was introduced to increase the reproduction rate and hence to favor those networks that exhibit high gene expression variability (information content) in their attractor states. If we perform the evolution of the networks solely by applying the ACC and the AIC but without using the α-fitness in the selection (which is equivalent to setting α = 1 for all networks), then all the surviving networks at each generation will generate the same number of descendants, equally contributing to the population at the next generation. Under such circumstances, the attractors in all the networks of the population will end with only zero values for σ, as shown in Fig. 9A (only the first 10 genes show some activity because they were the only ones present in the initial generation). This is mainly due to steps 1 and 3 of the mutation algorithm presented in the Methods Section which, together with the ACC, introduce a bias towards the state 0 in the Boolean functions. This in turn is needed to consider the physical meaning of the new Boolean functions: Each time a new gene is added to the network (through a gene duplication), the extension of the Boolean functions of the target genes that have accepted the new gene as their new regulator (input) is carried out by expanding the Boolean function' s truth tables of each target gene as follows: Where in the configuration of the new expanded input vector (row in truth table) the new gene has value the output of that target gene is assigned 1 or 0 randomly; whereas when in the input vector, the output is kept equal as it was before the addition of the new gene because in that input configuration the new regulator is in the off state and does not contribute to the regulation. Consequently, it follows that a trivial way to fulfill the ACC and preserve the old attractors after the duplication event is by selecting networks in which the new gene is inactive (i. e.) in all the attractors, since in this case the new part of the Boolean function is never used. Thus, without the α-fitness filter, all the new genes would appear in the 0 state in all the attractors. (This does not mean that in the transient states before the attractor is reached, the new genes cannot take, transiently, the value 1.) However, it should be noted that even without the α-fitness we can still obtain critical networks as a result of the evolutionary process. To show that criticality does not depend on α-fitness we enforced the evolution of criticality and show that such networks do not exhibit α-fitness. Thus, we evolved populations of networks subjected to the ACC and the AIC as usual. But instead of using the gene expression variability α as the additional fitness parameter, we demanded sensitivity S to be close to 1 as a selection criterion. Specifically, we explicitly selected for criticality by making the replication rate of the networks proportional to. Thus, networks with S≈1 were replicated at a higher rate than networks with S far from 1 (networks with negative values of did not replicate). Fig. 9B shows the evolution of the average network sensitivity using this “S-fitness” criterion (together with the ACC and the AIC). As expected, the average sensitivity of the population very quickly approaches 1 and remains close to 1 throughout the evolutionary process. Fig. 9C shows the histogram of sensitivities in the final population (generation g = 200000). It is clear that this process generates critical networks with S≈1, although their attractor landscape (shown in Fig. 9A) has no information content whatsoever. Very remarkably, however, the networks produced in this way always exhibited random topologies with no hubs at all (see the inset in Fig. 9C). The networks developed hubs only when the α-fitness was used (together with the ACC and the AIC) and consequently the attractors exhibited genetic variability distributed around α = 0. 5, as in Fig. 4A. Since the evolved networks were selected to optimize the evolutionary trade-off, it is important to determine the robustness of their attractor landscapes under mutations. This robustness should be compared against the one observed in networks that are also critical, but that did not go through the evolutionary process. To measure such robustness, we removed one gene from the network and computed the probability that a percentage q of the existing attractors is conserved as a result of this mutation. (We also implemented other types of mutations, such as rewiring or removing some input or output connections of one gene, or changing its Boolean function, and the results are qualitatively similar.) Each gene in the network and each network in the population was subjected to such a deletion mutation and analysis of its consequence. It should be mentioned that the networks in the final population had between 100 and 500 attractors (most likely the total number of attractors per network was higher but we worked with no more than 500 attractors per network—see the Methods section for a description about the search of new attractors). We also computed for critical networks of the same size (N = 100) as the evolved, but that were constructed de novo to be critical, namely, networks that were constructed with an initial sensitivity and did not undergo any selection process. Fig. 10 shows the probability for the evolved networks (panel A) and the de novo networks (panel B). Note that following deletion of one gene, the de novo critical networks either conserve the entire attractor landscape (), or none of the existing attractors is conserved (). There are almost no other choices for these networks because for intermediate values of q between 0 and 100. In contrast, the evolved critical networks do not exhibit such all-or-none behavior. Instead, with a very high probability all the existing attractors in the evolved networks are conserved (), whereas for q<100 the probability, although small, was appreciably larger than zero. Remarkably, it never happened in our simulations that all the attractors of the evolved networks changed after the deletion of one gene, as it is apparent from Fig. 10A, i. e. , . This last result indicates that, under the deletion of one gene, the evolved networks change only one fraction of their attractors but not all of them and that most likely, they will not change anything. The above behavior epitomizes mutational robustness and is consistent with knockout experiments in many organisms, which reveal that the knockout (or mutation) of one gene (node of the network) most of the time does not cause gross phenotype change. Dynamical (phenotypic) robustness, the return to attractor states following perturbations of gene activities, and flexibility, the ability to switch between attractors, are two central properties common to all living organisms. While apparently opposed to each other, they jointly guarantee developmental robustness and homeostasis while allowing for developmental change and physiological adaptation to (i. e. intra-individual coping with) environmental fluctuations within the lifetime of an individual. Networks that are dynamically in the critical phase are poised between such phenotypic robustness and flexibility and have been shown to exhibit maximal information diversity to cope with changing environments [27], [34]. Here we show that the ontogenetically important coexistence of dynamical robustness and flexibility (the developmental trade-off) is related to an analogous balance between the opposing phenomena at the phylogenetic time scale: mutational robustness (preservation of attractor landscape following mutational network rewiring) and innovation (expansion of the attractor landscape). The selection for these two properties using the attractor conservation (ACC) and innovation (AIC) criteria as biologically plausible fitness filters in simulated network evolution led to networks whose structural properties and Boolean functions dictated a dynamically critical behavior (Fig. 3). We should note that for the AIC in this evolutionary scheme, innovation of phenotypes occurs in two distinct ways. On the one hand, the generation of new attractors can be considered as the emergence of new phenotypes. On the other hand, the addition of new genes to the network also adds new information to the already existing attractors by modifying the attractor states and their basins. In either case, for this information to be useful and contribute to an organism' s discriminatory response to variable environments, the new genes must have an activity that changes from one attractor to another. Therefore, a third ad hoc biologically motivated selection constraint we used was that the average variability of the genes in the attractor landscape must be significantly different from 0. Although this constraint is biologically important, it is not required for the evolution towards criticality because one can construct critical networks whose attractors have a nearly zero genetic variability (Fig. 9). However, only when the genetic variability of the attractor landscape was appreciably different from 0 did the networks evolve structures with global regulators (hubs). Indeed, based on our numerical simulations we can assert that the hubs emerge with high probability whenever the networks are forced to conserve attractors with high genetic variability, namely, with high information content. At this point this provocative statement is an empirical observation that we have not been able to fully quantify and deserves much more study. In any case, our results indicate that the emergence of hubs throughout the evolutionary processes is a consequence of the constraints imposed on the network dynamics and not of the structure of the common ancestor network. This is consistent with studies carried out for other types of networks in which the dynamical constraints strongly determine the network architecture [56], [57]. It is important to mention that we have not been able to identify any special structural property of the initial nodes of the common ancestor network that can predict which particular nodes will eventually become global regulators in the final networks. In fact, when we explicitly provided some of the initial nodes with a special property, such as with high output connectivity or a special type of Boolean function, that property was lost through the evolutionary process. We should also note that the mutation algorithm alone does not generate hubs either. For networks that undergo mutations and gene duplication events but without selection do not acquire this topological feature (see Fig. S3). Another remarkable property of the global regulators was the high persistence of their links across the networks in the population. Although the attractor landscape was the same for all the networks in the final population (as they stem from the same common ancestor), there were structural differences between them (Fig. 7). These genotypic differences are reflected in the fact that some regulatory interactions (links) between pairs of genes are present in some networks but not in others. However, strikingly, the regulatory links invariably present in all networks of the population mostly belong to the global regulators, as Fig. 7D shows. This strongly suggests that these global regulators play a fundamental role in maintaining the phenotypic traits (attractors) across the population in spite of the small differences in network structure, and may be one of the reasons why this type of topology has been developed in real genetic networks. The importance of the hubs to maintain the phenotypic landscape across the population, as revealed in Fig. 7, is not a trivial result. For it has been shown that very often the hubs are not the key elements that influence dynamics of the network [58]. It is worthwhile calling into attention that in our simulations the attractor conservation and innovation criteria are not as stringent as one may think. The reason is that, due to computer limitations, the attractor landscape can be evaluated in full only for small networks of (e. g.). Thus, we completely determined the attractor landscape of all networks in the population only for the first generation. Thereafter, identification of new attractors was achieved by sampling a small fraction of the state space (at most states for each network). Obviously we can apply the ACC and the AIC only to the attractors identified in such sampling and “hidden” attractors may exist that have been destroyed or created by mutation. In our simulations we worked with a maximum of 500 attractors per network. However, a more thorough search revealed that at the end of the evolutionary process, the evolved networks can have more than or even attractors (see Fig. S4). Thus, apparently we applied our selection criteria ACC and AIC to a small fraction of the attractor landscape, underestimating innovation and overestimating conservation. Quite remarkably, this was enough to generate criticality, robustness and hub-like structures. The under-sampling of the state space in our numerical simulations has a biological equivalent because in reality selection does not act on all attractors which represent potentially realizable cellular phenotypes but rather, on the effectively existing phenotypes. For instance, for an organism like E. coli, with regulatory genes, it is very unlikely that all the possible gene expression configurations have been explored by evolution – consistent with the notion that evolution is a quasi-non-ergodic process. Most likely, the search of new phenotypes (attractors) occurs by perturbing the already existing and occupied attractors. Thus the search is conducted in their state space neighborhood, which precisely reflects the algorithm we used to find new attractors (as described in detail in the Methods section). The idea that novel attractors must be reachable from existing ones that are already occupied by cells has wide-reaching consequence for the evolution of multi-cellularity and development [59]. Our approach does not study the evolution of evolvability per se but complements several studies of this question that use gene network-based computational models because we reverse the question: First, we do not impose an artificial “optimal“ phenotype (such as an arbitrary “equilibrium state” or attractor which networks are selected to maintain or evolve towards). Much to the contrary, the critical dynamics of our resulting networks was not an explicit selection criterion but is an independently known property of some networks that exhibit naturally high fitness. Second, we instead selected directly for properties related to evolvability, namely conservation and innovation of attractors, of which the former is of course directly related to phenotypic robustness. By not selecting for a particular phenotype through an artificially defined fitness value (as in [14]–[19]), we avoid exposing mutational robustness that simply reflects the known convergent mapping of many distinct network structures into one same phenotype (“equilibrium state”). Third, our selection criteria introduce the notion of global dynamics, embodied by the multi-attractor landscape as phenotype which, in contrast to the use of a single expression pattern as target phenotype, captures phenotypic adaptability and versatility of an organism. Quite interestingly, the critical networks produced by our evolutionary algorithm exhibit considerably higher mutational robustness than critical networks constructed de novo. Indeed, there is a high probability for the latter to change all their attractors after a simple mutation, whereas for the former this never happens (Fig. 4). In conclusion, we do not present here a “molecular” mechanism, based on particular topological structures or mutations, to generate critical networks with global regulators. Instead, we propose a “dynamical” mechanism based on the conservation, innovation and information content of the attractor landscape. Our results show that dynamical criticality, a central property for the functioning of a living organism, naturally emerges as a consequence of evolution that favors evolvability. In other words, such an evolutionary process is sufficient for and robust in producing dynamical criticality. In our model such criticality appears as a coextensive property of evolvability and is not a direct adaptive phenotype. Whether evolvability in terms of our criteria ACC and AIC is necessary remains open. Specifically, we cannot exclude that in another evolution scenario there could be an adaptive component in the evolution of criticality, i. e. natural selection may directly favor networks with. Then, the resulting networks could be associated with evolvability, in which case evolvability would be coextensive to a selected dynamical network property, i. e. a consequence of it rather than a direct result of selection, as proposed in other network models. The mutual relationship between dynamical criticality and evolvability remains thus to be evaluated carefully. But because evolution of multi-attractor dynamics and evolvability by network growth produces criticality, and since experimental evidence that existing gene networks are dynamically critical continues to accumulate, it is very likely that organisms that have evolved under the inevitable constraints of evolvability became critical. To describe the mutagenic algorithm, in what follows we denote as the gene that has been chosen for mutation and as its set of regulators whose activities {0,1} define the “input configurations” (or “entries”) to each of which the output value of is assigned: . Each gene in the network has two parts, a regulatory region and a coding region, as described and illustrated in Fig. 1. The regulatory region contains the binding sites for the regulators. Point mutations can be divided into two types, depending on whether they affect either the regulatory or the coding region of a gene, leading to distinct effects on the network and will be dealt with separately below. In the initial population, each gene has only one binding site (BS) for each of its upstream regulators. However, this situation changes through generations since new BS can be added, or some existing BS can be removed. The addition or removal of BS in the regulatory region is a mutation process that can be subdivided into the six types of mutation described below and illustrated in Fig. 11. We implement gene duplication followed by divergence by randomly choosing one gene in the network and duplicating it. Let be the gene chosen for duplication. A duplication event increases the number of genes in the network from to, and is the duplicated copy. The duplication can be performed in two different ways. In the simulations, the entire attractor landscape was known for all the networks in the initial population only. This is possible because the initial networks are relatively small () and the state space can be completely evaluated to find all the attractors. However, as the networks increase in size eventually only a random sample representing a small fraction of the state space can be evaluated. The search for new attractors was conducted only immediately after the gene duplication events, every 2000 generations. To find new attractors we dynamically explored the neighboring states of the attractors that we already had identified. This was achieved as follows: We set the network state to a state in one of its attractors. Then, we randomly perturbed 10% of the genes by bit-flips (change of activity values of the genes in that state) and evaluated the relaxation dynamics from that “perturbed” state for 60 time steps followed by evaluation of whether or not a new attractor was found. We performed this perturbation-based attractor search procedure 20 times for each network state in each of the attractors. When a maximum of new attractors were found for a given network, the search was stopped for that network and the new attractors were incorporated in the expanded attractor landscape of that network. The attractor search was then continued with the next network in the population. For the results presented here we used, but similar results are obtained for ranging from up to. The reason for stopping the attractor search (for a given network) when at most 10 new attractors were found was to keep the computing time within reasonable limits (the whole evolutionary process for a population of 1000 networks took on average 1. 5 weeks). However, in addition to those attractors that we found and included in the attractor landscape of the networks, many more attractors were created after the duplication events. To obtain an idea of how many attractors were left out of the analysis of the evolving attractor landscapes, at the end of the simulation we performed a more exhaustive search by randomly sampling 106 initial states in networks randomly chosen from the final population. Fig. S4 shows the number of attractors that were discovered in this search as a function of the number of sampled initial states. Surprisingly, in sampling 106 initial states almost 150000 attractors were been found within only one network! ! This number was much larger than the <200 attractors used to evolve these networks under the Attractor Conservation and Attractor Innovation criteria. It is remarkable that evaluating a small fraction of the attractor landscape (through the ACC and ACI) was sufficient to produce robust critical networks with global regulators. Note that this “agglomerative search” for new nearby attractor through single bit flip perturbations of existing attractors automatically detects “dynamically accessible” attractors – precisely as evolution of new phenotypes would have occurred that has to ensure that the latter are developmentally realizable [59].

Dynamically critical systems are those which operate at the border of a phase transition between two behavioral regimes often present in complex systems: order and disorder. Critical systems exhibit remarkable properties such as fast information processing, collective response to perturbations or the ability to integrate a wide range of external stimuli without saturation. Recent evidence indicates that the genetic networks of living cells are dynamically critical. This has far reaching consequences, for it is at criticality that living organisms can tolerate a wide range of external fluctuations without changing the functionality of their phenotypes. Therefore, it is necessary to know how genetic criticality emerged through evolution. Here we show that dynamical criticality naturally emerges from the delicate balance between two fundamental forces of natural selection that make organisms evolve: (i) the existing phenotypes must be resilient to random mutations, and (ii) new phenotypes must emerge for the organisms to adapt to new environmental challenges. The joint effect of these two forces, which are essential for evolvability, is sufficient in our computational models to generate populations of genetic networks operating at criticality. Thus, natural selection acting as a tinkerer of evolvable systems naturally generates critical dynamics.

Abstract Introduction Results Discussion Methods

systems biology evolutionary modeling genetics regulatory networks biology computational biology gene networks genetics and genomics

2012

Criticality Is an Emergent Property of Genetic Networks that Exhibit Evolvability

Non-structural protein 1 (NS1) is one of the most enigmatic proteins of the Dengue virus (DENV), playing distinct functions in immune evasion, pathogenesis and viral replication. The recently reported crystal structure of DENV NS1 revealed its peculiar three-dimensional fold; however, detailed information on NS1 function at different steps of the viral replication cycle is still missing. By using the recently reported crystal structure, as well as amino acid sequence conservation, as a guide for a comprehensive site-directed mutagenesis study, we discovered that in addition to being essential for RNA replication, DENV NS1 is also critically required for the production of infectious virus particles. Taking advantage of a trans-complementation approach based on fully functional epitope-tagged NS1 variants, we identified previously unreported interactions between NS1 and the structural proteins Envelope (E) and precursor Membrane (prM). Interestingly, coimmunoprecipitation revealed an additional association with capsid, arguing that NS1 interacts via the structural glycoproteins with DENV particles. Results obtained with mutations residing either in the NS1 Wing domain or in the β-ladder domain suggest that NS1 might have two distinct functions in the assembly of DENV particles. By using a trans-complementation approach with a C-terminally KDEL-tagged ER-resident NS1, we demonstrate that the secretion of NS1 is dispensable for both RNA replication and infectious particle production. In conclusion, our results provide an extensive genetic map of NS1 determinants essential for viral RNA replication and identify a novel role of NS1 in virion production that is mediated via interaction with the structural proteins. These studies extend the list of NS1 functions and argue for a central role in coordinating replication and assembly/release of infectious DENV particles. Dengue is the most prevalent arthropod-borne viral disease affecting around 400 million people worldwide and causing around 25,000 deaths per year [1]. Dengue virus (DENV) infections can lead to a wide range of clinical manifestations, ranging from asymptomatic to life-threatening dengue hemorrhagic fever and shock syndrome. However, in spite of its high medical relevance, no prophylactic vaccines or antiviral therapies are currently available and therefore a better understanding of the flavivirus life cycle is essential to promote the development of effective therapeutic regimens. DENV has a single stranded RNA genome of positive polarity, encoding for a polyprotein that is co- and post-translationally processed into three structural proteins (capsid, prM, and envelope) and seven nonstructural proteins (NS1-NS2A-NS2B-NS3-NS4A-NS4B-NS5) [2]. After viral entry and release of the genomic RNA into the cytoplasm of infected cells, newly synthesized viral proteins induce massive remodeling of intracellular membranes, creating distinct intracellular structures where viral RNA replication and virion assembly take place [3,4]. Nucleocapsid formation, thought to occur in close proximity to replication sites, is likely accompanied by acquisition of a lipid envelope via budding into endoplasmic reticulum (ER) membranes enriched in the envelope protein E and prM [5,6], through as yet undefined mechanisms. Assembled virions, stored within ER stacks in highly ordered arrays, are then released from the cell via the conventional secretory pathway, where cleavage of the prM protein by furin, a protease residing in the trans-Golgi network (TGN), renders the viral particles infectious. Flavivirus NS1 is a multifunctional 48-kDa glycoprotein that is translocated into the ER lumen co-translationally. Within the ER, NS1 promptly dimerizes upon addition of high-mannose carbohydrates [7], and is targeted to three destinations: the viral replication sites, the plasma membrane and the extracellular compartment. The majority of secreted NS1 is a soluble, proteolipid particle forming an open-barrel hexameric shell with a central channel occupied by lipids [8]. The three-dimensional high-resolution structure of the DENV NS1 dimer was recently solved by X-ray crystallography [9], providing valuable insights into the complex NS1 fold (Fig 1). The dimer contains three domains: first, a small β-roll domain formed by two intertwined β-hairpins; second, a Wing domain, composed of an α/β subdomain and a discontinuous connector that sits against the β-roll; third, a β-ladder domain, formed by 18 antiparallel β-strands (9 contributed by each monomer) assembled in a continuous β-sheet that runs along the whole length of the dimer (Fig 1A, left panel). The protrusion created by the β-roll and the connector subdomain renders one side of the dimer hydrophobic, and has been proposed to face the ER membrane and to interact with other transmembrane viral proteins [9,10]. Conversely, within the NS1 hexamer, the β-roll faces the interior of the lipoparticle, where it associates with the central lipid core (Fig 1A, right panel). On the opposite side of the β-roll, both in the dimeric and the hexameric form, the distal tips of the β-ladder and the Wing domain loops point outward, and are therefore exposed to the solvent. Secreted NS1 as well as NS1 residing on the plasma membrane and within cells, plays important roles in immune evasion via binding to complement proteins and modifying or antagonizing their functions [11–14]. Besides its immune evasive functions, NS1 modulates early events in viral RNA replication, was shown to co-localize with double strand RNA (dsRNA) and to interact with NS4B [10,15–17]. Indeed, deletion of NS1 from the viral genome completely abrogates replication, but ectopic expression of NS1 in trans can efficiently rescue NS1-deleted (ΔNS1) viruses [18–21]. Because of its essential role early in RNA replication, genetic studies have thus far provided limited information on the molecular determinants of NS1 responsible for the viral replication cycle and did not investigate possible functions of the protein for assembly and release of infectious virus particles. By using a combination of genetic, high-resolution imaging and biochemical approaches we discovered a novel role of NS1 for the production of infectious DENV particles that is linked to NS1 interaction with the structural proteins, but independent from NS1 secretion. Sequence analysis and visual inspection of the recently solved three-dimensional crystal structure [9] of NS1 were performed to assess the degree of conservation of amino acid residues and to identify the most relevant positions to be targeted by site-directed mutagenesis (Fig 1B). Based on their distribution within the NS1 dimer and their relative conservation across the Flavivirus genus, we selected 46 residues for alanine scanning mutagenesis, including five invariant cysteine residues (C4, C55, C179, C291, C312), recently shown to be engaged in disulfide bonds and playing an essential role in stabilizing the protein fold [9,22]. To dissect the impact of each individual mutation on the different steps of the viral replication cycle, we assessed viral RNA replication and virus spread by taking advantage of a DVR2A luciferase reporter virus genome (Fig 2A). VeroE6 cells were electroporated with in vitro transcripts of wild-type (WT) or a given NS1 mutant and viral replication was assessed by luciferase activity 24,48,72,96 and 120 h later (Fig 2B). Additionally, a replication-deficient NS5 mutant (GND) with a lethal mutation affecting the RNA-dependent RNA polymerase activity was included as negative control. Based on the replication phenotypes, half of the NS1 mutants displayed only minor defects or replicated comparably to WT (Table 1). Conversely, 23 mutations, including those affecting cysteine residues engaged in disulfide bonds, severely or completely blocked viral RNA replication (Fig 2B and Table 1, underlined mutants). Most of these mutations clustered on the core of the Wing and β-ladder domain, affecting residues that point towards the β-roll (S1 Fig), which has been proposed to face the ER membrane [9]. Interestingly, in close proximity to the previously reported di-amino acid motif (N10-K11) suggested to mediate interaction with NS4B and association with ER membranes [10], a mutation targeting W8 within the NS1 β-roll domain completely abrogated viral RNA replication. Similarly, mutations within the greasy finger loop of the β-ladder, namely Y158A and G161A, resulted in a lethal phenotype as already shown for alanine substitutions at residues G159 and F160 [9]. Altogether, these results provide a comprehensive map of molecular determinants within NS1 essential for viral RNA replication, highlighting an important role for selected residues of the β-roll and β-ladder domains, clustering on hydrophobic protrusions within the NS1 dimer structure (S1 Fig). To determine the impact of each mutation on the production of infectious virus particles, culture supernatants of transfected cells (Fig 2B) were harvested 72 h after transfection and used to infect naïve VeroE6 cells. Virus production was determined by luciferase assay 48 h later (Fig 3A). This experiment revealed a group of mutations (S114A, W115A, D180A, T301A) with minor effects on RNA replication, but massive impairment of virus production (up to ~2. 5 Log10 reduction compared to WT) (Fig 3B). Noteworthy, a mutation targeting T117 within the unresolved stretch of the Wing domain and in close proximity to S114 and W115, slightly enhanced particle production. Altogether, these results suggest a previously undiscovered role of NS1 for the production of infectious DENV particles. Next we wanted to corroborate this observation and rule out that impaired virus production was an indirect consequence of diminished replication fitness rather than a specific defect in assembly or release of viral progeny. To this end, we assessed the impact of these mutations on RNA replication in the context of a sub-genomic reporter replicon (sgDVR2A) that does not support virus production, thus measuring replication independent from a possible contribution of virus spread (Fig 4A). In this and all subsequent analyses we focused on mutants with selective alterations of virus production (S114A, W115A, D180A, T301A; Fig 3B) in order to avoid possible indirect effects resulting from impaired replication fitness (Fig 3B). Therefore, NS1 mutants with strong replication defects were excluded (Table 1). Moreover, since several NS1 mutants with a defect in virus production were slightly impaired in RNA replication we included as control the NS1 mutation R314A that caused minimal defect in replication, but did not affect virus production (Table 1). Furthermore, mutant T117A was included in the analysis because of its increased capacity to produce infectious DENV particles and its close proximity to some of the sites where mutations caused a selective reduction of virus production. VeroE6 cells were transfected with in vitro transcripts of the replicon constructs and replication was measured 24,48 and 72 h later (Fig 4A). While at early times post transfection we observed a moderate reduction in luciferase activity for S114A, W115A, D180A and T301A compared to WT (2 to 5-fold; 24 h. p. t.), all mutants replicated comparably at 48 and 72 h. p. t. , arguing for a minor contribution of RNA replication to the observed reduction in infectious particle production. As already observed in the context of the full-length reporter virus, the T117A mutant exhibited a replication profile comparable to WT also within the subgenomic sgDVR2A replicon whereas R314A exhibited an overall decrease in replication fitness, with a 50% to 75% reduction at any time point. Collectively, these results demonstrate a selective defect for a sub-group of NS1 mutants in assembly and/or release of virus particles. To investigate further the phenotype of the NS1 mutants with respect to the production of infectious intra- and extracellular virus particles, and to corroborate these observations in human cells, we determined the infectivity profiles of each mutant in the context of a full-length DENV genome transfected into human hepatoma Huh7 cells. Three days post-transfection, titers of infectious virus released into the culture media or contained in cells were determined by limiting dilution assay with naïve cells. The results shown in Fig 4B demonstrate that the amounts of intra- and extracellular infectivity were altered in all mutants, albeit to very different degrees. In agreement with the results obtained with the reporter virus genome, we found that alanine substitutions at residues S114, W115, D180 and T301 reduced extracellular infectivity titers up to 100-fold, confirming an essential role of these amino acid residues in NS1 for particle production. Interestingly, the amounts of intracellular virus particles were also reduced 5- to 10-fold. While this impairment argued for a defect of the NS1 mutants in virus assembly or maturation, the higher reduction of extracellular virus titers suggested an additional effect on particle release as inferred from the ratio of intra- to extracellular virus titers and comparison with the WT (Fig 4C). The R314A mutation reduced the virus titer only ~3. 5-fold and did not affect the ratio of intra- to extracellular infectivity, consistent with a subtle effect on assembly or virus maturation (Fig 4C). Interestingly, the T117A mutant produced 12-fold more intracellular virus than WT, concomitant with a ~5-fold higher titer of extracellular virus particles, indicating accelerated assembly or infectivity maturation and reduced virus particle release. In conclusion, these results suggest that alanine substitutions at position S114, W115, T117, D180 or T301 of NS1 alter the production of infectious virus, supporting the notion that NS1 is a critical determinant for assembly or release of infectious virus particles. NS1 accumulates in extracellular fluids as a homo-hexamer with a lipidic core [8,23,24] and besides its immune evasive functions [11–13] was shown to enhance virus attachment upon entry [25]. In addition, some studies hypothesized a link between NS1 secretion and virus assembly or release [26]; however the lack of genetic tools in those days allowing the selective block of NS1 secretion in the context of a complete replication cycle precluded any functional investigation. Prior studies have shown that YFV, KUNV and WNV mutants lacking NS1 do not replicate, but can be rescued when NS1 is complemented in trans by ectopic expression of the full-length protein [17,18,21,27]. To elucidate the possible role (s) of NS1 secretion in the DENV replication cycle, we utilized a similar approach and generated a DENV genome containing a 97 amino acids in-frame deletion within the NS1 gene (DVR2AΔNS1). This mutant retained the N-terminal 156 and the C-terminal 99 residues, respectively (Fig 5A, left panel). In parallel, we engineered a set of helper VeroE6 cell lines, constitutively expressing different NS1 variants after lentiviral transduction of expression vectors containing the complete NS1 coding region (NS1WT) or the empty pWPI vector (CTRL) that served as positive and negative controls, respectively. Furthermore, we engineered C-terminally tagged variants carrying a HA-affinity epitope (NS1HA) or the well-described KDEL motif (NS1KDEL), responsible for retrieval of ER luminal proteins from the Golgi apparatus by retrograde transport (Fig 5A, right panel). Correct protein expression and secretion of each NS1 variant was confirmed by western-blotting (Fig 5B). As expected, NS1HA had a slower electrophoretic mobility than the WT. Furthermore, both NS1WT and NS1HA were readily detected in the culture media, while NS1KDEL was effectively retained in the ER. Next, we assessed rescue of viral RNA replication and particle production by the NS1 variants by using transfection of the DVR2AΔNS1 genome into each helper cell line. As shown in Fig 5C, the ΔNS1 genome was able to replicate in NS1WT, NS1HA and NS1KDEL helper cells, while no luciferase activity could be detected in CTRL cells lacking NS1. Of note, when each naïve helper cell line was infected with virus-containing culture fluids harvested 72 h. p. t, comparable levels of luciferase activity were detected in all conditions, indicating that C-terminally HA-tagged NS1 is fully functional (Fig 5D). Importantly, the rescue of particle production by NS1KDEL shows that secretion of NS1 is dispensable for infectious DENV particle production. To address the relative efficiency of trans-complementation upon virus infection, NS1WT-, NS1HA- and NS1KDEL-expressing VeroE6 cells were infected with trans-complemented DVR2AΔNS1 particles (ΔNS1TCP), produced in VeroE6_NS1WT cells. Culture fluids were harvested 24,48 and 72 h later and the amounts of produced particles were determined by focus-forming unit (FFU) assay on NS1WT cells (Fig 6A). Consistent with the luciferase assay data, ΔNS1TCP did not produce infectious virus on CTRL cells that do not express NS1, confirming that this mutant fails to replicate in the absence of NS1. Titers of infectious ΔNS1TCP particles released from VeroE6_NS1WT, NS1HA or NS1KDEL cells were higher than DVR2A wild-type infection on CTRL cells at early times post-infection (24 and 48 h p. i.), with no appreciable differences observed at later time points (72 h p. i.) (Fig 6B). Interestingly, while none of the trans-complemented TCPs gave rise to clearly visible plaques as previously reported [17], DENV-containing foci detected by immunostaining were larger than those produced by the full-length wild-type virus (Fig 6B, lower panel). Additionally, lysates and culture supernatants of infected cells were harvested 24,48 and 72 h p. i. to evaluate protein expression and secretion upon ΔNS1TCP infection. Under these conditions, virus replication in the various helper cell lines was comparable as judged by the intracellular protein levels of NS5 (Fig 6C; “Intra“). Furthermore, secretion profiles of HA- WT- and KDEL-tagged NS1 resembled those of uninfected cells, with the latter being efficiently retained intracellularly also upon DENV infection (Fig 6C; “Extra“). To unequivocally confirm that the ΔNS1TCP system fully recapitulates wild-type DENV infections, we additionally investigated the ultrastructural morphology of NS1HA helper cells upon infection with DVR2AΔNS1 TCPs by using transmission electron microscopy. In agreement with the replication and infectious particle production data, NS1HA cells infected with DVR2AΔNS1 TCPs contained the characteristic membrane invaginations which have been proposed to represent viral replication factories (vRFs) [5,6] and electron-dense virus particles forming regular arrays within the ER (Fig 6D). Moreover, a large number of virus particles were observed on the plasma membrane or accumulating within the extracellular space between adjacent cells (Fig 6E). These structures were absent in both uninfected NS1HA cells and ΔNS1TCP-infected CTRL cells (S2 Fig). In conclusion, these results demonstrate full functionality of HA-tagged and ER-retained NS1 supporting both DENV RNA replication and production of infectious virus particles. Since NS1 secretion appeared functionally unlinked to infectious particle production, we next hypothesized that NS1 function (s) required for the late steps of the viral replication cycle might involve interactions between NS1 and the structural DENV proteins. To address this hypothesis we took advantage of our ΔNS1TCP system using HA-tagged NS1 for trans-complementation. DVR2AΔNS1 TCP stocks produced and titered in NS1WT helper cells were used to infect NS1HA target cells at an MOI of 1. Forty-eight hours later, cell lysates were subjected to HA-affinity capture using anti-HA agarose beads, and purified NS1 protein complexes or whole cell lysates were analyzed by western-blot using C-, prM-, E- and NS5-specific antibodies. Specificity of western-blot and immunoprecipitation analysis was monitored by including VeroE6_CTRL and VeroE6_NS1WT cells, respectively. As shown in Fig 7A, upon infection of NS1 helper cells with ΔNS1TCP, comparable amounts of structural proteins accumulated in both NS1WT and NS1HA cells, confirming that DVR2AΔNS1 replicates efficiently in both cell lines. Most interestingly, upon HA-immunoprecipitation, all three structural proteins (E, prM and C) specifically co-precipitated with NS1HA arguing for an interaction between NS1 and DENV virions and possibly also subviral particles. In spite of comparable protein amounts in the cell lysates, no specific signal was detected in case of the non-tagged NS1WT or for the NS5 protein, confirming specificity of the NS1HA-immunoprecipitation. To corroborate the interaction between NS1 and the structural proteins, we performed analogous pull-down experiments using cells that had been infected with wild-type DENV. VeroE6 cells were mock-infected or infected with DENV-2 at an MOI of 1 and 48 hours later cell lysates were subjected to immunoprecipitation using a rabbit pre-immune serum (PIS) or an NS1-specific polyclonal antiserum (Fig 7B). Interestingly, also under these conditions E and prM were specifically co-immunoprecipitated with NS1 whereas C was not detected. The absence of C might be due to the overall lower efficiency of this immunocapture approach, to suboptimal conditions for antibody binding or to an altered ratio of subviral particles to infectious virions in the TCP system as compared to wild-type virus-infected cells (see discussion). Nevertheless, the interaction between E and NS1 was confirmed in a reciprocal approach by which the NS1 protein in DENV-2 infected cells could be specifically co-immunoprecipitated with envelope (S3 Fig). Altogether, these results support the notion that NS1 interacts with the viral envelope glycoproteins. Based on the results described above, we hypothesized that the selective defect in infectious particle production observed for some of the NS1 point mutants was due to an altered association with the envelope glycoproteins. To investigate this hypothesis, we analyzed the association of selected NS1 mutants with the structural proteins by co-immunoprecipitation in the DVR2AΔNS1 TCP system, because it allowed highly efficient pull-down of NS1. We engineered stable VeroE6 cell lines expressing HA-tagged forms of the NS1 mutants S114A, W115A, T117A, D180A, T301A and R314A and infected these cell lines with ΔNS1TCP at an MOI of 1. Samples harvested 48 h. p. i. , together with positive and negative controls, were subjected to HA-specific pull-down and cell lysates, immunocomplexes or culture supernatants were analyzed as described above. All cell lines expressed HA-tagged NS1 variants and despite small variations in the expression levels rescued DVR2AΔNS1 replication to similar extents as judged by the abundance of E, prM and C (Fig 8A) and the luciferase activity in the cell lysates (S5 Fig). Interestingly, analyses of NS1HA-immunocomplexes revealed marked differences in the interaction profiles of these mutants with the structural proteins. In case of NS1 mutants S114A and W115A, a significant reduction in E and prM co-precipitation, concomitant with a loss of the C-specific signal was observed, arguing for impaired association of these NS1 variants with DENV virions (Fig 8B; reduction of Envelope: ~7. 5- and 3. 8-fold, respectively). In contrast, mutants D180A and T301A exhibited a selective loss of co-precipitated C while retaining high prM and E interaction, suggesting that NS1 contributes to virus production in an additional manner that is independent from interaction with the envelope glycoproteins. Consistent with their higher competence in supporting infectious particle production, T117A and R314A were still able to bind all three structural proteins, although the C-specific signal was lower as compared to wild-type. These results were further corroborated by co-localization analyses of HA-tagged NS1 mutants with the envelope glycoproteins revealing reduced co-localization in case of the two mutants that were impaired in virus production (S114A and W115A; S6 Fig), but no significant change in case of all the other mutants. Importantly, no major differences could be observed with respect to NS1 secretion into the culture supernatants (Fig 8C), strengthening the notion that NS1 secretion is functionally unlinked to its role in the production of infectious extracellular virus particles. To further investigate the interaction between NS1 and the structural proteins, we used immunofluorescence to visualize the sub-cellular colocalization of NS1 with the structural proteins. To allow detection of putative DENV assembly sites and/or assembled virus particles, we aimed to perform simultaneous immunostaining of capsid, envelope and NS1. Since we were limited by the availability of antibodies allowing for triple staining, we engineered a C-terminally mCherry-tagged NS1 variant (NS1mCherry) to be visualized without requirement for antibodies (Fig 9A). In the initial set of experiments, we confirmed correct NS1mCherry expression and sub-cellular distribution. Importantly, mCherry-tagged NS1 supported efficient DENV RNA replication and infectious particle production upon infection with ΔNS1TCP demonstrating full functionality of the fluorescently tagged NS1 (Fig 9B and 9C). Taking advantage of this approach, we analyzed infected cells by confocal microscopy and observed several discrete structures where Envelope, Capsid and NS1 colocalized (Fig 9D). Although the nature of these structures that might correspond to putative assembly sites or assembled virus particles is not clear, this colocalization provides additional evidence for a previously unreported association between NS1 and the structural DENV proteins. To overcome the inherent limitations of fluorescence microscopy in allocating specific proteins to distinct subcellular structures and to elucidate the nature of these NS1-positive structures we performed correlative light-electron microscopy (CLEM) of VeroE6_NS1mCherry cells infected with ΔNS1TCP (Fig 10). MCherry-fluorescent structures were allocated by confocal microscopy of fixed cells grown on photo-etched gridded coverslips (Fig 10A and 10B), which were subsequently processed for EM (Fig 10C). After alignment of confocal and electron microscopy images we were able to identify NS1-enriched structures. As observed with NS1HA, NS1mCherry-infected cells also contained the characteristic DENV-induced membrane rearrangements. Indeed, we found that highly fluorescent mCherry-positive areas corresponded to ER, vesicle packets (VPs) and ER-associated bags containing arrays of DENV particles (Fig 9D and 9E). These results are consistent with our previous studies describing the association of NS1 with VPs as determined by immuno-EM [5]. Most importantly, the present data further support our assumption that NS1 associates with assembled virus particles (Fig 10Db and 10Ed), in agreement with our results from immunoprecipitation and confocal microscopy experiments. Flavivirus NS1 has emerged as one of the most enigmatic proteins forming distinct intra- and extracellular complexes and contributing to pathogenesis as well as viral replication. While soluble and cell-surface–associated NS1 was shown to modulate complement activation pathways through interactions with host proteins, such as the regulatory protein factor H, complement factor C4 or clusterin [11–13,28] and to induce cross-reacting antibodies to human proteins [29–32], the role of NS1 in the viral replication cycle has so far been elusive. NS1 was initially thought to be involved in virus assembly or maturation, given its subcellular localization within the ER lumen and its secretion profile, largely mirroring that of the structural proteins prM and E [33,34]. However, this hypothesis was challenged by the co-localization of NS1 with dsRNA, a marker for RNA replication sites and by biochemical and genetic evidences supporting an essential role of NS1 in viral RNA replication [10,35–37]. Additionally, NS1 has been reported to interact with NS4B and/or NS4A, which are assumed to relay NS1 signals to other components of the viral replicase through their ER luminal segments. In the present study, we used a combination of genetic, biochemical, and imaging approaches to investigate the functional role of intra- and extracellular NS1 in the DENV replication cycle. In addition to the previously reported mutations within the N-linked glycosylation sites [35,38] and the cysteine residues engaged in disulfide bonds [39], we identified 18 additional residues that completely or severely reduced DENV replication. Interestingly, most of these mutations clustered on residues located towards the proposed ER binding site (S1 Fig). Among these, a conserved tryptophan residue at amino acid position 8 (W8A) was found to be essential for RNA replication. This residue is located in close proximity to the previously reported di-amino acid motif (N10K11) within the β-roll domain of WNV NS1 that is also critical for efficient RNA replication [10,40]. These residues are located in an exposed region of the NS1 dimer suggested to face the ER membrane and possibly mediating the interaction with NS4B. Moreover, within the NS1 hexamer, these residues point towards the inner cavity of the barrel and might contribute to the association of NS1 with its lipid cargo (S1B Fig) [9]. While these results are consistent with the proposed way how the NS1 dimer associates with intracellular membranes and provide a detailed genetic map of NS1 residues essential for viral RNA replication, further studies are required to decipher the impact of these mutations on protein-protein interactions and induction of ultra-structural membrane rearrangements. Most interestingly, our mutagenesis approach identified a group of NS1 mutants with selective defects in infectious particle production. These mutants (S114A, W115A, D180A and T301A) had only minor or negligible defects in NS1 protein stability and viral RNA replication as judged by their replication kinetics in the context of a sub-genomic replicon (Fig 4A), but released up to 100-fold lower amounts of infectious DENV particles than the wild-type (Figs 3A and 4B). Of note, amounts of intracellular infectivity were reduced ~5- to 10-fold, arguing that the NS1 mutations affected both assembly (as evidenced from intracellular virus titers) and release of infectious DENV particles (indicated by titer reduction in cell culture supernatants). Although we cannot precisely comment on the individual contribution of NS1 to virus assembly and particle release, the identification of an additional NS1 mutant (T117A) producing ~12-fold higher amounts of intracellular infectious virus strongly hints towards a primary role of NS1 in the assembly of infectious DENV particles. Based on previous studies reporting the ability of ectopically-expressed NS1 to trans-complement YFV or WNV NS1 deletion mutants [18,21], we established a DENV-based ΔNS1TCP system to characterize NS1 functions. Taking advantage of this method, we probed the capacity of an intracellularly retained KDEL-tagged NS1 (NS1KDEL) to support production and release of viral particles and demonstrate that secretion of NS1 is dispensable for these functions. Of note, KDEL-tagged proteins are shuttling between the ER and the Golgi apparatus and thus, NS1KDEL could still assist early vesicular traffic of viral particles, i. e. from virion budding sites (ER) to early secretory compartments (ERGIC) (Fig 11). Recently it has been reported that depletion of the KDEL receptor or a subset of class II Arf proteins reduces secretion of non-infectious sub-viral particles and that prM—KDEL receptor interaction plays a role in virus secretion, arguing that flavivirus release from infected cells is assisted by specific sorting mechanisms [41,42]. However, additional viral or host factors appear to be required to coordinate these processes, since knock-down of KDEL receptor or Arf4+5 expression reduced YFV and DENV1-3 secretion less than 10-fold and had no effect on DENV4 and WNV particle production. Further studies will be needed to elucidate the possible contribution of NS1 trafficking from the ER to early or intermediate secretory compartments for DENV particle release. By using HA-tagged NS1 variants (NS1HA) and ΔNS1 virus-infected cells, we identified a previously unreported interaction between NS1 and the envelope glycoproteins E and prM. Although more than two decades ago E—NS1 complexes were identified in insect cells infected with a selected group of flaviviruses, the interaction was suggested to result from unspecific protein aggregation, lacking functional significance and not reproducible in DENV-2- and YFV-infected cells [26]. In contrast, several lines of evidence suggest that NS1 interaction with these structural proteins is specific and of importance for the production of infectious DENV particles: (i) the co-immunoprecipitation of NS1 with prM and E and, possibly indirectly, with C; (ii) the identification of triple-positive structures enriched in NS1, E and C by confocal microscopy; (iii) the detection of NS1 in compartments containing assembled virions by CLEM. Importantly, immunoprecipitation experiments using lysates of wild-type DENV-2-infected cells and an NS1-specific antibody confirmed the interaction of NS1 with both prM and E glycoproteins. Of note, while in this experimental set-up no specific signal could be detected for the capsid protein, it was well co-precipitated in the DVR2AΔNS1 TCP system. This discrepancy might be due to the overall lower NS1 capture efficiency in case of wild-type virus-infected cells. Alternatively, it is possible that in wild-type virus-infected cells subviral particles (SVPs) are produced in higher abundance relative to virus particles, whereas in the DVR2AΔNS1 TCP system production of virus particles might be favored relative to SVPs. Consistent with this assumption we observed a much faster and more efficient production of infectious virus particles in our TCP system than with wild-type virus-infected cells (Fig 6B). Assuming that NS1 can also interact with prM/E present on the surface of SVPs, in case of their excess production we would still observe coprecipitation of NS1 with the envelope glycoproteins, whereas C would no longer be co-precipitated. In contrast, when virus particles are produced in excess over SVPs, we would observe NS1 co-precipitation with prM/E and C, i. e. virions. This is consistent with the observed colocalization of NS1 with C and E (Fig 9) and the enrichment of NS1 in areas containing fully assembled DENV particles (Fig 10). Moreover, the conclusion is consistent with the membrane topology of E, prM and NS1 that are located in the lumen of the ER, while capsid resides on the cytoplasmic side of ER membranes (Fig 11). Since no tangible interactions between capsid and the glycoproteins were reported to date, the most likely explanation for the apparent NS1-C co-precipitation is an interaction between NS1 and assembled virions. While the precise mechanism remains to be determined, the specificity and functional relevance of these interactions was corroborated by the interaction profiles of NS1 mutants having defects in infectious particle production. Mutations affecting highly conserved residues of the flexible solvent-exposed loop in the NS1 Wing domain (S114A and W115A; Fig 8D) simultaneously abrogated C, prM and E association arguing that this NS1 loop is engaged in interaction with DENV virions. In contrast, mutations affecting residues in the β-ladder domain (D180A and T301A) preserved glycoprotein binding, but prevented association with capsid arguing for a second function of NS1 in the assembly of DENV particles that is independent from envelope protein interaction. While the exact mechanism remains to be established, it is tantalizing to speculate that NS1 might, directly or indirectly via these residues, interact with other viral or cellular factor (s) required for the formation of infectious DENV particles. Alternatively, NS1 might assist membrane budding or conformational changes in prM/E required for the envelopment of nucleocapsids. This hypothesis is supported by the localization of “capsid non-binder” mutants in the NS1 dimer structure. In fact, D180 resides at the intersection of the Wing and β-ladder domain and T301 points towards the ER membrane and is solvent exposed (Fig 8D). These mutations might affect the NS1 dimer fold, its affinity for membranes or its membrane-bending ability, while preserving prM/E-association through the distal tips of the Wing domain that points towards the ER lumen. Alternatively, these mutations might induce conformational constraints in prM/E complexes, reducing their plasticity and capability to envelope budding nucleocapsids, affect the recruitment of prM/E to assembly sites or only indirectly affect association with capsid, as a consequence of altered interactions with other viral [43,44] or host factors playing critical roles in virion morphogenesis. However, ultrastructural analysis of NS1 mutants failed to identify assembly intermediates, the only striking difference to the wild-type being a marked reduction in the overall amount of electron-dense particles found in proximity to or directly at the plasma membrane (S2 Fig). These results suggest that nucleocapsid formation and envelopment are coupled or that naked nucleocapsids have an extremely short half-life. Further experiments will be required to dissect the exact role of NS1 in the assembly process of infectious virus particles. Assembly of flavivirus particles is a poorly understood process (reviewed in [45]). High-resolution imaging approaches have shed some light on the topological arrangement of viral RNA replication and virus particle assembly. The viral replicase machinery is assumed to reside in highly organized membranous structures, designated as vesicle packets (VPs). These structures are formed by ER invaginations containing pores that would allow the release of newly synthesized viral RNA to be used for packaging into virions [5,6]. While genetic evidence suggests the involvement of several non-structural proteins and host factors in flavivirus assembly [41,43,44,46–48], the underlying molecular mechanisms are not known. Besides the lack of tangible interactions between C and the prM/E complex, major limitations are posed by the difficulty to visualize assembly intermediates and the seemingly unspecific incorporation of genomic RNA into nucleocapsids [49,50]. Based on the results presented here, it is tempting to speculate that NS1 might assist virion morphogenesis via its lipid-remodeling activity [8,9], its affinity for membranes [9] and its ability to interact with both non-structural [10,15] and structural proteins. In this respect NS1 might provide essential lipids or recruit essential host factor required for the biogenesis of the replication complex, while coordinating the recruitment of E and prM to assembly sites juxtaposed to the viral replicase [5] (Fig 11). In conclusion, the present study provides a comprehensive genetic map of NS1 determinants important for viral RNA replication and identifies a novel role of NS1 for the production of infectious DENV particles. We demonstrate that NS1 interacts with the envelope glycoproteins presumably on the surface of virions and these interactions are required for efficient production of infectious virus particles. Given its multiple roles in counteracting host defense, promoting RNA replication and enhancing production of virus particles, NS1 exemplifies the genetic economy of flaviviruses and emerges as attractive target for antiviral drugs. The mouse monoclonal antibody recognizing human GAPDH (sc-47724/0411) was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). The mouse anti-Envelope monoclonal antibody (3H5-1) was purchased from ATCC. The mouse monoclonal antibody 6F3. 1 reacting with the capsid protein was a kind gift of Dr. John G. Aaskov (Queensland University of Technology, Australia); rabbit polyclonal serum anti-capsid was a kind gift of Dr. Andrea Gamarnik (Fundación Instituto Leloir, Argentina). The rabbit polyclonal antibodies recognizing Envelope, NS1, NS5 and prM were previously described [5]. Rabbit anti-HA antibody (Ab9110) and mouse anti-NS1 antibody (ab41623) were purchased from Abcam. The mouse monoclonal anti-HA (H3663) and anti-Flag (F1804) antibodies, agarose anti-HA conjugated beads and secondary anti-mouse and anti-rabbit horse-radish peroxidase-conjugated antibodies were purchased from Sigma-Aldrich (Sigma-Aldrich, Saint Louis, MO). Huh7 [51], HeLa [52], VeroE6 (ATCC #CRL-1586), and BHK-21 (ATCC #CCL-10) cells were maintained in Dulbecco' s modified Eagle medium (DMEM; Invitrogen, Karlsruhe, Germany) supplemented with 2 mM l-glutamine, nonessential amino acids, 100 U/ml penicillin, 100 μg/ml streptomycin and 10% fetal calf serum. VeroE6 stably expressing NS1WT or tagged derivatives were cultured in the presence of 10 μg/ml of puromycin. Samples were denatured in 2x protein sample buffer (200 mM Tris [pH 8. 8], 5 mM EDTA, 0. 1% Bromophenolblue, 10% sucrose, 3% SDS, 1 mM DTT) and incubated for 5 min at 95°C. Proteins were separated by SDS-polyacrylamide gel electrophoresis (PAGE) and transferred onto polyvinylidene difluorid membranes by using a MINI-SDS-PAGE wet-blotting apparatus (Bio-Rad, Munich, Germany). Membranes were blocked with 5% non-fat dry milk in PBS/0. 5% Tween-20 (PBST) and incubated with primary antibodies (capsid 1: 50; GAPDH 1: 1,000; HA 1: 1,000; E 1: 1,000; prM 1: 500; NS1 1: 500) by over-night incubation at 4°C or for 1 h at room temperature. After 3 washes with PBST, membranes were incubated with secondary horse radish peroxidase-conjugated antibodies, developed with the Western Lightning Plus-ECL reagent (Perkin Elmer; Waltham, MA) and bands were imaged using an Intas ChemoCam Imager 3. 2 (Intas, Göttingen). VeroE6 helper cells expressing different forms of NS1 were infected with ΔNS1TCP at an MOI of 1. Two days later, cells were fixed with 4% PFA for 10 min at room temperature, permeabilized with 0. 5% (vol/vol) Triton X-100 in PBS and aspecific biding sites blocked with PBS containing 5% FBS for 30 min at RT. For staining of NS1mCherry cells, rabbit anti-NS1 (S3 Fig), or rabbit anti-C and mouse anti-Envelope antibodies (Fig 8) were used in combination with goat anti-mouse Alexa 647-conjugated and donkey anti-rabbit Alexa488-conjugated secondary antibodies. For staining of NS1HA cells (wild-type and mutants) (S4 Fig), rabbit anti-HA and mouse anti-Envelope antibodies were used in combination with goat anti-mouse Alexa 568-conjugated and donkey anti-rabbit Alexa488-conjugated secondary antibodies. Nuclear DNA was stained with 4′, 6-diamidino-2-phenylindole (DAPI) (Molecular Probes, Karlsruhe, Germany). Coverslips were mounted in Fluoromount-G mounting medium (Southern Biotechnology Associates, Birmingham, AL). For 3D visualization of NS1mCherry, E and C, samples were imaged with an Ultraview ERS spinning disk (PerkinElmer Life Sciences) on a Nikon TE2000-E inverted confocal microscope using a Plan-Apochromat VC 100× objective (numeric aperture [NA], 1. 4). Optical sections of 0. 13 μm were acquired separately for each channel. Z-stacks were deconvolved with a theoretical point-spread function, and chromatic shifts between green and far-red dyes were corrected using Autoquant X3 software. 3D reconstructed images were created using the Imaris 8 software package. For colocalization analyses of HA-tagged NS1 mutants and envelope fluorescence signals, Pearson' s correlation coefficient was calculated on single plane images, by using the integrated function in Fiji (ImageJ). For determination of virus titers by limiting dilution assay, Huh7 target cells were seeded into 96-well plates (104 cells/well) the day before infection. Cells were inoculated with serial dilutions of virus-containing supernatants that had been filtered through a 0. 45-μm-pore-size filter. Infected cells were detected by immune staining of the E protein using the mouse anti-E antibody (3H5-1; diluted 1: 500) and secondary horseradish peroxidase-conjugated antibody (1: 200). Virus titers (expressed as 50% tissue culture infective dose [TCID50]/ml) were calculated as previously reported [53]. For determination of virus titers by Focus forming unit (FFU) assay, VeroE6 or VeroE6_NS1WT cells (2x105 cells/well) seeded into 24-well plates, were infected with serial dilutions of 0. 45-μm-filtered supernatants and incubated in the presence of 0. 8% methylcellulose for 5 days. Monolayers were rinsed twice in PBS, fixed with 5% PFA and permeabilized with 0. 2% (v/v) TritonX-100 in PBS for 15 min. Infected foci were detected by immune staining of the E protein using the mouse anti-E antibody (3H5-1; diluted 1: 1,000 in PBS) and secondary horseradish peroxidase-conjugated antibody (1: 200). Alternatively, virus titers were determined by standard plaque assay (PFU) on target VeroE6 cells as previously described [54]. To determine intracellular infectivity titers, transfected cells were disrupted by several freeze–thaw cycles as described earlier [55]. In brief, transfected Huh7 cells were extensively washed with PBS, scraped off the plate into PBS and centrifuged for 5 min at 700 × g. Cell pellets were resuspended in complete DMEM (containing 15 mM HEPES, pH 7. 2–7. 5) and subjected to three cycles of freezing and thawing by using liquid nitrogen and a thermo block set to 37°C. Cell debris was removed by centrifugation at 20,000 × g for 10 min at 4°C. Virus-containing culture supernatants from transfected cells were treated in the same way and infectivity was determined in parallel by limiting dilution assay as described above. VeroE6 cells were seeded into 15-cm2 dishes (7. 5x106 cells/dish). Twenty-four hours later, cell monolayers were either mock-infected or infected with DENV-2 (MOI = 1). Forty-eight hours later, cell monolayers were scraped into 1 ml lysis buffer (50 mM Tris-HCl [pH 8. 0], 0. 5% NP-40,150 mM NaCl and protease inhibitor cocktail (cOmplete, Roche) ). After 30 min incubation on ice, cell debris was removed by 15 min centrifugation at 13,800xg and 400 μl of clarified cell lysate was incubated with rabbit pre-immune serum (PIS) or rabbit anti-NS1 antiserum for 6 hours at 4°C in a head-to-head shaker. Samples were incubated with 40 μl of protein A beads slurry (Sigma Aldrich, St. Louis, USA) for 1 hour at 4°C, washed three times with 1 ml of lysis buffer and protein A-bound complexes were transferred into a fresh tube. After a final wash with 1 ml of lysis buffer for 15 min at 4°C, protein complexes were eluted at room-temperature by two consecutive steps with 75 μl of 0. 1 M glycine [pH 2. 5] for 5 min. Collected supernatants were immediately neutralized by adding 7. 5 μl 1 M Tris-HCl [pH 8] and denatured for 5 min at 95°C in the presence of 33 μl of 6X SDS sample buffer. Alternatively, mouse anti-HA or mouse anti-E antibodies were used in combination with protein G beads slurry (Sigma Aldrich, St. Louis, USA) and the immunoprecipitation was carried exactly as described above. For ΔNS1TCP experiments, VeroE6 cells stably expressing an empty pWPI vector (CTRL), or wild-type NS1 (NS1WT), or HA-tagged NS1 (NS1HA) or derivatives thereof were seeded into 10-cm2 dishes (3x106 cells/dish). Twenty-four hours later, cell monolayers were infected with DVR2AΔNS1 (MOI = 1) for 4 h at 37°C. Forty-eight hours post-infection, cell monolayers were scraped into 1 ml lysis buffer (50 mM Tris-HCl [pH 8. 0], 0. 5% NP-40,150 mM NaCl and protease inhibitor cocktail (cOmplete, Roche) as recommended by the manufacturer). After 30 min incubation on ice, cell debris was removed by 15 min centrifugation at 13,800xg. For HA-specific affinity capture, samples were incubated with HA-specific agarose beads (Sigma-Aldrich, St. Louis, USA) for 5 h by continuously inverting the tubes at 4°C. Beads were washed three times for 20 min with large volumes of lysis buffer at 4°C and samples were eluted at room-temperature in two consecutive steps with 3% SDS in PBS for 5 min and PBS for 5 min, respectively. The two eluates were pooled and precipitated over-night at -20°C with 4 volumes of ice-cold acetone. Samples were centrifuged for 30 min at 20,000xg, air-dried, resuspended in 2x SDS sample buffer and boiled for 5 min at 95°C. Alternatively, 4 h before ΔNS1TCP infection, VeroE6 cells stably expressing wild-type NS1 (NS1WT) or HA-tagged NS1 (NS1HA) were transfected with 10 μg of pcDNA 3. 1 (+) empty vector or pCMV_NS4B-FLAG (encoding a C-terminally Flag-tagged NS4B protein of Hepatitis C virus) by using the TransIT-LT1 transfection reagent (MirusBio LLC, Madison, WI, USA) as recommended by the manufacturer. Infection and HA-specific immunoprecipitation were carried out exactly as described above. Eluted proteins were further analyzed by western blot as specified in the results section. For analysis of secreted NS1, supernatants were clarified through 0. 45 μm filters and incubated with 40 μl mouse anti-HA slurry beads over-night at 4°C, in a head-to-head shaker. Immunoprecipitates were washed three times with lysis buffer and eluted as described above. Alternatively cell culture supernatants containing NS1 were used undiluted for SDS-PAGE. In vitro transcripts were generated as previously described [56]. For RNA transfection, single-cell suspensions were prepared by trypsinization, washed with PBS, and resuspended at a concentration of 1x107 cells (Huh7) or 1. 5x107 cells (VeroE6) per ml in Cytomix, supplemented with 2 mM ATP and 5 mM glutathione. Five to 10 μg of subgenomic or genomic in vitro transcript was mixed with 400 μl of the cell suspension and transfected by electroporation using a Gene Pulser system (Bio-Rad) and a cuvette with a gap width of 0. 4 cm (Bio-Rad) at 975 μF and 270 V. Cells were immediately diluted into 20 ml of DMEM cplt and seeded in the appropriate format (1ml/well in 24-well plates; 2 ml/well in 12-well plates; 15 ml/dish in 15 cm-diameter dishes). Human immunodeficiency virus (HIV) -based particles that were pseudotyped with the vesicular stomatitis virus glycoprotein (VSV-G) were generated by transfection of 293T cells as described previously [53]. For production of transducing lentiviral particles, 293T cells were co-transfected with a transfer vector encoding the gene of interest and a puromycin resistance gene (pWPI_Puro), the HIV-1 packaging plasmid (pCMV) and a VSV-G expression vector (pMD. G) (ratio 3: 3: 1). Cells were transfected using the CalPhos mammalian transfection kit as recommended by the manufacturer (Becton Dickinson). After 48 and 72 h, supernatants were harvested, clarified through 0. 45 μm pore size filters, pooled and stored in aliquots at -20°C until use. Titers of lentiviral particles were estimated by colony-forming unit (CFU) assay using HeLa cells and serial dilutions of each lentiviral stock. Inoculated cells were subjected to selection using the appropriate antibiotic for 5–7 days and surviving cell colonies were stained for 15min with a 1% crystal violet solution. Colonies were counted under a bright-field inverted microscope and lentivirus titers were calculated as CFU/ml. Huh7 or VeroE6 cells transfected with full-length or subgenomic DVsR2A in vitro transcripts were seeded as specified in the results section (typically 12- or 24-wells plates). Replication was determined by measuring luciferase activity in cell lysates 4,24,48 and 72 h after transfection. For determination of luciferase activity, cells were washed once with PBS and lysed by adding 200 μl of luciferase lysis buffer as previously described [57]. Cells were frozen immediately at −70°C and after thawing, lysates were resuspended by gentle pipetting. For each well 20 μl lysate, mixed with 400 μl assay buffer (25 mM glycylglycine, 15 mM MgSO4,4 mM EGTA, 1 mM DTT, 2 mM ATP, 15 mM K2PO4 [pH 7. 8], 1. 42 μM coelenterazine H), were measured for 10 sec in a tube luminometer (Lumat LB9507, Berthold, Freiburg, Germany). In some cases (24-well plates, 100 μl lysis buffer per well) a plate luminometer was used (Mithras LB940, Berthold, Freiburg, Germany). Each well was measured in duplicate. To determine the amount of infectious virus particles released into culture supernatants 72 h after electroporation, naïve VeroE6 cells were inoculated with culture supernatants of transfected cells and 48 h later luciferase activity was determined. Kinetics of virus replication were calculated by normalizing the relative light units (RLU) measured at a given time point to the respective 4 h value. The plasmid containing a synthetic version of the full-length DENV-2 strain 16681 (pFK-DVs) and subgenomic constructs derived therefrom without or with luciferase reporter were previously described [56]. For NS1-targeted site-directed mutagenesis external primers NS1_BamHI_f (5’-CTG GGA TTT TGG ATC CTT GGG AGG AG-3’) and NS1_KasI_r (5’-TCC GTC ATA GTG GCG CCT ACC ATA AC-3’) were used in combination with mutagenic forward and reverse primers (the full list of primers is available upon request). Amplicons containing the desired point mutation in the NS1 coding region were inserted into the full-length DVsR2A constructs via the BamHI-KasI restriction sites. Full-length non-reporter constructs containing selected NS1 mutations were generated by insertion of a DNA fragment that was excised via BamHI and KasI from pFK-DVsR2A into pFK-DVs; subgenomic luciferase reporter constructs were generated by DNA fragment exchange using MluI and KasI and insertion into pFK-sgDVsR2A. The pCMV_NS4B-FLAG expressing C-terminally Flag-tagged NS4B protein of Hepatitis C virus was previously described [58]. The pWPIpuro-based NS1 constructs used for production of lentiviral vectors were generated by PCR-based amplification of the NS1 encoding sequence plus the last 24 codons of the envelope coding region, by using full-length genomic constructs as template and the following primers: pWPIpuro_BamHI_NS1_HA_frw (5’-GCT GGG ATC C ACC ATG AGC ACC TCA CTG TCT GTG ACA CTA GTA TTG GTG-3’) and pWPIpuro_NS1_Stop_NheI_rev (5’-AGA TAG CTA GCC TAA GCT GTG ACC AAG GAG TTG ACC AAA TTC-3’). Amplicons were inserted into pWPI-Puro via BamHI and SpeI restriction sites. For variants encoding the C-terminal HA epitope or the KDEL ER retrieval sequence, the primer pWPIpuro_BamHI_NS1_HA_frw was used in combination with NheI_NS1-HA_stop_rev (5’-ATA GCT AGC CTA AGC GTA ATC TGG AAC ATC GTA TGG GTA TGA TCC AGC TGT GAC CAA GGA GTT GAC CAA ATT CTC TTC TTT CT-3’) or pWPIpuro_NS1_Stop_NheI_KDEL_rev (5’-AGA TAG CTA GCC TAT AGC TCG TCC TTA GCT GTG ACC AAG GAG TTG ACC-3’), respectively. The pWPIpuro-based NS1mCherry expression construct was generated by amplifying the mCherry sequence contained in pFKI389neoNS3-3′δg_JFH-1_NS5A-aa2359_mCherry [59] with primers pWPIpuro_SpeI_mCherry_frw (5´-TCA ACT CCT TGG TCA CAG CTA CCG GTG GAT CGA TGG TGA GCA AGG GCG AGG A-3´) and pWPIpuro_SpeI_mCherry_rev (5´-AAA ACT AGT CTA CTT GTA CAG CTC GTC CAT GC-3´) and the NS1 coding sequence contained in pWPIpuro_NS1 with primers pWPIpuro_SpeI_NS1_frw (5´-GAC ACT AGT ATT GGT GGG AAT TGT GAC AC-3´) and pWPIpuro_SpeI_NS1_rev (5´-TCC TCG CCC TTG CTC ACC ATC GAT CCA CCG GTA GCT GTG ACC AAG GAG TTG A-3´). The two PCR fragments were used to generate an intermediate amplicon using primers pWPIpuro_SpeI_NS1_frw and pWPIpuro_SpeI_mCherry_rev, which was inserted into pWPIpuro via SpeI. Finally, the envelope leader peptide sequence was excised from pWPIpuro_NS1 wild-type and inserted upstream of the NS1mCherry sequence via BamHI-MluI. The ΔNS1 full-length DVsR2A genome used for trans-complementation studies in NS1 helper cell lines, was created by insertion of a 97 codon in-frame deletion into the NS1 open reading frame using NS1_BamHI_f and NS1_KasI_rev as external primers and NS1_156_frw (5’-GAA TTC GTT GGA AGT TGA ACA CAA CTA TAG ACC AGG CTA-3’) and NS1_156_rev (5’-TAG CCT GGT CTA TAG TTG TGT TCA ACT TCC AAC GAA TTC-3’) as internal primers. Thus, in the final construct, the first 156 codons and the last 99 codons of NS1 were retained, respectively. Sequence alignment of NS1 open-reading frames were performed using the ClustalW algorithm available in the JalView Desktop software and a ClustalW scoring algorithm, with the following isolates (UniprotKB/Swiss-Prot accession numbers are given): DV-1 (Brazil/97-11/1997) P27909; DV-2 (Thailand/16681-PDK53) P29991; DV-3 (Martinique/1243/1999) Q6YMS3; DV-4 (Thailand/0348/1991) Q2YHF0; West Nile virus P06935; Yellow Fever (Ivory Coast/1999) Q6J3P1; Japanese Encephalitis virus (SA-14) P27395; Kunjin virus (MRM61C) P14335; St. Louis Encephalitis virus (MS1-7) P09732. Molecular graphics were performed with the UCSF Chimera package developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco [60] on the DENV-2 NS1 crystal structure (Protein Data Bank [PDB] accession no. 4O6B). VeroE6-based helper cell lines expressing different NS1HA mutants were seeded onto glass coverslips (5x104 cells/well) and 16 h later, infected with 1 MOI of DVR2AΔNS1. After a 48 h incubation period, cells were fixed and prepared for transmission electron microscopy as described previously [61]. For correlative light-electron microscopy, VeroE6-based helper cell lines expressing NS1mCherry were seeded into glass-bottom culture dishes containing photo-etched gridded coverslips (MatTek Corporation, Ashland, MA) and infected with 1 MOI of DVR2AΔNS1. After 48 hours, cells were fixed with 4% PFA and 0. 2% glutaraldehyde in PBS for 30 min at room temperature, washed three times with PBS, stained with DAPI and analyzed by fluorescence microscopy to acquire optical sections of 0. 13 μm as described above. NS1mCherry-positive cells were imaged and their position on the gridded coverslip was recorded. Cells were then processed for analysis as described previously [61]. The DAPI signal was used for correlation purposes and images were adapted by using the Image J (version 1. 46r) and Adobe Photoshop (version 12. 1. 1) software packages. Statistical analyses were performed by applying the two-tailed, unpaired Student’s t-test available within the GraphPad Prism (ver. 5. 0) software.

Dengue virus (DENV) is a major arthropod-borne human pathogen, infecting more than 400 million individuals annually worldwide; however, neither a therapeutic drug nor a prophylactic vaccine is currently available. Amongst the DENV proteins, non-structural protein 1 (NS1) is one of the most enigmatic, being required for RNA replication, but also secreted from infected cells to counteract antiviral immune response, thus contributing to pathogenesis. Despite its essential role at early stages of the viral replication cycle, the molecular determinants governing NS1 functions are unknown. Here, we used a combination of genetic, high-resolution imaging and biochemical approaches and found that NS1 additionally plays an important role for the production of infectious virus particles. By using a novel trans-complementation system with fully functional epitope-tagged NS1, we show that NS1 interacts with the structural proteins residing in the envelope of the virus particle. An NS1 variant retained in the endoplasmic reticulum still supported efficient DENV particle production, demonstrating that secretion of NS1 is dispensable for virion production. This study expands the list of functions exerted by NS1 for the DENV replication cycle. Given this multi-functional nature, NS1 appears to be an attractive target for antiviral therapy.

Abstract Introduction Results Discussion Materials and Methods

2015

Dengue Virus Non-structural Protein 1 Modulates Infectious Particle Production via Interaction with the Structural Proteins